Do Global Factors Impact Bitcoin Prices

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/329308550
Do global factors impact bitcoin prices? evidence from wavelet

approach
Article · November 2018
CITATIONS READS
20 1,872
2 authors:
Debojyoti Das Manoharan Kannadhasan

Indian Institute of Management Bangalore Indian Institute of Management Raipur
32 PUBLICATIONS 959 CITATIONS 52 PUBLICATIONS 601 CITATIONS
SEE PROFILE SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Do emerging stock markets react to international economic policy uncertainty, geopolitical risk and financial stress alike? A quantiles-
based approach View project
All content following this page was uploaded by Manoharan Kannadhasan on 30 November 2018.
The user has requested enhancement of the downloaded file.

D. Das and M. Kannadhasan / Journal of Economic Research 23 (2018) 227-264 227
Do global factors impact bitcoin prices? evidence from

wavelet approach1
Debojyoti Dasⅰ
Finance and Accounting Area, Indian Institute of Management Raipur
M. Kannadhasanⅱ
Finance and Accounting Area, Indian Institute of Management Raipur
Abstract
In this article, we attempt to delineate the relationship between bitcoin

prices and global factors such as stock index, economic policy
uncertainty, gold spot prices, implied volatility and crude oil prices in a
time-frequency domain. We resort to wavelet-based analysis to
capture the multiscale interactive behavior of bitcoin with global
factors. We primarily show that bitcoin is insulated from global factors
in the short run. However, the existence of a significant relationship of
bitcoin with global factors cannot be denied in the medium to long run,
which could be attributed to the endogenous and intertwined
economic system. Among the global factors considered in the study, we
find the impact of economic policy uncertainty and crude oil prices to
be more prominent on bitcoin. Our study offers some interesting
insights on multiscale sensitivity of bitcoin to global factors, which may
be useful for investors for taking informed decisions.
Keywords: Bitcoin; Co-movement; Wavelet; Quantile Regression; Global Factors
JEL Classification: G15, Q02
i) Finance and Accounting Area, Indian Institute of Management Raipur, Atal Nagar, P. O. -
Kurru (Abhanpur), Raipur 493661, Chhattisgarh, India, Email: debojyoti. [email protected]
ii) Finance and Accounting Area, Indian Institute of Management Raipur, Atal Nagar, P.
O. - Kurru (Abhanpur), Raipur 493661, Chhattisgarh, India, Email: [email protected]
First received, July 5, 2018; Accepted, November 5, 2018.

228 Do global factors impact bitcoin prices? evidence from wavelet approach
1 Introduction
1.1 Background
Bitcoin is a most innovative form of “digital currency” first introduced

by an individual (or a group of individuals) under the pseudonym of
“Satoshi Nakamoto” (Nakamoto, 2008) during the Global Financial Crisis
(GFC) in November, 2008. Bitcoin had emerged as a safe haven resort at
the crucial juncture of the global economic holocaust of 2008 when
confidence in the formal banking system was evanescing (Bouri, Gupta,
Tiwari, & Roubaud, 2017). Owing to the safe haven properties of Bitcoin
during the periods of high economic stress, they were also referred to as
‘Digital Gold’ (Popper, 2015; Rogojanu & Badea, 2014).
The popularity of Bitcoin in recent years has attracted interest in
knowing about various perspectives of Bitcoin, particularly related to its
use as a monetary or investment instrument. For instance, Bitcoin has been
viewed from a macroeconomic perspective as an alternative monetary
system (Rogojanu & Badea, 2014), as an investment or speculative vehicle
(Baek & Elbeck, 2015; Yermack, 2013), as hedging instrument for global
uncertainty (Bouri, Gupta, et al., 2017), a diversifier, a hedge and safe haven
for energy commodities (Bouri, Jalkh, Molnár, & Roubaud, 2017) and as
enhancing the efficiency of an investor’s portfolio (Wu & Pandey, 2014).
Furthermore, Bitcoin has also been recognized as a complementary
instrument to emerging market currencies (Carrick, 2016). In addition,
past studies have also attempted to identify Bitcoin’s own price-specific
factors. For example, market forces of demand and supply (Bouoiyour &
Selmi, 2015; Buchholz, Delaney, Warren, & Parker, 2012), attractiveness for
investors (Bouoiyour & Selmi, 2015; Kristoufek, 2013, 2015), the overall
price formation economics of Bitcoin (Ciaian, Rajcaniova, & Kancs, 2016),
its informational efficiency (Katsiampa, 2017; Nadarajah & Chu, 2017;
Tiwari, Jana, Das, & Roubaud, 2018; Urquhart, 2016, 2017) and global
macro financial development (van Wijk, 2013) that are affecting the price
of Bitcoin.
Bitcoin has also leapt to prominence in portfolio and investment
management literature very recently. The studies suggest that despite
high volatility and risk (Valstad & Vagstad, 2014) use of Bitcoin in
investment portfolio as a diversifying instrument yields considerable

benefits (Eisl, Gasser, & Weinmayer, 2015; Gandal & Halaburda, 2014).
More importantly, besides commodities, bonds and exchange rates, Bitcoin
has also been referred as an effective diversifier to major stock indices
(Bouri, Molnár, Azzi, Roubaud, & Hagfors, 2017). Traditionally, emerging
markets are viewed as an efficient diversifier for investors from developed
markets since the seminal work of Grubel (1968) and Levy and Sarnat
(1970) on international portfolio diversification. Investments in emerging
markets is channeled from the developed world owing to its higher
returns and lower correlations with the developed markets (Bekaert &
Harvey, 1997). Similarly, Wu and Pandey (2014) report lower correlations
of Bitcoin with major world currencies and asset classes. Further, Bitcoin
have also demonstrated phenomenal performance in terms of price
returns. Figure 1 panel (a) below demonstrates the annualized risk-returns
characteristics to facilitate comparability amongst tradable commodities
i.e. Crude-Oil (WTI), Gold, financial assets; Dow Jones Industrial Average
(DJIA) index as proxy for developed market stocks and the American
emerging market stock indices1. Unquestionably, Bitcoin outperform both
tradable and financial assets. Nevertheless, the risk is also the maximum
in the case of Bitcoin. The phenomenon of higher risks could be attributed
to the tremendous volatile behavior exhibited by the Bitcoin price. On July
18, 2010 Bitcoin was priced at US$0.09, which skyrocketed to US$1124.76
by November 30, 2013 followed by a steep downfall to US$226.98 by
January 13, 2015 finally reaching a record high price by March 03, 2017 of
US$1290.79 (Figure 1 panel (b)). By virtue of higher returns and lower
correlations with the major asset classes (Wu & Pandey, 2014), Bitcoin may
also appear to be a potential contender for an effective diversifier
alternative investment avenue. While the returns from emerging markets
are weakening since the post GFC era (obvious economic underlying
reason could be the law of diminishing marginal returns to investments).
Besides, the gradual higher market interconnectedness (stronger
cross-country investment linkages in financial or real assets) induces the
1
The stocks are selected for emerging American markets and according to the market
classification of Morgan Stanley Capital International (MSCI) as on July 22, 2017. The
constituent indices are MSCI Brazil, MSCI Chile, MSCI Colombia, MSCI Mexico and MSCI Peru.
The observations are collected from Bloomberg and for the same time-period as the other key
variables considered in the study described in section 3, which is devoted to data description.
potential dangers of propagating financial crisis from developed to

emerging and markets in the rest of the world (Lehkonen, 2015). The
demons of financial crisis endow devastating impact on real economic
variables of a country, such as sluggishness in potential output and Gross
Domestic Products (GDP) (Gros & Alcidi, 2010; Yap, Cuenca, & Reyes,
2009), which may hinder the growth of a nation as a whole. Thus, the
benefits of international portfolio diversification also involves certain
macroeconomic costs. Bitcoin being an instrument, which is decentralized
and detached from the real economy, can substantially minimize the evils
concerning the domino effects of crisis transmission. Nonetheless, evidence
in respect of Bitcoin’s sensitivity to global economic factors is a crucial
issue, which is imperative for taking investment decisions. In the recent
past, scholars have examined the safe haven properties of emerging and
frontier stock markets by analyzing co-movement and dependence
structure with global economic factors (Mensi, Hammoudeh, Reboredo,
& Nguyen, 2014; Mensi, Hammoudeh, & Tiwari, 2016). However, there is a
significant omission about how various global economic factors impact
the price of Bitcoin. Thus, in this endeavor we attempt to model the
relationship between Bitcoin and several explanatory economic and
financial variables both under extreme market conditions and also at
different timescales.
1.2 Motivation and literature
This section clearly indicates the motivation of our research i.e. the
research question and methodological choice. We intend to study the
interaction of five global factors with Bitcoin: (a) stock index, (b) gold
prices, (c) implied volatility, (d) oil prices and (e) economic policy
uncertainty (EPU). 2 The rationale of selection of these variables are
discussed as below:
(a) Stock indices: The relationship of Bitcoin with stock indices is studied in
the recent past (Ciaian et al., 2016; van Wijk, 2013) since it is assumed that
stock indices are a barometer of macroeconomic and financial
development. Well-developed and favorable macroeconomic fundamentals
2
The details regarding the indices are provided in section 3.
may stimulate the demand for Bitcoin (Ciaian et al., 2016). Moreover, the
rising correlations between Bitcoin and stock prices have also been a
matter of interest for analysts and academicians. The underlying
phenomenon is attributed to increasing intervention or migration of
traditional investors to crypto-asset market. 3 Thus, an analysis of
co-movement and dependence structure is expected to advance our
understanding of the relationship.
(b) Gold: The relationship of Bitcoin with gold has also been a matter of
considerable interest for the research community (Ciaian et al., 2016;
Dyhrberg, 2016; Zhu, Dickinson, & Li, 2017). Dyhrberg (2016) highlights
the similar traits of Bitcoin and gold, the author makes an important note
that both the commodities derive their value on account of their (a) scarcity,
(b) cost involvement in extraction and (c) no control of any specific
government. The author further classifies Bitcoin as a hybrid commodity,
which endows features of both gold and currency (US$). While gold acts as
a store of value, large price fluctuations can be observed for Bitcoin. As a
choice of an alternative investment, risk-averse investors may prefer gold,
however investors with speculative motive may prefer Bitcoin. Thus, the
interaction of price dynamics between Bitcoin and gold is an important
issue to be examined in the study.
(c) Volatility Index: Weber (2014) claims that the introduction of Bitcoin
profited amid and after the volatile period GFC of 2008. During the market
conditions of high volatility investors usually resort to an uncorrelated and
safer asset –a flight to safety phenomenon. The implied volatility index
(VIX) is a common benchmark of volatility that is often used as a reference
point of market vulnerability by investors. Since Bitcoin were uncorrelated
to mainstream assets (Wu & Pandey, 2014) (higher correlations are
observed recently) and also closely entail the features of gold (Dyhrberg,
2016), disentangling the relationship of Bitcoin with implied volatility
holds considerable value. The contemporary literature on Bitcoin has
attempted to model the relationship, and asserts that Bitcoin acts a hedge
to global uncertainty (Bouri, Gupta, et al., 2017).
(d) Oil prices: The oil prices are also considered as a determinant of Bitcoin
prices (Ciaian et al., 2016; van Wijk, 2013). Oil prices are assumed to be the
3
For full report refer link: https://www.bitcoinmarketjournal.com/bitcoin-vs-stock-market/,
accessed April 03, 2018, 19:03.
harbinger of inflationary conditions in an economy, besides it is also a

