Energy Economics 95 (2021) 105092

Contents lists available at ScienceDirect

Energy Economics

journal homepage: www.elsevier.com/locate/eneeco

A closer look into the global determinants of oil price volatility

Ioannis Chatziantoniou a,⁎, Michail Filippidis a, George Filis b, David Gabauer c
a
Economics and Finance Subject Group, University of Portsmouth, Portsmouth Business School, Portland Street, Portsmouth PO1 3DE, United Kingdom
b
Department of Economics, University of Patras, University Campus 265 04, Rio, Patras, Greece.
c
Software Competence Center Hagenberg, Data Analysis Systems, Softwarepark 21, 4232 Hagenberg, Austria

a r t i c l e i n f o a b s t r a c t

Article history: In this paper we investigate global determinants of oil price volatility by employing a time-varying parameter
Received 27 May 2020 vector autoregressive (TVP-VAR) model. We focus on realised volatility and consider the impact from a set of po-
Received in revised form 1 November 2020 tential determinants including oil supply, oil demand, oil inventory, financial market uncertainty, financial inter-
Accepted 27 December 2020
bank stress, as well as, financial trends in different currencies. We investigate the impact of these factors on
Available online 10 January 2021
realised volatility utilising monthly data over the period 1990:1–2019:5. Findings show that all factors can be
Keywords:
conducive to higher levels of realised oil price volatility particularly in the short run. What can further be noticed,
Financial indicators is that the magnitude of the corresponding impulse response functions may differ across time and this could
Oil-market specific fundamental factors largely be attributed to specific intervals of financial crises and economic recessions. Nevertheless, we show
Oil price volatility that shocks originating to the financial markets tend to be more important for oil price volatility. Our findings
TVP-VAR are closely linked to the implications regarding the financialisation of the oil market.
© 2021 Elsevier B.V. All rights reserved.
JEL codes:
C32
C51
Q40
Q43

1. Introduction & Brief Review of the Literature participants – including hedge funds, insurance companies and pension
funds (see, inter alia, Fattouh et al. 2013). It should also be noted that,
The examination of oil price volatility is essential as it helps improve existing literature in the field of forecasting has so far effectively re-
our understanding in connection with changes in the level of uncer- ported the importance of oil market specific variables such as world
tainty in the market for oil, which are typically reflected upon abrupt crude oil production, US crude oil inventories, US petroleum inventories
fluctuations in the price of oil. Clearly, oil price volatility contains infor- and NYMEX oil futures prices, for forecasting the price of oil (see, for in-
mation which is closely associated with developments in both the phys- stance, Baumeister and Kilian, 2012); nonetheless, such variables have
ical and the futures oil market. In turn, such information could rather been neglected when it comes to making predictions of oil
potentially be utilised by energy traders who seek to effectively manage price volatility. That is, existing literature in this field, mainly uses his-
their asset portfolios and design hedging strategies. It follows that, oil torical information in the form of price jumps rather than oil market
price volatility has received considerable attention by the research com- fundamentals to produce forecasts of oil price volatility (see
munity in recent years. Prokopczuk et al., 2016).
In fact, the investigation of oil price volatility has been the focal point Recent studies have also emphasized the role of oil price volatility as
of many studies in the field of forecasting. More particularly, we high- a leading macroeconomic indicator, since it provides significant infor-
light the recent studies by authors such as Efimova and Serletis mation to energy traders, financial market participants and
(2014); Phan et al. (2016); Degiannakis and Filis (2017); policymakers (see Efimova and Serletis 2014). Indeed, in an early
Chatziantoniou et al. (2019) who employ a variety of forecasting speci- study, Ferderer (1996) provides evidence that oil price volatility con-
fications in order to produce more accurate forecasts of oil price volatil- tains information that helps forecast industrial production growth in
ity. The general consensus from this strand of the literature is that oil the US. Similarly, Pindyck (2004) explains that higher oil price volatility
price volatility forecasting has become increasingly more important in has a positive impact on demand for storage and thus subsequently af-
recent years mainly due to the financialisation of the oil markets and fects both the spot price of oil and the convenience yield. Furthermore,
the fact that oil is now largely regarded as a financial asset by market in a recent study, Elder and Serletis (2010) find that oil price volatility
has a negative impact on the US aggregate output, investment and con-
⁎ Corresponding author. sumption. In addition, Henriques and Sadorsky (2011) indicate that oil
E-mail address: [email protected] (I. Chatziantoniou). price volatility has an impact on the strategic investment decisions of

https://doi.org/10.1016/j.eneco.2020.105092
0140-9883/© 2021 Elsevier B.V. All rights reserved.
I. Chatziantoniou, M. Filippidis, G. Filis et al. Energy Economics 95 (2021) 105092