prime source of demand and cost pressures (Krugman, 2008; Palombizio &
Morris, 2012). Thus, the price of Bitcoins may depreciate (or appreciate)
when a potential change in general price level is perceived by signals of oil
price oscillations (Ciaian et al., 2016).
(e) Economic Policy Uncertainty: Finally, we include the US EPU as

another critical determinant of Bitcoin prices. The EPU is a newspaper
based index of uncertainty constructed from News Bank Access World
News Database, for details refer to (Baker, Bloom, & Davis, 2016). This
index reflects the overall viability of the business environment and
investment and investment prospects, thus this variable has gained
considerable recognition in financial literature as an indicator of risk
(Mensi et al., 2014). Demir et al., (2018) analyzes the predictive power of
EPU on Bitcoin returns and concludes that Bitcoin is a hedging tool for
economic uncertainty.
Nevertheless, besides many theoretical assumptions and conjectures

scholars also opine that Bitcoin being an unconventional asset with
distinguishable characteristics, standard economic factors may deceive to
explain the causality in its prices (Bouoiyour & Selmi, 2015; Kristoufek,
2013). Thus, we attempt seek an answer to the question whether the global
factors impact the price of Bitcoin? Or is Bitcoin insulated from global
factors? Considering the multiscale nature of relationship (i.e. relationships
are dynamic and change across time and frequencies), we adopt a
multiscale approach of analysis. In addition, we also attempt to unravel the
asymmetric relationship between the variable, which is rarely dealt in
earlier literature. Thus, to fulfill our objective we employ a battery of tests;
they are described briefly as below:4
(a) Wavelet coherence analysis: In the first stage, we employ continuous

wavelet coherence analysis, which shows the nature of co-movements
between Bitcoin and global factors in different frequencies of time, i.e. short,
medium and long term. This test also shows the direction of co-movements
(positive/negative) between the variables.
4
We briefly describe the benefits of our methodological choice in this section. A detailed
discussion on the method is presented in section 2. We are thankful to the anonymous reviewers
for highlighting this point.
(b) Quantile regression analysis: Secondly, we employ the quantile regression

analysis to ascertain the dependence structure between the dependent and
independent variables. This method also enables us to understand the
changing sensitivity of Bitcoin to global factors in different market states
i.e. bullish, normal and bearish states.
(c) Wavelet-based quantile regression analysis: The hybrid model of discrete

wavelet transformation and quantile regression combines the benefits of
the two methods as described above. Thus, using this method we are able to
capture the dependence structure between the variables at different
market-states and also at varying timescales.
(d) Asymmetric quantile regression analysis: This model enables us to

understand the impact of positive and negative shocks in global factors on
Bitcoin. Hence, we can answer questions such as: if there is a positive (or
negative) shock in oil prices (and similarly other variables), what is the
impact on Bitcoin returns?
1.3 Main contributions
We contribute to the literature in the following way: (a) In the first

stance, we employ a wavelet coherence analysis to detect the co-movement
of Bitcoin with the selected global factors. We find weaker signs of
co-movement between Bitcoins and global factors in the short to
medium-run. Nevertheless, some significant co-movement patches are
observed at lower frequencies for EPU and WTI. (b) Secondly, wavelet
based quantile regression reveal economic policy uncertainty (EPU) and
oil prices (WTI) are significant predictors of Bitcoin prices across all
frequencies and both in bullish and bearish market states. However, in the
long run all factors tend to become statistically significant. (c) Finally, we
attempt to model the asymmetric relationship by decomposing the global
factors in positive and negative shocks, we again find that EPU and WTI
significantly impact the price of Bitcoin.
One of the crucial benefits of our study is also the methodological choice
of wavelet-based analysis which has certain benefits over the traditional
econometric techniques to analyze the data on both time and frequency
domain (Das, Kannadhasan, Tiwari, & Al-Yahyaee, 2018; Reboredo &
Rivera-Castro, 2013). Thus, we also understand the time-frequency varying
Figure 1. Risk-returns plot of assets and price series plot of bitcoin
(a) Risk-Returns Plot
(b) Bitcoin Price

Note: Panel (a) of figure 1 exhibits the risk-returns plot of tradable commodities; Crude
Oil (WTI) and Gold, financial assets; Dow Jones Industrial Index as proxy for developed
stock market. EM1…EM5 represent the notations for emerging stock market proxies, the
constituent country indices are: EM1: MSCI Brazil, EM2: MSCI Chile, EM3: MSCI
Colombia, EM4: MSCI Mexico and EM5: MSCI Peru.
sensitivity of Bitcoin with the global factors. This method is also preferred
since in the market there exist heterogeneous investors with different
investment frequencies. Wavelet methods are capable to capture such
multiscale behavior in the data (Rua & Nunes, 2009). We show that the
relationship of Bitcoin with the global factors is dynamic and varies across
the different frequencies. Overall, our results offer insightful findings, which
are also consistent with the previous literature (Ciaian et al., 2016; Demir et
al., 2018; van Wijk, 2013).
The rest of the article is structured as follows: section 2 describes the
methodology, whereas section 3 describes the data characteristics. Section
4 discusses the empirical results and section 5 presents the results for
robustness analysis, while section 6 concludes.
2 Estimation methodology
2.1 Wavelet coherence analysis
Wavelet Coherence Analysis (WCA) is one of the most preferred forms of

wavelet-based analysis for measuring co-movements in two different
time-series objects in a time-frequency space. For example, using WCA
scholars have examined co-movement between equity indices (Aloui &
Hkiri, 2014; Boako & Alagidede, 2017; Das, Bhowmik, & Jana, 2018; Das,
Kannadhasan, Al-Yahyaee, & Yoon, 2018; Das, Kannadhasan, Tiwari, et al.,
2018; Graham, Kiviaho, & Nikkinen, 2012; Rua & Nunes, 2009),
co-movement between several macroeconomic variables concerning
monetary policies and business cycles (Luís Aguiar-Conraria, Azevedo, &
Soares, 2008; Luis Aguiar-Conraria & Joana Soares, 2011), co-movement
between equity indices and economic policy uncertainties (Das & Kumar,
2018; Ko & Lee, 2015) and many others. Wavelet in a layman’s term refers
to ‘small waves’ that exhibit an oscillating pattern of growth and decay in a
time-period which is limited in nature. A wavelet transformation function
enables decomposition of a time series into some elementary function
known as daughter wavelets , ( ) , which results out of a mother
wavelet ( ). The mother wavelet ( ) may be expressed as a function of
time and scale i.e. a translation parameter , which is a function of time
and a dilation parameter s, which represents the scale that is associated

with the frequency-based information. The wavelet transformation
provides well time-localized information while maintaining entropy
conservation, which was a severe limitation for the Fourier based
approach5.The wavelets can be mathematically expressed as:
, ( )= (1)
√
The normalization factor is denoted by , which ensures the

√
comparability of wavelet transforms across time-scales. There are certain
conditions that are required to be complied with for becoming a mother
wavelet ( ) (Bruce & Gao, 1996; Gencay, Selcuk, & Whitcher, 2002;
Percival & Walden, 2000). The conditions are listed as below:
(i) The mother wavelet must have a mean of zero ( ) = 0.

(ii) The square of the mother wavelet must integrate to
one ( ) = 1 .6
| ( )|
(iii) The admissibility condition 0 < = < ∞, where
( ) is the Fourier transformation of ( ), which stands as ( ) =
( ) .This condition permits the restoration of a time series
( )from its continuous wavelet transformation (CWT) ( , ). Thus,
from the wavelet-based transformation it becomes possible to recover
( ) using the following formula:
( )= ( , ) (2)
√
Thus, the CWT of a time-series ( ) with respect to ( ) is given by

the following convolution:
5
The Fourier transformation is capable to decompose a time-series into the infinite length of sine
and cosine waves, however the time-localized information is discarded. It is a major drawback of
Fourier based analysis which can be resolved using wavelets (Gallegati, 2008; Khalfaoui &
Boutahar, 2011; Rua & Nunes, 2009).
6
It means ( ) limited to a specific time-interval.
( , )= ( ) ∗ ( ) ( ) ∗
, = (3)
√
The complex conjugate is denoted by * in equation (3). In case of a

discrete time-series ( ), = 1, … , we have:
( , )= ∑ ∗
( ) (4)
√
The computation of wavelet transform in time domain is possible using

the equation (4) above. However, it is more convenient to carry out the
wavelet transform operation in a Fourier space (Torrence & Compo, 1998).
The type of mother wavelet used for analysis is the Morlet wavelet on
account of its rich applicability in the field of economics and finance (Ko
& Lee, 2015). The following is the mathematical expression of Morlet:
( )= − (5)
For an appropriate the term − becomes negligible, hence the

equation can be modified as:
( )= (6)
The equation (6) with the corresponding Fourier Transformation process

is given by:
( )= √2 ( − ) (7)
From equation (7) it may be observed that a Morlet wavelet within a