US firms. They document that increased oil price volatility affects the futures markets. According to Tang and Xiong (2012), this investment
cost of oil inputs and generates uncertainty not only around investment has grown from $15 billion to $200 billion between 2003 and mid-
decisions but also with regard to firm valuation and firm profitability. In 2008, whereas Irwin and Sanders (2011) maintain that this investment
addition, Diaz et al. (2016) provide evidence that higher levels of oil raised to $250 billion in 2009, and further to $450 in 2011, as docu-
price volatility negatively affect stock market returns in G7 economies. mented by Bicchetti and Maystre (2013). Overall, this recent develop-
Finally, Bouri et al. (2018c) report a significant effect of oil price volatil- ment recognises that oil futures market derivatives can be used as
ity on the sovereign credit risk of BRICS countries. In contrast with stud- financial assets in the attention of energy traders and portfolio man-
ies that assume a leading indicator role for oil price volatility, one of the agers who appear to increase their positions in the oil futures market
objectives of this study is to investigate how well-established financial both prior to the Global Financial Crisis of 2007–08 (henceforth, GFC
indicators such as the global financial market volatility, the global inter- 2007) and the period afterwards.
est rate or the global exchange rate influence oil price volatility per se. It should also be noted that, we focus on realised oil price volatility as
In this relatively scarce strand of the literature in connection with opposed to conditional oil price volatility; although we do consider the
the investigation of factors that determine oil price volatility, Van latter alternative for robustness purposes. This decision is primarily
Robays (2016) opines that increased macroeconomic uncertainty trig- based on Andersen and Bollerslev (1998); Andersen et al. (2003) as
gered by recessions and financial crises is rather conducive to higher well as Hansen and Lunde (2006) who provide evidence that realised
levels of oil price uncertainty. More specifically, Van Robays (2016) ex- volatility is both a more accurate measure and more reliable in produc-
plains that during periods of increased macroeconomic uncertainty, oil ing forecasts. In a relevant study involving the oil market, Herrera et al.
price volatility is driven by fundamental factors underpinning oil supply (2018) also focus on realised oil price volatility arguing that the under-
and demand shocks, which in turn, are largely associated with delays in lying volatility of crude oil returns is rather unobserved.
production and consumption decisions from market agents. This is We contribute to existing literature in two ways. First, we investi-
mainly because, higher macroeconomic uncertainty lowers the price gate the said relationships utilising a time-varying parameter vector
elasticity of oil supply and oil demand and consequently increases oil autoregressive (TVP-VAR) model. By allowing all parameters to vary
price volatility. over time, we are able to capture regular or unexpected variations to
Another interesting aspect with regard to oil price volatility and its those parameters, securing both flexibility and accuracy. Second, we
determinants, relates to speculation in the market for oil and how the make use of a unique set of factors such as global oil supply, global oil
latter could potentially result in increased volatility. Van Robays demand, global oil inventory, global financial market uncertainty, global
(2016) argues that, changes in the price elasticity of oil supply and oil financial interbank stress and global financial trends in currencies. In
demand should not be attributed to changes in oil inventory holding- this regard, we provide additional evidence on oil price volatility consid-
levels and that speculation should not really be regarded as a factor ering relationships that have so far been largely under-researched.
that contributes to changes in oil price volatility. Nonetheless, authors Our empirical findings suggest that all potential determinants affect
such as Beidas-Strom and Pescatori (2014) who utilise data on global the realised volatility of oil, mainly at short run horizons. In addition,
oil inventories to approximate shifts in speculative demand for oil find impulse response functions vary over time and this might very well be