Gaussian envelope, it is a complex sine wave, where is the number of
waves (for a detailed explanation refer to Addison, 2002). To strike an
optimum balance of time-frequency localization the is usually set to 6
(Grinsted, Moore, & Jevrejeva, 2004).
Now, given a set of two time-series ( ) and ( ), the cross-wavelet

spectrum may be defined as for the wavelet
transforms and of respective time-series under
consideration. Similar to Fourier-based analysis, the wavelet-squared
coherence may be defined as the square of the absolute value of smoothed
cross-wavelet spectrum, the normalization of which is done using the
smoothed wavelet power spectra, which may be expressed as:
(8)
Where denotes the time and scale smoothing operator (Torrence

& Webster, 1999). As similar to Fourier analysis, the time-scale smoothing
also becomes essential in this case; otherwise the squared coherency would
always result to one (Priestley, 1981). Around each time-frequency
moments the wavelet coherence produce correlation coefficients. Thus, it
becomes convenient to show how a pair of time-series tends to move
together. The coefficient of squared wavelet coherence ranges
typically from 0 to 1, which indicate a weaker to higher co-movement.
Thus, through the mapping of the wavelet coherence contour the deeper
insights regarding co-movements may be captured. To make statistical
inferences, the Monte Carlo simulation on the time-series pair under
consideration with identical length is performed.
One of the advantages of using complex-valued wavelet is that the
phases can be computed for each series. In a time-frequency function the
possible oscillation delays in the two time-series can be obtained by
computing the phase-differences. The phase-difference may be expressed
mathematically as:
( , ( , ))
, ( , )= (9)
( , ( , ))
Where, I and R stand for the Imaginary and Real parts of the wavelet
that allows us to study both amplitude and phase. The arrows on the
wavelet coherence plots indicate the phase. Arrows in the coherence maps
designate the lead/lag relationship in the time-frequency space of a
continuous wavelet transform framework. The left-tailed (→)/ right-tailed
(←) arrows show the two time-series under consideration are in-phase/
anti-phase respectively. The in-phase and anti-phase phenomenon depicts
positive and negative co-movements respectively. The upward ( ↑ ),
right-upward (↗) and left-downward (↙) arrows represent the first time
series which? leads the second one. Similarly, the downward ( ↓ ),
right-downward ( ↘ ) and left-upward ( ↖ ) arrows indicate the first
time-series tends to follow the lead of the second series (Jiang, Nie, &
Monginsidi, 2017;Vacha & Barunik, 2012).
2.2 Wavelet multi-scale analysis
Wavelets are capable of separating a signal into multi-horizon

(multi-resolution) components. Splitting up a signal by wavelet
transformation technique can capture the finer details of a signal at
smaller time-scales (Lehkonen & Heimonen, 2014). According to Ramsey
(2002) any function of time can be represented as a sequence
of projections by father and mother wavelets. The long-scale
smooth components are represented by the father wavelets that integrate
to one. On the other hand, deviations from the smooth components are
represented by mother wavelets, which integrate to zero. The scaling
coefficients are generated by father wavelets, whereas the differencing
coefficients are generated by mother wavelets.
The father wavelet is mathematically expressed as follows:
(10)
whereas, the mother wavelet is defined as below:
(11)
where, j and k represent respective scale and translation parameters.

The scale or the width of the functions and is measured
by .
The smooth coefficients obtained from the father wavelet are as follows:
(12)
The detail coefficients of the mother wavelets are expressed as:
(13)
Thus, the wavelet representation of a function is defined

as the linear combination of wavelet function as below:
(14)
The equation (5) may be simplified as:
(15)
with the orthogonal components as below:
=∑ , , ( ), (16)
(17)
From equation (15), is the resulting multiscale

decomposition of . The level wavelet detail that corresponds to the
changes in the series is defined by . The aggregated sum of
variations at each detail scale is represented by , which becomes
smoother with higher levels of (Gencay et al., 2002).
The maximum overlap discrete wavelet transform (MODWT) is applied

to calculate the scale and wavelet coefficients. The decomposition of the
time-series data under consideration was done using Daubechies (a family
of orthogonal wavelets) least asymmetric filter of length eight [LA(8),
hereafter]. In comparison to Haar wavelet filters, the LA(8) filters are
smoother (Gencay et al., 2002). Further, better-uncorrelated coefficients
across scales are exhibited by LA(8) filters as compared to Haar filters
(Cornish, Bretherton, & Percival, 2006). The decomposition of series are
done to wavelet coefficients to following (Bouri, Gupta, et al., 2017).
The resolutions of the data under investigations are provided for scales at
to . The oscillation periods of 2-4, 4-8, 8-16, 16-32, 32-64 and 64-128
days corresponds to wavelet scales … respectively (Table 1).
Table 1.Time interpretation of different frequencies

Scale 1 2~4 days
Scale 2 4~8 days
Scale 3 8~16 days
Scale 4 16~32 days
Scale 5 32~64 days
Scale 6 64~128 days
2.3 Quantile regression
The traditional regression models establish a symmetric linear

relationship between the variables of interest. In general, to establish a
relationship between an independent variable and a set of variables a
model with linear specification is formulated which is conditional upon its
mean. Thus, the results essentially focus upon the mean value relationship.
However, the relationship at different levels of conditional distribution of
dependent variable remains unexplored. To captivate the complex
dependence of time-series Quantile Regression (QR) technique is applied
on time-varying wavelet crystals. The QR technique was introduced by
Koenker and Bassett (1978). Scholars to captivate the asymmetric and
non-linear dependence structure advocate the use of QR (Baur, 2013;
Mensi et al., 2014; Nusair & Al-Khasawneh, 2017).
Let be a dependent variable, which is assumed to be dependent on

linearly. Thus, the conditional quantile of function may be specified
as below:
(18)
where, the conditional distribution function of is represented by

. The relationship of dependence between vector and the
conditional quantile of function is determined by . The dependence
is conditional if the exogenous variables are added to , while the
dependence is unconditional if the exogenous variables are excluded from
. The complete dependence structure of is determined by for
. There could be three possible dependence structures between
and vector of independent variables : (a) where the value of for
different values of is similar (dissimilar) the relationship is symmetric
(asymmetric) for lower and higher quantiles, (b) where the value of
does not change, the relationship is constant and (c) where the values of
increases (decreases) with the values of , the relationship is
monotonically increasing (decreasing).
For a given , the coefficients of are estimated by minimization of

the weighted absolute deviations between and :
(19)
where, the usual indicator function is denoted by . The

solution to this problem is arrived by using linear programming algorithm
(Koenker & D’Orey, 1987). To obtain the standard error for the estimated
coefficients the pair-bootstrapping procedure proposed by Buchinsky
(1995) is used. The asymptotically valid standard errors under hetero
skedasticity and misspecifications of the QR function are obtained by this
pair-bootstrapping procedure.
The quantile regression process is used on the wavelet-decomposed values
of the variables. This enables to capture the interaction dynamics between the
variables across different timescales and also at different market conditions.
This model has recently been used to model the dependence structure among
the variables (Das, Bhatia, Pillai, & Tiwari, 2018; Karlsson, Karlsson, Månsson,
& Sjölander, 2017; Mensi et al., 2016). Whereas, in order the gauge the impact
of positive and negative shocks in isolation the explanatory variables are
decomposed into positive and negative components, then the quantile
regression process is used to capture the relationship at different conditional
values of the series under consideration following the recent literature (Nusair
& Al-Khasawneh, 2017).
3 Data
We examine the dependence structure between Bitcoin and the global

factors by considering the following proxies: (a) Dow Jones Industrial
Average (DJIA) as a representative of major stock market, (b) West Texas
Intermediate (WTI) Brent [crude oil] priced in US dollar (US$) per barrel,
(c) gold spot prices in US dollar (US$) per ounce, (d) CBOE volatility index
(VIX) as a measure of implied volatility and (e) US Economic Policy
Uncertainty (EPU) Index7 (Baker et al., 2016) for a period spanning over
July 18, 2010 to March 27, 2017 (1,553 daily observations), which is
constrained by availability of Bitcoin price data. The non-synchronous
data points are removed to avoid the problem of underestimation of true
correlations, refer to (Martens & Poon, 2001). The logarithmic
transformation of all the series is done taking the first difference. The
Bitcoin price is obtained from https://www.coindesk.com/price/. The DJIA and
VIX data is extracted from Bloomberg database. The gold spot and WTI
series are collected from the websites of World Gold Council and Energy
Information Administration respectively. Table 2 presents the descriptive
statistics of the data. It may be observed that the mean returns are highest
for Bitcoin, whereas gold returns are minimum. The maximum
unconditional volatility is exhibited by WTI expressed in terms of
standard deviation, whereas DJIA exhibits minimum volatility. The
tradable assets Bitcoin, DJIA, gold and WTI show negative skewness
coefficients, which suggests negative values are more frequent than
7
The EPU Index for the US is gathered from http://www.policyuncertainty.com/index.html.
positive values. Excess kurtosis coefficients for all the variables (except for
WTI) confirm the presence of fat tails. Hence, the Jarque-Bera (JB) test
rejects the null hypothesis of normality. The Ljung-Box test at lag order of
10 reveals serial autocorrelation for all the variables except WTI. Thus, the
variables exhibit a pattern of non-linear temporal dependencies up to 10
daily observations. The null hypothesis of unit roots are finally tested using
the conventional Augmented Dickey Fuller (ADF) (Dickey & Fuller, 1979)
and Philips Perron (PP) (Phillips & Perron, 1988) test. The results show
that the returns series for all the variables are stationary at 1% level. In
addition, no significant correlations are reported with the global factors.
Table 2. Summary statistics

(a) Descriptive Statistics
Bitcoin DJIA EPU Gold VIX WTI
Mean (×100) 0.38 0.01 -0.02 -0.04 0.03 0.01

SD 0.06 0.01 0.02 0.06 0.01 0.49
Skewness -0.39 -0.94 0.15 0.87 -0.43 -0.11
Kurtosis 12.51 11.11 5.73 6.40 6.99 2.00
JB 7837.50*** 1119.97*** 322.01*** 587.27*** 873.24*** 2946.98***
LB Q-Stat
68.77*** 21.22** 167.13*** 18.68** 22.92** 14.05
(10)
ADF -35.24*** -41.48*** -54.08*** -42.72*** -42.01*** -41.28***
PP -35.60*** -41.63*** -58.83*** -42.68*** -42.72*** -41.25***
(b) Unconditional Correlations
Bitcoin DJIA EPU Gold VIX WTI
Bitcoin 1
DJIA 0.0487 1
EPU 0.0113 -0.025 1
Gold 0.0233 0.3612*** -0.0217 1

VIX -0.0364 -0.7968*** 0.027 -0.2893*** 1
WTI 0.0157 -0.0185 -0.0108 0.1567*** 0.009 1

Note: The critical value of Jarque-Bera (JB) test at 5% level is 5.99. Ljung-Box (LB) test
was performed taking lag of 10.
* Indicates significance at the 10 per cent level.
** Indicates significance at the 5 per cent level.
*** Indicates significance at the 1 per cent level.
4 Empirical results and discussion
In this segment, we stepwise discuss the empirical findings of our study.