a link between an increase in oil inventories and that in the price of attributed to changing economic conditions across time. More particu-
oil. They conclude that financial speculation triggers short term oil larly, we note that the impact is stronger and shows peaks during severe
price fluctuations between 3 and 22%. episodes of economic recessions and financial crises. Overall, our results
What is more, Robe and Wallen (2016) investigate whether physical indicate that high levels of oil price volatility might indeed be deter-
market fundamentals, financial and macroeconomic conditions drive oil mined both by oil fundamental and financial factors. More importantly,
price volatility at horizons of one to six months. Robe and Wallen it has been illustrated that the greatest impact on oil price volatility is
(2016) use the term structure of oil option-implied volatilities and re- received by financial factors rather that fundamental factors.
port that the VIX index significantly affects oil implied volatility, The remainder of the paper is structured as follows: In Section 2 we
whereas both other macroeconomic variables and speculative activity describe the dataset and discuss the variables under investigation. In
do not seem to have a significant impact. On a parallel note, Caldara Section 3 we present the econometric methods that we employ in this
et al. (2019) examine sources of oil price movements. They indicate study. In turn, in Section 4 we report and discuss the empirical results
the importance of oil supply shocks and global demand shocks in driv- of the study. Finally, Section 5 summarises and concludes the paper.
ing oil price fluctuations. Specifically, they report that oil supply shocks
and global demand shocks contribute to explain 50 and 35% of oil price 2. Data Description & Preliminary Analysis
fluctuations respectively.
Following from the above, the objective in this study is to focus on 2.1. Explanatory variables
global determinants of oil price volatility in an effort to shed additional
light upon the underlying factors of oil price fluctuations. To this end, In this study, we use monthly data from January 1990 to May 2019
we consider a broad set of factors such as global oil supply, global oil de- which is translated into 353 observations. The choice of the monthly
mand, global oil inventories, global financial market uncertainty, global time series and the sample period is governed by the data availability.1
financial interbank stress and global financial trends in currencies. To be Additionally, the choice of the time period is motivated by the most sig-
more explicit, we examine whether unanticipated changes both in nificant changes and developments in the crude oil market.2 We employ
global, crude oil market-specific fundamental factors and in financial in- a set of potential global determinants of the oil price volatility which is
dicators are useful in explaining movements in oil price volatility and used as a proxy of the crude oil market uncertainty. More specifically,
whether they subsequently help capture the uncertainty generated in we collect daily data on the Brent crude oil price which will be used to
the crude market for oil. compute their monthly volatilities. The choice of Brent is attributed to
Our decision to employ fundamental and financial factors is strongly
motivated by the fact that recent empirical evidence indicates that oil 1
More specifically, the start date is attributed to the fact that the VIX index data series is
price fluctuations are not only traditionally determined by the funda- available from January 1990, whereas the end date is chosen due to the fact that data on
mental factors of oil supply and oil demand but also driven by financial the global economic activity index were available until May 2019 the period when we col-
activity (see, for example Tang and Xiong 2012; Irwin and Sanders lected the sample data.
2
Crude oil prices react to a variety of geopolitical, financial and macroeconomic events.
2011). Indeed, the financialisation of the oil market refers to the in- For more information about the oil price chronology from 1970, the reader is referred to
creased participations of hedge funds, insurance companies and pen- the EIA in the following link: http://www.eia.gov/finance/markets/reports_
sion funds, among other financial participants, in the commodity presentations/eia_what_drives_crude_oil_prices.pdf