Figure 2 exhibits the wavelet coherence maps of Bitcoin’s interaction with
the global factors. In order to facilitate ease of interpretation, the scale
(legend on the vertical left axis) is converted to the number of days. Overall,
in higher frequencies no significant co-movements can be observed, the
limited patches (red and black bounded contour) of significant
co-movement is observed only in medium and lower frequencies. The
significant co-movements are more noticeable during the year 2013, of
which Bitcoin’s co-movement is most coherent with EPU.8 Stanchfield
(2016) discusses that an uncertain economic policy environment has a
tendency to drive up Bitcoin prices (flight to safety phenomenon). This
feature indicates that Bitcoin are beyond a mere diversifier, as argued by
Bouri, Molnár, et al., (2017) and Demir et al., (2018). The year 2013 had been
a period of economic chaos with the US Debt Ceiling Crisis9. On the eve of
the crisis, prior to credit rating downgrade episode in October 2013, the
Bitcoin prices spiked up by US$140 by the month of May.10 Experts such
as Roger Ver, the president of the bitcoinstore.com and member of the
BitAngels investment team opined that there would be no massive impact
on Bitcoins due to US default crisis in the short-term. Ver further
comments, “There will never be an actual default. They will print more
dollars to pay the debt, causing massive inflation.”11 The US$ would
8
Significant co-movements may also be observed for DJIA, EPU, Gold and WTI at medium and
lower frequencies. We refrain to consider those patches as significant co-movements since the
information is plagued by edge effects and falls out of statistical significance.
9
The US debt-ceiling crisis of 2013 was concerned with the policy debate of raising the federal
government debt ceiling. While the President argued that not raising the debt ceiling would delay
the payments (including the benefits of Government) and the economy would also be susceptible
to default on Government debt. Ben Bernanke, the Chairman of the Federal Reserve also
supported the raise in the debt ceiling. However, the Republicans opposed and argued that the
ceiling should not be raised unless the spending is curbed by an equal or greater margin of ceiling
increase. The Fitch Ratings warned the Government that delays in debt ceiling review might lead
to down gradation of US credit rating from AAA status. The Fitch Ratings on October 15, 2013
placed US as “Rating watch Negative” in response to the crisis. The Dagong Global Credit
Rating downgraded the credit rating of US from A to A-.
10
Though a mild market crash was experienced in BitCoin prices when the Silk Road was taken
down, however the market recovered very quickly. For full report refer https://www.coindesk.
com/us-debt-crisis-bitcoin/, accessed August 09, 2017.
11
For full report refer https://www.coindesk.com/us-debt-crisis-bitcoin/, accessed August 09, 2017.
appear less trustworthy, since more money printed would drive up prices
of commodities. It is a well-known fact that inflation in general would
discourage investments in instruments with a fixed income such as bonds
and currencies. As in an inflationary period the local currency is devalued
and the actual worth (of currency/money) is not denominated. In a case
where faith in currency is shaken, Bitcoin may tender a store of value.12
Nevertheless, gold is also a considerable alternative with lower risk, while
Bitcoin is highly speculative. 13 Our results conform to the standard
economic arguments. However, the information regarding the direction of
relationship (positive or negative) is mixed and inconclusive. The other
significant co-movement contours are observed around the period of 2015,
especially for gold and WTI. In 2015, the drawdowns in oil prices ensued
volatilities in the foreign exchange markets.14 Since, under volatile market
conditions market participants tend to flee to safer assets like gold or
Bitcoin, thus, co-movement between these assets may be expected. In
addition, the co-movement between Bitcoin and WTI shows a negative
relationship around 2015 as represented by right-tailed (←) arrows. The
negative relationship may be justified since the falling oil prices induced
volatilities in the foreign exchange markets, hence stimulating the demand
for Bitcoin.
12
The head economist at Buttercoin, Kevin Zhou comments “One of the reasons that bitcoin is
touted as a way to preserve your capital and retain value is that it isn’t just centrally controlled,
and so it can’t be freely printed and debased”. For full report refer https://www.coindesk.com/
us-debt-crisis-bitcoin/, accessed August 09, 2017.
13
For full report refer https://www.coindesk.com/us-debt-crisis-bitcoin/, accessed August 09, 2017.
14
For full report refer http://www.marketwatch.com/story/the-events-that-rocked-financial-
markets-in-2015-2015-12-22, accessed August 09, 2017.
Figure 2. Wavelet coherence maps
(a) Dow Jones Industrial Average Index (b) US Economic Policy Uncertainty
(c) Gold (d) CBOE Volatility Index
(e) WTI Brent

Note: The 5% significance level estimated from Monte Carlo Simulations is designated by
the black contour. The red (blue) colors signify region with high (low) coherency. The
coherence power color bar (on the right) shows the power of coherence coefficients.
To understand the relationship further by unraveling the dependence

structure between global factors and Bitcoin, we employ QR analysis.
Table 3 presents the OLS and QR results for the baseline model. In QR
analysis we report the results for seven quantiles of the relationship (0.05,
0.10, 0.25, 0.50, 0.75, 0.90, 0.95), the quantiles 0.05, 0.10 and 0.25 are
considered as lower quantiles or bearish market conditions, on the other
hand, quantiles 0.75, 0.90 and 0.95 represent higher quantiles or bullish
state of the market. Intermediate quantile is represented by 0.50, which
depicts a normal market state. The Bitcoin depict significant dependence
on the global factors except for EPU at some higher quantile of 0.75 and
0.90. Oil prices (WTI) are found positively related to Bitcoin prices at a
bullish market state. The underlying reason may be attributed to the fact
that oil prices are forerunners of inflationary conditions (as mentioned
before). Inflation devalues the currency as the general prices are driven up
and hence Bitcoin could be used as a store of value with flexibility for
speculation. Figure 3 exhibits the plot of coefficients QR analysis.
Table 3. Baseline OLS and QR Model

OLS 0.05 0.10 0.25 0.50 0.75 0.90 0.95
DJIA -5.399*** -7.911*** -5.875*** -2.289*** -1.852*** -4.241*** -10.47*** -10.87***

(0.424) (0.307) (0.948) (0.676) (0.472) (0.598) (0.940) (0.770)
EPU 0.0766*** 0.00146 0.0468*** 0.0933*** 0.0594*** 0.0466 0.0442 0.0900***
(0.00993) (0.0159) (0.0164) (0.00885) (0.00764) (0.0308) (0.0289) (0.0262)
Gold 0.715*** 3.246*** 2.462*** 0.907*** 0.168 0.581*** 0.892*** 0.333
(0.153) (0.127) (0.260) (0.191) (0.170) (0.153) (0.248) (0.375)
VIX 0.481*** 0.498*** 0.407*** 0.123 0.209*** 0.544*** 0.833*** 0.470***
(0.0522) (0.0388) (0.0890) (0.0744) (0.0635) (0.0775) (0.127) (0.106)
WTI 4.956*** 3.142*** 2.801*** 2.261*** 2.469*** 2.161** 8.030*** 11.06***
(0.363) (0.426) (0.318) (0.545) (0.397) (1.093) (1.152) (1.099)
Constant 1.15e-10 -0.012*** -0.010*** -0.006*** -0.0012*** 0.00370*** 0.0121*** 0.0178***
(0.0001) (0.001) (0.004) (0.002) (0.002) (0.0003) (0.0007) (0.0008)
R2 0.205
Pseudo R2 0.015 0.012 0.011 0.001 0.018 0.017 0.023
Note: The dependent variable is Bitcoin. In parentheses the standard errors associated with each of the
coefficients are shown. The headers in the top row represent level of quantiles.
Figure 3. Coefficient Plot of Predictors
Note: The figure exhibits estimate of OLS and coefficients of quantile regression for
predictor variables indicated on the y-axis. The solid red horizontal lines represent
the coefficient of OLS at its 95 per cent confidence intervals respectively. The shaded
areas around the quantile regression coefficient plotline represent the coefficient of
quantile regression at its 95 per cent confidence intervals respectively.
Table 4 presents the results of OLS and QR for multiscale model15.

Tables 4(a) to 4(f) represents the results for wavelet scale 1 to 6 (see Table 1
for time-interpretations). Scale 1 and 2 represents short-run, scale 3 and 4
corresponds to medium-run and scale 5 and 6 denotes the long run. Table
4(a) reports the results for Scale 1 (2~4 days). In this scale Bitcoin prices
show positive and higher tail dependence to EPU, except for quantile 0.95.
This finding has the important implication that higher EPU contributes
considerably to bullish Bitcoin prices. In addition, Bitcoin prices also
exhibit both positive higher and lower tail dependence on oil prices (WTI).
Similarly, Table 4(b) shows the results for Scale 2 (4~8 days). Though the
OLS model shows a positive and significant relationship of Bitcoin with
EPU and VIX, the QR model do not depict the same and at any quantile.
15
By multiscale model we mean the OLS and QR analysis on the wavelet decomposed returns
series data.
The OLS model in this scale shows negative and is highly significant, in
this case the QR model conforms to the OLS. The QR model shows
Bitcoin’s significant negative dependence both at lower tail (except
quantile 0.05) and at higher tail. Overall, in the short-run WTI and EPU
can be regarded as important determinants of Bitcoin prices.
Table 4(c) exhibits the results for Scale 3 (8~16 days). In this scale, DJIA
appears to impact Bitcoin prices significantly and negatively at quantile
0.50 and 0.75. The negative relationship of Bitcoin prices with stock (DJIA)
could be explained in two ways: (a) as mentioned earlier, Bitcoin may be
preferred when economic downturns prevail. Thus, investors may trade
more on Bitcoin when stocks depict plummeting tendencies. (b) Secondly,
on the occasions of economic upturns investors may tend to fly back to
stocks. Several concerns over Bitcoin’s illiquidity issues has been expressed
in the past16. Thus, investors may also prefer stocks to Bitcoins under
stable economic conditions on account of liquidity –a flight to liquidity
phenomenon. The OLS model shows EPU to be highly positively
significant. However, the QR model reflects only significant lower tail
dependence, nonetheless the positive relationship persists at higher
quantiles as well. The other important finding is Bitcoin prices show
higher tail dependence to implied volatility (VIX) in the market. It implies
the positive impact of stock market volatility on Bitcoin prices (leading to
the bullish Bitcoin market), this result is consistent with previous studies
(Bouri, Gupta, et al., 2017). Oil prices depict lower tail dependence,
whereas Gold shows significant positive relationship at quantile 0.25 and
0.75. Table 4(d) reports the results for Scale 4 (16~32 days). This scale shows
Bitcoin’s lower tail dependence upon DJIA (also higher tail dependence,
however only at quantile 0.95). The EPU is positive and significant both at
lower (quantile 0.05) and higher quantiles (quantile 0.75 and 0.95). The VIX
is positive and significant at quantiles 0.50 and 0.75, whereas, the oil prices
(WTI) are significant and negative at quantile 0.50 however significant and
positive at quantile 0.95. The OLS model shows DJIA, EPU and Gold are
significant predictors of Bitcoin prices. However, Gold is found significant
only at quantile 0.75. Overall, in the medium term DJIA, EPU and WTI
may be assumed to influence Bitcoin prices, whereas gold exhibits a
16
For full report refer to https://ftalphaville.ft.com/2016/07/12/2169101/bitcoins-unfortunate-
liquidity-problem-as-inadvertently-highlighted-by-an-sec-order/, accessed August 11, 2017.
weaker causality.
Table 4(e) presents the results for Scale 5 (32~64 days). Bitcoins depict
significant relationships with DJIA, EPU, VIX and WTI. Similarly, in Table
4(f) which represents Scale 6 (64~128 days) shows that all the variables
become largely significant. In addition, the predictive ability of the model
also improves at the longest timescale both for OLS and QR as given by
the R2 and Pseudo R2 respectively.
Table 4. Multiscale Model

(a) Scale 1
OLS 0.05 0.10 0.25 0.50 0.75 0.90 0.95
DJIA 0.221 0.741 0.443 0.181 0.0411 0.0565 -0.270 -0.0494
(0.272) (0.838) (0.533) (0.262) (0.140) (0.359) (0.757) (1.040)
EPU 0.00532 0.00937 0.00361 0.00390 0.00473** 0.0118*** 0.0277** 0.00774
(0.00428) (0.0155) (0.00811) (0.00350) (0.00194) (0.00390) (0.0118) (0.0163)
Gold -0.0391 0.0282 -0.0375 -0.00580 0.0324 0.0158 0.0781 0.00600
(0.0919) (0.236) (0.186) (0.0583) (0.0451) (0.0709) (0.285) (0.406)
VIX 0.0359 0.0895 0.0182 0.0449* 0.000860 0.0157 0.0997 0.110
(0.0332) (0.0893) (0.0661) (0.0243) (0.0158) (0.0284) (0.0650) (0.152)
WTI 0.571*** 0.704** 0.812*** 0.425*** 0.232*** 0.338** 0.736** 1.346**
(0.150) (0.329) (0.162) (0.106) (0.0816) (0.153) (0.299) (0.544)
Constant -0 -0.037*** -0.024*** -0.011*** -0.0012*** 0.00845*** 0.0264*** 0.0422***
(0.000711) (0.00274) (0.00138) (0.000626) (0.000383) (0.000710) (0.00172) (0.00314)
R2 0.015
Pseudo
0.024 0.014 0.013 0.004 0.009 0.013 0.017
R2
(b) Scale 2
OLS 0.05 0.10 0.25 0.50 0.75 0.90 0.95

DJIA 0.0410 -1.058 -0.325 0.209 0.330* 0.211 -0.254 -1.375
(0.354) (0.911) (0.756) (0.326) (0.176) (0.308) (0.612) (1.116)
EPU 0.00817** -0.000775 -0.00157 0.00185 0.00250 0.00451 0.00655 -0.00386
(0.00416) (0.0158) (0.00935) (0.00357) (0.00257) (0.00455) (0.00988) (0.0185)
Gold -0.0920 -0.149 0.0900 -0.0439 -0.0373 -0.0188 0.0252 0.153
(0.107) (0.502) (0.218) (0.0712) (0.0565) (0.0673) (0.305) (0.447)
VIX 0.0793** 0.192 0.0859 0.0195 -0.00686 0.0247 0.163 0.328*
(0.0388) (0.170) (0.0894) (0.0368) (0.0213) (0.0348) (0.1000) (0.172)
WTI -0.827*** -0.556 -0.660** -0.630*** -0.410*** -0.627*** -1.139*** -1.643**
(0.188) (0.711) (0.277) (0.179) (0.0997) (0.224) (0.422) (0.736)
Constant -0 -0.025*** -0.016*** -0.0059*** -0.00057*** 0.00461*** 0.0152*** 0.0260***

(0.000460) (0.00217) (0.000911) (0.000385) (0.000213) (0.000480) (0.00107) (0.00170)
R2 0.018
Pseudo
0.010 0.009 0.007 0.007 0.008 0.015 0.024
R2
(c) Scale 3
OLS 0.05 0.10 0.25 0.50 0.75 0.90 0.95
DJIA 0.321 -0.0364 0.121 -0.314 -0.483* -0.758* -0.564 2.640
(0.363) (0.851) (0.775) (0.392) (0.267) (0.398) (1.290) (1.914)
EPU 0.0117*** 0.0180* 0.00194 -0.000105 0.00357 0.00633 -0.00197 0.0136
(0.00384) (0.0101) (0.00629) (0.00338) (0.00248) (0.00405) (0.0108) (0.0153)
Gold 0.0932 -0.215 0.179 0.163*** 0.0437 0.202* 0.125 0.0837
(0.0996) (0.592) (0.158) (0.0583) (0.0389) (0.119) (0.413) (0.581)
VIX 0.0497 0.135* 0.0477 0.0244 0.0491** 0.0663* 0.140* -0.0908
(0.0360) (0.0808) (0.0518) (0.0272) (0.0194) (0.0359) (0.0829) (0.148)
WTI 0.0744 1.691*** 0.930** -0.187 -0.266 -0.340 0.798 0.728
(0.206) (0.546) (0.417) (0.154) (0.172) (0.239) (0.558) (0.804)
Constant 0 -0.018*** -0.011*** -0.0044*** -0.00045*** 0.00368*** 0.0111*** 0.0199***
(0.000304) (0.00131) (0.000661) (0.000263) (0.000168) (0.000251) (0.000909) (0.00152)
R2 0.015
Pseudo
0.038 0.013 0.004 0.004 0.006 0.010 0.023
R2
(d) Scale 4
OLS 0.05 0.10 0.25 0.50 0.75 0.90 0.95