2
I. Chatziantoniou, M. Filippidis, G. Filis et al. Energy Economics 95 (2021) 105092

the fact that it is widely considered as the global crude oil benchmark. (2013); Singleton (2014); Büyüksahin et al. (2009). Even more, an in-
Specifically, Brent is used to price crude oil that is produced and traded depth review on these effects can be found in Haase et al. (2016).
not only in Europe, the Mediterranean, and Africa, but also in Australia In addition, the volatility index (VIX) is used to approximate the
and some countries in Asia.3 Data for the Brent crude oil price are ex- global uncertainty in financial markets. The VIX index is provided by
tracted from Datastream and are expressed in dollar terms. the CBOE and serves as a key measure of the expected volatility of the
The crude oil-market specific factor of global oil supply is repre- underlying S&P 500 stock index that reflects the market's expectation
sented by the world oil production. Based on Kilian (2009) empirical of 30-day volatility. However, the VIX index is also regarded as a bench-
work, we use this variable to evaluate unexpected changes in global mark not only for the S&P 500 stock index but also for the whole US
oil supply. For example, an unexpected disruption in the world oil pro- stock market and further as a global measurement of market stress. In
duction attributed to OPEC's decision to cut its production quotas, leads particular, Cai et al. (2009) argue that the VIX index is considered as
to increases in oil prices and linked with fluctuations in oil price volatil- the world's premier barometer to gauge investor fear. In addition,
ity. Data for the world oil production have been obtained from the En- Sarwar (2012) documents that the VIX index measures the market
ergy Information Administration (EIA). fear in stock markets outside the US, such as those of Brazil, China and
In the same vein, the global economic activity index is used to ap- India. Finally, Bouri et al. (2018a, 2018b) argue that the US VIX index
proximate the crude oil-market specific factor of global oil demand. Ac- is a gauge of fear for emerging BRICS economies due to the massive
cording to Kilian (2009), this index is designed to record shifts in the size of the US stock market and the dominant role of the US VIX index
demand for industrial commodities responding to the global business to highly predict stock market returns and volatility in emerging mar-
cycle. Kilian and Murphy (2014) consider that this index is the only kets. Given these characteristics, an increase in the expected stock mar-
one appropriate indicator among other alternative proxies in the area ket volatility results in negative market returns in international
of industrial commodities of capturing shifts in global demand. An ad- markets. Therefore, decreasing asset prices (including commodity/oil
vantage of this index is the inclusion of the real economic activity in prices) are anticipated which implies deviations in oil price volatility.
the emerging economies of India and China (Kilian and Park 2009). Data for the VIX are provided from Datastream.
We expect that an unexpected rise in global economic activity will Moreover, we use the TED spread as a proxy of global financial inter-
drive up oil prices and associated with fluctuations in oil price volatility. bank stress which denotes interest rate fluctuations. The TED spread is
The data for the global economic activity index are retrieved from Lutz the difference between the three-month US London Interbank Offered
Kilian's website.4 Rate (LIBOR) and the three-month US Treasury bill rate. It is used as
Based on the fact that there is no open data on global oil inventories, an indicator of global interbank market stress which means that a rising
we follow Wang et al. (2017) who use the total US crude oil inventory TED spread signifies a higher risk of bank defaults and hence a lower
including strategic reserves to approximate the factor of global oil in- economic activity. According to Basher et al. (2012), the TED spread
ventories. The authors justify this choice by arguing that this is because can be used to approximate global interest rate movements and there-
the world total level of inventory is unavailable. We suggest that global fore to capture developments in future economic activity. In this regard,
oil inventories can be regarded as an additional crude oil-market spe- unexpected higher interest rates tend to rise the borrowing costs of oil
cific factor. Since oil is a storable asset, unexpected changes in the de- companies and therefore affect negatively their revenues and profitabil-
mand for inventory are expected to affect movements in oil prices. For ity. Since oil extraction requires a huge amount of capital, higher inter-
example, if shortages of barrels of oil are more likely to occur in the est rates reduce such investment and therefore limit future economic
near future, an increase in the inventory demand in the oil physical activity. This process is translated into a lower demand for oil and
spot market is likely to happen. The accumulation of inventories leads hence a fall in oil price which suggests an alteration in oil price volatility.
to a decline in the availability of crude oil for current use and tends to Data on TED spread are extracted from Datastream.
put an upward pressure on the oil price and related to significant varia- What is more, the trade-weighted exchange rate index represents
tions in oil price. It should be mentioned that Kilian and Murphy (2014) the weighted average of the foreign exchange value of the US dollar
provide evidence that fluctuations in global oil inventories are mainly against the most widely traded currencies in the financial markets and
driven by speculative activity and therefore reflect speculative trading. hence it reflects global financial trends in currencies. Specifically, this
In other words, they argue that the use of global oil inventories help index denotes financial trader's assessment of the dollar and used to ap-
to identify the speculative component of the real price of oil. These proximate movements in global foreign exchange rates.5 Even more,
data are collected from the EIA database. this index is used to measure financial stress on US dollar. Thus, a sud-
To elaborate further on speculative activity, Fattouh (2012) and den higher value of this index denotes that the US dollar appreciates.
Fattouh et al. (2013) document that this can be associated with the in- Since crude oil is priced in US dollars, a stronger US dollar implies a
creased participation of non-commercial traders, e.g. financial agents, higher oil supply and a lower oil demand. Therefore, the fall in oil
in the oil market. In particular, non-commercial traders, such as hedge price causes shifts in oil price volatility. Data on the trade-weighted ex-
funds or money managers, tend to take long position in the oil futures change rate index are obtained from the Federal Reserve of St. Louis da-
or accumulate oil inventories, without having a commercial interest in tabase (FRED).
oil market. Such activity tends to (i) significantly rise oil spot prices,
(ii) increase oil price volatility and consequently cause greater uncer-
tainty in the oil market, and (iii) cause a breakdown of the oil price-oil 2.2. Preliminary analysis
inventory relationship. For a detailed discussion of these aforemen-
tioned effects of speculative activity on oil markets the reader is directed Fig. 1 illustrates diagrammatically visual representation of the series
to authors such as Adams et al. (2020); Nguyen et al. (2020); Baur and under consideration. It is evident that the realised oil price volatility ex-
Dimpfl (2018); Zhang et al. (2017); Basak and Pavlova (2016); Adams hibits a peak during the GFC 2007 which begins to emerge in the second
and Glück (2015); Büyükşahin and Robe (2014); Cheng et al. (2015); half of 2008 and accords with the most serious stage of this crisis. It is
Hamilton and Wu (2014); Morana (2013); Silvennoinen and Thorp worth mentioned that a significant peak is also observed during the pe-
riod of 1990–1991 which is associated with Iraq's invasion of Kuwait