DJIA 1.268*** 2.226** 0.879 0.855** 0.150 -0.562 0.635 3.016*
(0.401) (1.034) (0.759) (0.360) (0.280) (0.591) (1.447) (1.642)
EPU 0.0140*** 0.0263** 0.00756 -0.00187 0.000923 0.00836** 0.00754 0.0406**
(0.00419) (0.0103) (0.00589) (0.00416) (0.00222) (0.00418) (0.0142) (0.0186)
Gold 0.286** 0.141 0.205 0.128 0.0594 0.276* 0.159 0.489
(0.115) (0.272) (0.154) (0.111) (0.0708) (0.148) (0.193) (0.499)
VIX 0.0610 -0.000429 0.0164 0.0182 0.0566* 0.132*** 0.154 0.0773
(0.0421) (0.107) (0.0789) (0.0343) (0.0314) (0.0419) (0.108) (0.203)
WTI 0.121 -0.492 -0.765 -0.287 -0.505*** -0.235 1.638 2.088*
(0.234) (0.532) (0.484) (0.325) (0.175) (0.325) (0.997) (1.206)
Constant -0 -0.014*** -0.0091*** -0.0043*** -0.00096*** 0.0025*** 0.011*** 0.018***
(0.000244) (0.000712) (0.000530) (0.000184) (0.000110) (0.000293) (0.000796) (0.00143)
R2 0.054
Pseudo
0.058 0.016 0.010 0.007 0.012 0.035 0.094
R2
(e) Scale 5
OLS 0.05 0.10 0.25 0.50 0.75 0.90 0.95
DJIA 2.039*** 7.774*** 3.957*** 0.694** 0.375 -0.224 0.739 5.750**
(0.448) (1.215) (0.957) (0.335) (0.342) (0.440) (1.127) (2.294)
EPU 0.0136*** 0.0117 -0.000610 0.00366 0.0160*** 0.0153** -0.0128 0.0196
(0.00519) (0.0106) (0.00835) (0.00538) (0.00265) (0.00600) (0.0106) (0.0143)
Gold -0.287** -0.329 0.275 -0.0706 -0.0840 0.0806 -0.0803 -1.311***
(0.133) (0.280) (0.303) (0.101) (0.0584) (0.137) (0.229) (0.502)
VIX -0.00838 -0.367*** -0.170** 0.0681* 0.0756* 0.110** 0.227 -0.268
(0.0484) (0.120) (0.0819) (0.0355) (0.0450) (0.0469) (0.141) (0.222)
WTI 3.427*** 1.495** 2.353*** 1.530*** 1.589*** 2.809*** 5.632*** 6.209***
(0.302) (0.601) (0.509) (0.472) (0.314) (0.546) (0.600) (0.469)
Constant -0 -0.0149*** -0.0109*** -0.0043*** -0.00056*** 0.00393*** 0.0117*** 0.0165***
(0.000233) (0.000778) (0.000538) (0.000285) (0.000110) (0.000357) (0.000562) (0.00132)
R2 0.131
Pseudo
0.135 0.078 0.022 0.029 0.043 0.125 0.195
R2
(f) Scale 6
OLS 0.05 0.10 0.25 0.50 0.75 0.90 0.95
DJIA -5.399*** -7.911*** -5.875*** -2.289*** -1.852*** -4.241*** -10.47*** -10.87***
(0.424) (0.349) (0.858) (0.681) (0.516) (0.654) (1.002) (0.636)
EPU 0.0766*** 0.00146 0.0468*** 0.0933*** 0.0594*** 0.0466* 0.0442** 0.0900***
(0.00993) (0.0149) (0.0137) (0.00856) (0.00677) (0.0274) (0.0223) (0.0242)
Gold 0.715*** 3.246*** 2.462*** 0.907*** 0.168 0.581*** 0.892*** 0.333
(0.153) (0.173) (0.240) (0.194) (0.166) (0.141) (0.204) (0.351)
VIX 0.481*** 0.498*** 0.407*** 0.123 0.209*** 0.544*** 0.833*** 0.470***
(0.0522) (0.0463) (0.0842) (0.0746) (0.0686) (0.0811) (0.108) (0.0744)
WTI 4.956*** 3.142*** 2.801*** 2.261*** 2.469*** 2.161** 8.030*** 11.06***
(0.363) (0.389) (0.308) (0.554) (0.345) (0.986) (0.974) (0.876)
Constant 1.15e-10 -0.012*** -0.0094*** -0.0052*** -0.0012*** 0.0037*** 0.0121*** 0.0178***
(0.000218) (0.000423) (0.000420) (0.000273) (0.000209) (0.000325) (0.000813) (0.000707)
R2 0.205
Pseudo
0.268 0.197 0.099 0.040 0.045 0.192 0.288
R2
Table 5 represents the results for the asymmetric model17. This model
shows the impact of global factors on Bitcoin prices, after decomposition
of variables into positive and negative shocks (Granger & Yoon, 2002). The
result for OLS shows a positive shock in EPU and VIX leads to fall in
Bitcoin prices. The coefficients are not significant though and the
magnitude of impact is also diminutive. The possible explanation for this
phenomenon could be the risk aversive behavior of the investors during
turbulent economic conditions (Hoffmann, Post, & Pennings, 2013). Both
EPU and VIX relates to unstable economic and financial environment.
Thus, in such economic condition investors may take passive investment
decisions i.e. investors might tend to redeem or hold back their funds until
the economic environment stabilizes. The QR model corresponds the
finding of OLS only for VIX. In addition, the QR model shows a negative
shock in VIX significantly and negatively impacts Bitcoin prices. The
negative shock in VIX implies a fall in implied market volatility, thus when
the economy is more certain lesser will be the preference to invest in
Bitcoin. A positive shock in gold prices depicts a negative impact on
Bitcoin prices at quantile 0.50. The underlying reason could be attributed
to the fact that investors may prefer to invest in gold over Bitcoin, when
gold prices show rising trajectories (coupled up with obvious reason of
lower risks). On the other hand, a negative shock in gold prices also
illustrates a negative impact at quantile 0.90. As discussed earlier, the
features of gold and Bitcoin are somewhat identical (Dyhrberg, 2016), both
of them belong to the class of alternative investment options (hedge/safe
haven). A negative shock in gold prices may be stimulated due to investors
impending tendencies to invest in mainstream assets. Bitcoin being an
instrument of identical class, the prices might become susceptible to a fall.
The results further show a significant negative relationship of Bitcoin in
the lower tail of positive shock in crude oil (WTI). The probable underlying
reason could be ascribed to the fact that in the bearish (lower tail) Bitcoin
market conditions might discourage investors to trade in Bitcoin.
Alternatively, investors might also be interested to trade in commodities
like crude oil (with speculative motive) when a positive wave hit the
markets. However, the relationship becomes positive and significant at
17
By asymmetric model we mean the OLS and QR analysis after the decomposition of the
returns series data into positive and negative shocks.
higher quantiles (or in bullish market conditions), on account of obvious

intents to hedge inflationary conditions. The negative shock in WTI
exhibits a negative impact on Bitcoin prices at higher quantile of 0.90. One
of the potential reasons could be the fact that if Bitcoin is used as an
inflationary hedge, a fall in oil prices may be perceived as a potential
stabilization in the economy and hence Bitcoin prices are impacted
negatively.
Table 5. Asymmetric Model

OLS 0.05 0.10 0.25 0.50 0.75 0.90 0.95
DJIA (+) 0.00169 0.00879 0.00210 -5.82e-05 0.00125 0.00101 0.00129 -0.00101
(0.00301) (0.00558) (0.00472) (0.00132) (0.000855) (0.00311) (0.00735) (0.00909)
DJIA (-) -0.00148 -0.00454 -0.00201 -0.000427 -0.00103 -0.000519 0.000814 -0.0178
(0.00273) (0.0132) (0.00444) (0.00289) (0.00122) (0.00239) (0.0128) (0.0215)
EPU (+) -0.00606* -0.0103 -0.0170 -0.00208 -0.00122 -0.00298 -0.00828 -0.0136
(0.00330) (0.0235) (0.0114) (0.00283) (0.00116) (0.00306) (0.00782) (0.0263)
EPU (-) -0.00225 -0.0113 -0.00494 -0.000928 -0.000705 0.00143 0.00415 0.00945
(0.00294) (0.00808) (0.00626) (0.00127) (0.00112) (0.00128) (0.00488) (0.0295)
Gold (+) 0.00176 0.0114 -0.00632 -0.00243 -0.00141* -0.00457 -0.00202 0.0213
(0.00323) (0.00713) (0.00562) (0.00233) (0.000820) (0.00319) (0.0121) (0.0187)
Gold (-) -0.00382 0.00482 0.00152 -0.000900 -0.00149 -0.00280 -0.0230* -0.0218
(0.00346) (0.0161) (0.0128) (0.00164) (0.00116) (0.00460) (0.0119) (0.0173)
VIX (+) -0.00546* -0.0155 -0.0135 -0.00116 -0.000574 -0.000887 -0.00387 -0.0103**
(0.00290) (0.0238) (0.0107) (0.00294) (0.00109) (0.00204) (0.00276) (0.00444)
VIX (-) -0.000834 -0.00385 -0.00423 -0.00298*** -0.000611 -0.000195 9.76e-05 0.00218
(0.00304) (0.00963) (0.00492) (0.000920) (0.00128) (0.00160) (0.00514) (0.00926)
WTI (+) 0.000318 -0.0155** -0.0111 -0.00285* -0.00198 0.0105* 0.0108*** 0.00618
(0.00272) (0.00766) (0.00952) (0.00145) (0.00146) (0.00582) (0.00397) (0.00772)
WTI (-) -0.00147 0.0178 0.00120 0.000345 -0.00120 -0.000990 -0.0164** -0.00762
(0.00275) (0.0235) (0.0102) (0.00314) (0.00159) (0.00289) (0.00756) (0.0163)
Constant 0.00465** -0.0798*** -0.0386*** -0.0133*** 0.00209*** 0.0226*** 0.0557*** 0.0907***
(0.00210) (0.0104) (0.00395) (0.00142) (0.000721) (0.00232) (0.00486) (0.0133)
R2 0.007
Pseudo R2 0.034 0.014 0.005 0.002 0.005 0.021 0.026
In order to test the robustness, the “scale-by-scale”stability of results is

checked through the application of non-parametric regression analysis.
More specifically, the Wald robustness test is carried out as suggested by
Koenker and Bassett (1982). The test is helpful to gauge all parameter
heterogeneity across any two given quantiles. Thus, the null hypothesis of
equality of slope across quantiles is tested against the alternative
hypothesis of significantly different slopes across quantiles. Table 6
exhibits the results of heterogeneity tests for the linear and multiscale
models. The results are reported for the differences between the quantiles
of 0.05, 0.25 and 0.50 with the highest quantile of 0.95.18 The result for
heterogeneity test clearly rejects the null hypothesis of equality of slopes
for the linear model and multiscale model at scale 4, 5 and 6. The results
also confirm the significant deviances from the OLS model estimates as
shown in Figure 3, thus the application of QR model is justified. Hence, it
can be concluded that the pattern of dependency between Bitcoin and
global factors is dynamic across market conditions and investment
horizons.
Table 6. Heterogeneity Test (Wald Test) for Equality of Slopes

(a) Linear Model
TS p-Value
0.05-0.95 38.42*** 0.00
0.25-0.95 59.10*** 0.00
0.50-0.95 59.20*** 0.00
(b) Multiscale Model

Scale 1 Scale 2 Scale 3 Scale 4 Scale 5 Scale 6
TS p-Value TS p-Value TS p-Value TS p-Value TS p-Value TS p-Value
0.28 0.93 0.38 0.86 0.87 0.50 2.72** 0.02 9.62*** 0.00 54.75*** 0.00
0.64 0.67 1.49 0.19 1.60 0.16 6.04*** 0.00 18.76*** 0.00 71.93*** 0.00
1.10 0.36 1.92 0.09 1.71 0.13 7.06*** 0.00 23.17*** 0.00 73.48*** 0.00
Note: TS stand for the Test Statistic.

18
The full result of the Wald test is available upon request to the authors.
5 Conclusions
Since 2008 a remarkable upsurge in usage of virtual currencies such as

Bitcoin is evident on account of its peculiar features (Dyhrberg, 2016).
Among the other prominent utilities of Bitcoin, usage of Bitcoin for the
purpose of investment has attracted considerable attention of economic
participants and the research community. Further, Bitcoin has gained the
recognition of an effective diversifier instrument (Bouri, Molnár, et al.,
2017), during the turbulent economic environment (Stanchfield, 2016).
Thus, including Bitcoin in a portfolio as an investment commodity has
some potential benefits such as higher returns and lower correlations with
other major asset classes (Eisl et al., 2015; Wu & Pandey, 2014).
A prudent management of portfolio risks and sound allocation of assets
stress upon the need for modeling the dependence structure (between an
asset of interest with influential macroeconomic factors) across
market/economic conditions and investment horizons (Mensi et al., 2016).
Hence, we attempt to model the relationship between Bitcoin and the
global factors. In analyzing the relationships, we adopt a multiscale
approach owing to the presence of heterogeneous groups of investors in
the market. The market participants differ in terms of risk appetite,
expectations, investment horizons and sentiments. The aggregate
information set that an investor holds may stand significantly different
from the other. Thus, for a given market event, the reactions might be
distinctive and so the reaction time may vary. The difference in reaction
time leads to the multiscale feature of the economic relationships i.e.
economic relationships vary across timescales (short-run to long-run).
Therefore, the multiscale analysis is imperative to capture the dynamic
economic relationships. In addition to multiscale analysis, we also examine
the asymmetric relationship since price response of an asset differs with
positive and negative economic shocks.
Our results for wavelet coherence analysis show that the global factors
are significantly coherent in the medium to longer-run at timescales of
64-256 days. Thus, in the short run Bitcoin prices do not appear to co-move
with global factors. Hence, Bitcoin can be of use to diversify the portfolio
of investors in the short run.
Similar results are observed when the data is decomposed in various
frequencies to accommodate the heterogeneous market hypothesis and

analyzed on a regression-based framework. The results report significant
temporal cyclical fluctuations. In the short run the Bitcoin’s higher tail
dependence is observed, which signifies a crucial role of EPU in driving up
the Bitcoin prices, which is consistent with the findings of Demir et al.,
(2018).
On the other hand, both lower and higher tail dependence of Bitcoin
prices on WTI is revealed. It implies that the Bitcoin price is impacted by
oil price (WTI) changes both in bullish and bearish market conditions. This
phenomenon appears logical if the hypothesis of “oil price rise leading
inflation”is held valid. In the medium run DJIA also becomes a significant
variable to influence Bitcoin prices. All the variables become significant in
the long run.
Moreover, an asymmetric relationship is also observed significant for
WTI. Though scholars (Bouoiyour & Selmi, 2015) argue that Bitcoin is an
unconventional commodity with features distinguishable from other
standard assets, thus the macroeconomic factors may fail to explain the
causality in price fluctuations. We argue that Bitcoin being a part of the
economic system (formal or informal) cannot be completely isolated from
other influential economic factors. Thus, the decision of market
participants to trade on Bitcoin may be stimulated by a host of economic
and visceral determinants. Our result models the relationship of Bitcoin
with global factors, which can be of great importance for the market
participants who trade in Bitcoin to understand the market dynamics and
to take informed decisions.
References
Addison, P., The illustrated wavelet transform handbook, Institute of Physics
Publication, Philadelphia, 2002.
Aguiar-Conraria, L., Azevedo, N. and Soares, M. J., “Using wavelets to
decompose the time–frequency effects of monetary policy,”Physica A:
Statistical mechanics and its Applications 387(12), 2008, 2863–2878.
Aguiar-Conraria, L. and Joana Soares, M., “Business cycle synchronization
and the Euro: A wavelet analysis,” Journal of Macroeconomics 33(3),
2011, 477–489.
Aloui, C. and Hkiri, B., “Co-movements of GCC emerging stock markets:

New evidence from wavelet coherence analysis,”Economic Modelling
36, 2014, 421–431.
Baek, C. and Elbeck, M., “Bitcoins as an investment or speculative vehicle? A
first look,”Applied Economics Letters 22(1), 2015, 30–34.
Baker, S. R., Bloom, N. and Davis, S. J., “Measuring Economic Policy
Uncertainty,” The Quarterly Journal of Economics 131(4), 2016, 1593–
1636.
Baur, D. G., “The structure and degree of dependence: a quantile regression
approach,”Journal of Banking & Finance 37(3), 2013, 786–798.
Bekaert, G. and Harvey, C. R., “Emerging equity market volatility,”Journal of
Financial Economics 43(1), 1997, 29–77.
Boako, G. and Alagidede, P., “Co-movement of Africa’s equity markets:
Regional and global analysis in the frequency–time domains,”Physica
A: Statistical Mechanics and its Applications 468, 2017, 359–380.
Bouoiyour, J. and Selmi, R., “What Does Bitcoin Look Like?,” Annals of
Economics & Finance 16(2), 2015, 449-492.
Bouri, E., Jalkh, N., Molnár, P. and Roubaud, D., “Bitcoin for energy
commodities before and after the December 2013 crash: diversifier,
hedge or safe haven?,”Applied Economics 49(50), 2017, 5063-5073.
Bouri, E., Gupta, R., Tiwari, A. K. and Roubaud, D., “Does Bitcoin hedge
global uncertainty? Evidence from wavelet-based quantile-in-quantile
regressions,”Finance Research Letters 23, 2017, 87–95.
Bouri, E., Molnár, P., Azzi, G., Roubaud, D. and Hagfors, L., “On the hedge and
safe haven properties of Bitcoin: Is it really more than a diversifier?,”
Finance Research Letters 20, 2017, 192–198.
Bruce, A. and Gao, H. Y., Applied wavelet analysis with S-plus.
Springer-Verlag NewYork, Inc. 1996.
Buchholz, M., Delaney, J. and Warren, J., “Bits and Bets, Information, Price
Volatility, and Demand for Bitcoin,”Economics 312, 2012, 1-48.
Buchinsky, M., “Estimating the asymptotic covariance matrix for quantile
regression models a Monte Carlo study,”Journal of Econometrics 68(2),
1995, 303-338.
Carrick, J., “Bitcoin as a Complement to Emerging Market Currencies,”
Emerging Markets Finance and Trade 52(10), 2016, 2321–2334.
Ciaian, P., Rajcaniova, M. and Kancs, D. A., “The economics of BitCoin price
formation,”Applied Economics 48(19), 2016, 1799–1815.
Cornish, C. R., Bretherton, C. S. and Percival, D. B., “Maximal overlap wavelet
statistical analysis with application to atmospheric turbulence,”
Boundary-Layer Meteorology 119(2), 2006, 339–374.
Das, D., Kannadhasan, M., Al-Yahyaee, K. H. and Yoon, S. M., “A wavelet
analysis of co-movements in Asian gold markets,” Physica A:
Statistical Mechanics and its Applications 492, 2018. 192–206.
Das, D., Kannadhasan, M., Tiwari, A. K. and Al-Yahyaee., “Has co-movement
dynamics in emerging stock markets changed after global financial
crisis? New evidence from wavelet analysis,” Applied Economics
Letters 25(20), 2018, 1447-1453.
Das, D., Bhatia, V., Pillai, J. and Tiwari, A. K., “The relationship between oil
prices and US economy revisited,”Energy Sources Part B: Economics,
Planning, and Policy 13(1), 2018, 37–45.
Das, D., Bhowmik, P. and Jana, R. K, “A multiscale analysis of stock return
co-movements and spillovers: Evidence from Pacific developed
markets,” Physica A: Statistical Mechanics and its Applications, 502,
2018, 379-393.
Das, D. and Kumar, S. B., “International economic policy uncertainty and
stock prices revisited: Multiple and Partial wavelet approach,”
Economics Letters 164, 2018, 100–108.
Demir, E. et al. (2018) ‘Does economic policy uncertainty predict the Bitcoin
returns? An empirical investigation’, Finance Research Letters 26, 2018,
145-149.
Dickey, D. A. and Fuller, W. A., “Distribution of the estimators for
autoregressive time series with a unit root,” Journal of the American
statistical association 74(366a), 1979, 427–431.
Dyhrberg, A. H., “Bitcoin, gold and the dollar–A GARCH volatility analysis,”
Finance Research Letters 16, 2016, 85–92.
Eisl, A., Gasser, S. M. and Weinmayer, K., “Caveat Emptor: Does Bitcoin
Improve Portfolio Diversification?,”SSRN, 2015, 1-21.
Gallegati, M., “Wavelet analysis of stock returns and aggregate economic
activity,” Computational Statistics and Data Analysis, 52(6), 2008,
3061–3074.
Gandal, N. and Halaburda, H., “Competition in the Cryptocurrency Market,”

Bank of Canada Working Paper 2014-33, 2014.
Gencay, R., Selcuk, F. and Whitcher, B. An introduction to wavelets and other
filtering methods in finance and economics, Academic Press, 2002.
Graham, M., Kiviaho, J. and Nikkinen, J., “Integration of 22 emerging stock
markets: A three-dimensional analysis,” Global Finance Journal 23(1),
2012, 34–47.
Granger, C. W. and G. Yoon., “Hidden cointegration. University of California
San Diego,”Economics working paper series 2, 2002, 1-48.
Grinsted, A., Moore, J. C. and Jevrejeva, S., “Nonlinear Processes in
Geophysics Application of the cross wavelet transform and wavelet
coherence to geophysical time series,” Nonlinear Processes in
Geophysics 11(5/6), 2004, 561–566.
Gros, D. and Alcidi, C., “The impact of the financial crisis on the real economy’,
Intereconomics 45(1), 2010, 4–20.
Grubel, H. G., “Internationally Diversified Portfolios: Welfare Gains and
Capital Flows,” The American Economic Review 58(5), 1968, 1299–
1314.
Hoffmann, A. O. I., Post, T. and Pennings, J. M. E., “Individual investor
perceptions and behavior during the financial crisis,” Journal of
Banking & Finance 37(1), 2013, 60–74.
Jiang, Y., Nie, H. and Monginsidi, J. Y., “Co-movement of ASEAN stock
markets: New evidence from wavelet and VMD-based copula tests,”
Economic Modelling 64, 2017, 384–398.
Karlsson, H. K., Karlsson, P., Månsson, K. and Sjölander, P., “Wavelet quantile
analysis of asymmetric pricing on the Swedish power market,”
Empirica 44(2), 2017, 249–260.
Katsiampa, P., “Volatility estimation for Bitcoin: A comparison of GARCH
models,”Economics Letters 158, 2017, 3–6.
Khalfaoui, R. and Boutahar, M., A time-scale analysis of systematic risk:
wavelet-based approach. University Library of Munich, Germany,
2011.
Ko, J. and Lee, C., “International economic policy uncertainty and stock
prices: Wavelet approach,”Economics Letters 134, 2015, 118–122.
Koenker and D’Orey, V., “Algorithm AS 229: Computing Regression Quantiles,”
Journal of the Royal Statistical Society. Series C (Applied Statistics)

36(3), 1987, 383–393.
Koenker, R. and Bassett, G., “Regression Quantiles,”Econometrica, 46(1), 1978,
33–50.
Koenker, R. and Bassett Jr, G., “Robust tests for heteroscedasticity based on
regression quantiles,” Econometrica: Journal of the Econometric
Society 50 (1), 1982, 43–61.
Kristoufek, L., “BitCoin meets Google Trends and Wikipedia: Quantifying the
relationship between phenomena of the Internet era,”Scientific reports
Nature Publishing Group 3 (3415), 2013, 1-7.
Kristoufek, L., “What are the main drivers of the bitcoin price? Evidence from
wavelet coherence analysis,”PLoS ONE, 10(4), 2015, 1–15.
Krugman, P. R., International economics: Theory and policy, 8/E. Pearson
Education India, 2008.
Lehkonen, H., “Stock Market Integration and the Global Financial Crisis,”
Review of Finance 19(5), 2015, 2039–2094.
Lehkonen, H. and Heimonen, K., “Timescale-dependent stock market
comovement: BRICs vs. developed markets,” Journal of Empirical
Finance 28, 2014, 90–103.
Levy, H. and Sarnat, M., “International Diversification of Investment
Portfolios,”The American Economic Review 60(4), 1970, 668–675.
Martens, M. and Poon, S. H., “Returns synchronization and daily correlation
dynamics between international stock markets,” Journal of Banking
and Finance 25(10), 2001, 1805–1827.
Mensi, W., Hammoudeh, S., Reboredo, J. C., & Nguyen, D. K., “Do global
factors impact BRICS stock markets? A quantile regression approach,”
Emerging Markets Review 19, 2014, 1–17.
Mensi, W., Hammoudeh, S. and Tiwari, A. K., “New evidence on hedges and
safe havens for Gulf stock markets using the wavelet-based quantile,”
Emerging Markets Review 28, 2016, 155–183.
Nadarajah, S. and Chu, J., “On the inefficiency of Bitcoin,”Economics Letters
150, 2017, 6–9.
Nakamoto, S., “Bitcoin: A peer-to-peer electronic cash system,”www.bitcoin.
org, 2008, 1-9.
Nusair, S. A. and Al-Khasawneh, J. A., “Oil price shocks and stock market
returns of the GCC countries: empirical evidence from quantile

regression analysis’, Economic Change and Restructuring 51, 2018,
339–372.
Palombizio, E. and Morris, I., “Forecasting exchange rates using leading
economic indicators,”Open Access Scientific Reports 1(8), 2012, 1–6.
Percival, D. B. and Walden, A. T., Wavelet methods for time series analysis
Volume 4, Cambridge University Press, 2000.
Phillips, P. C. B. and Perron, P., “Testing for a unit root in time series regression,”
Biometrika. Oxford University Press 75(2), 1988, 335–346.
Popper, N., Digital gold: The untold story of Bitcoin. Penguin UK, 2015.
Priestley, M. B., Spectral analysis and time series, Academic press, 1981.
Ramsey, J., “Wavelets in economics and finance: Past and future,” Studies in
Nonlinear Dynamics & Econometrics 6(3), 2002, 1–27.
Reboredo, J. C. and Rivera-Castro, M. A., “A wavelet decomposition approach
to crude oil price and exchange rate dependence,” Economic
Modelling 32(1), 2013, 42–57.
Rogojanu, A. and Badea, L., “The issue of competing currencies,”Theoretical
and Applied Economics 21(1), 2014, 103–114.
Rua, A. and Nunes, L. C., “International comovement of stock market returns:
A wavelet analysis,”Journal of Empirical Finance 16(4), 2009, 632–639.
Stanchfield, D., “The Strange Positive Effect Political Uncertainly Has on
Bitcoin,” Available at: https://www.entrepreneur.com/article/280779,
2016.
Tiwari, A. K., Jana, R. K., Das, D. and Roubaud, D., “Informational efficiency
of Bitcoin—An extension,”Economics Letters 163, 106–109.
Torrence, C. and Compo, G. P., “A practical guide to wavelet analysis,”Bulletin
of the American Meteorological Society 79(1), 1998, 61–78.
Torrence, C. and Webster, P. J., “Interdecadal changes in the ENSO-monsoon
system,”Journal of Climate 12(8-2), 1999, 2679–2690.
Urquhart, A., “The inefficiency of Bitcoin,” Economics Letters 148, 2016, 80–
82.
Urquhart, A., “Price clustering in Bitcoin,”Economics Letters 159, 2017, 145–
148.
Vacha, L. and Barunik, J., “Co-movement of energy commodities revisited:
Evidence from wavelet coherence analysis,” Energy Economics 34(1),

2012, 241–247.
Valstad, O. C. A. and Vagstad, K., A bit risky? A comparison between Bitcoin
and other assets using an intraday Value at Risk approach. Norwegian
University of Science and Technology, Norway, 2014.
Weber, B., “Bitcoin and the legitimacy crisis of money,”Cambridge Journal of
Economics, Oxford University Press UK 40(1), 2014, 17–41.
van Wijk, D., “What can be expected from the BitCoin,”Erasmus Universiteit
Rotterdam Working Paper 345986, Erasmus University: Rotterdam,
Netherlands, 2013.
Wu, C. Y. and Pandey, V. K., “The value of Bitcoin in enhancing the efficiency of
an investor’s portfolio,”Journal of financial planning 27(9), 2014, 44–52.
Yap, J. T., Cuenca, J. S. and Reyes, C. M.,“Impact of the global financial and
economic crisis on the Philippines,”Philippine Institute of Development
Studies Discussion Paper Series (2009-30). Philippines, 2009.
Yermack, D., “Is Bitcoin a real currency? An economic appraisal,” NBER
Working Paper 19747, National Bureau of Economic Research:
Cambridge, Massachusetts, 2013.
Zhu, Y., Dickinson, D. and Li, J., “Analysis on the influence factors of Bitcoin’s
price based onVEC model,”Financial Innovation 3(3), 2017, 1-13.
View publication stats

Do Global Factors Impact Bitcoin Prices

Uploaded by

Copyright:

Available Formats

Do Global Factors Impact Bitcoin Prices

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Do Global Factors Impact Bitcoin Prices

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Do global factors impact bitcoin prices? evidence from wavelet

Article · November 2018

Debojyoti Das Manoharan Kannadhasan

SEE PROFILE SEE PROFILE

The user has requested enhancement of the downloaded file.

Do global factors impact bitcoin prices? evidence from

In this article, we attempt to delineate the relationship between bitcoin

Keywords: Bitcoin; Co-movement; Wavelet; Quantile Regression; Global Factors

JEL Classification: G15, Q02

First received, July 5, 2018; Accepted, November 5, 2018.

Bitcoin is a most innovative form of “digital currency” first introduced

investment portfolio as a diversifying instrument yields considerable

potential dangers of propagating financial crisis from developed to

1.2 Motivation and literature

harbinger of inflationary conditions in an economy, besides it is also a

(e) Economic Policy Uncertainty: Finally, we include the US EPU as

Nevertheless, besides many theoretical assumptions and conjectures

(a) Wavelet coherence analysis: In the first stage, we employ continuous

(b) Quantile regression analysis: Secondly, we employ the quantile regression

(c) Wavelet-based quantile regression analysis: The hybrid model of discrete

(d) Asymmetric quantile regression analysis: This model enables us to

1.3 Main contributions

We contribute to the literature in the following way: (a) In the first

Figure 1. Risk-returns plot of assets and price series plot of bitcoin

(a) Risk-Returns Plot

(b) Bitcoin Price

2.1 Wavelet coherence analysis

Wavelet Coherence Analysis (WCA) is one of the most preferred forms of

and a dilation parameter s, which represents the scale that is associated

The normalization factor is denoted by , which ensures the

(i) The mother wavelet must have a mean of zero ( ) = 0.

Thus, the CWT of a time-series ( ) with respect to ( ) is given by

The complex conjugate is denoted by * in equation (3). In case of a

The computation of wavelet transform in time domain is possible using

For an appropriate the term − becomes negligible, hence the

The equation (6) with the corresponding Fourier Transformation process

From equation (7) it may be observed that a Morlet wavelet within a

Now, given a set of two time-series ( ) and ( ), the cross-wavelet

Where denotes the time and scale smoothing operator (Torrence

2.2 Wavelet multi-scale analysis

Wavelets are capable of separating a signal into multi-horizon

The father wavelet is mathematically expressed as follows:

whereas, the mother wavelet is defined as below:

where, j and k represent respective scale and translation parameters.

The detail coefficients of the mother wavelets are expressed as:

Thus, the wavelet representation of a function is defined

The equation (5) may be simplified as:

with the orthogonal components as below:

From equation (15), is the resulting multiscale

The maximum overlap discrete wavelet transform (MODWT) is applied

Table 1.Time interpretation of different frequencies

Scale 2 4~8 days

Scale 3 8~16 days

Scale 4 16~32 days

Scale 5 32~64 days

Scale 6 64~128 days

2.3 Quantile regression

The traditional regression models establish a symmetric linear

DJIA -5.399* -7.911* -5.875* -2.289* -1.852* -4.241* -10.47* -10.87*

Constant -0 -0.025* -0.016* -0.0059* -0.00057* 0.00461* 0.0152* 0.0260***