5
Based on the Board of Governors of the Federal Reserve System, the trade weighted
3
The source of the information can be found on EIA: exchange rate index of major currencies (TWEXMMTH) includes the Euro Area (euro),
https://www.eia.gov/todayinenergy/detail.php?id=18571 Canada (dollar), Japan (yen), United Kingdom (pound), Switzerland (frank), Australia
4
Global economic activity index is constructed by Lutz Kilian and the data can be found (dollar), and Sweden (krona). For more information about trade-weighted indexes, please
in his personal website: see https://sites.google.com/site/lkilian2019/research/data-sets. see http://www.federalreserve.gov/pubs/bulletin/2005/winter05_index.pdf.

3
I. Chatziantoniou, M. Filippidis, G. Filis et al. Energy Economics 95 (2021) 105092

Fig. 1. Time series employed in the study. Note: This Figure exhibits the evolution of the series during the sample period. In the first column, world crude oil production, global crude oil
inventories, TED spread and realised oil price volatility are represented. In the second column, the global economic activity index, VIX index and trade-weighted exchange rate index are
depicted. The sample period runs from January 1990 to May 2019.

and the collapse of the former Soviet Union which was one of the largest absence of financial turmoil, the VIX index experienced very low levels.
oil-producer in the world. However, a higher oil price volatility is also This low risk environment was put at an end during the GFC 2007, when
spotted in 1998–1999 which coincides with two oil production cuts the VIX index reached unprecedented levels. Since then, the VIX index
by OPEC in order to cause an end to declining oil prices, in late-2001 has reverted back to its pre-crisis levels. Similarly, the TED spread
which is associated with 9/11 terrorist attack and 2016 which is attrib- shows a historically significant peak during the GFC 2007 period attrib-
uted to speculation followed the OPEC's decision to cut production uted to solvency problems and restrictions to interbank lending by lead-
quota. ing banks. It is also evident that both in the period before and after the
With reference to world crude oil production, there is a gradually in- GFC 2007, the TED spread was trading at fairly low basis points which
creasing pattern for most of the period of investigation. However, there is indicative of stability in the financial system and the lack of financial
are only few periods of notable declining patterns such as the early- stress. Indeed, although the aforementioned episodes of the Asian finan-
1998 (OPEC decisions), in 2000 (recession in the US market), during cial crisis, the internet bubble burst and the 9/11 terrorist attack were
the GFC 2007 and the beginning of the Arab Spring in 2011. Global eco- suggestive of a relatively greater financial uncertainty prior to the GFC
nomic activity index experiences a boom since 2002 which is 2007, it appears that markets recognise only a small amount of financial
characterised by a surge in demand for industrial commodities mainly risk after the GFC 2007 and even during the European sovereign debt
by emerging economies such as China and India. Nevertheless, global crisis.
economic activity index exhibits a significant drop during the GFC Finally, turning to the trade-weighted exchange rate index, a peak is
2007 and further in 2016 which can be influenced by a temporary detected in early-2002, which was the consequence of a persistent in-
weakness of the Chinese economy and other large emerging economies. creasing trend since 1995 and attributed to the strengthening of the
Regarding the global crude oil inventories, the period from the mid- US economy compared to the economic weakness of the rest of the
1990 through the early 2000 is mainly associated with low levels due world.6 Furthermore, a stronger US dollar is detected during the second
to OPEC's production cuts. In addition, the accumulation of global half of 2008, whereas a sharp upward trend is detected in mid-2014 on-
crude oil inventories in 2003 is closely associated with the rapid growth wards. According to Fratzscher (2009), the appreciation of the US dollar
in US shale oil production which was triggered by technological ad- since July 2008 is driven by the severe global financial turmoil origi-
vances in drilling. In particular, shale oil production experienced an in- nated from a negative US shock linked with the collapse of Lehman
crease in 2003 and a rapid expansion in 2009, which resulted in Brothers. This resulted in capital repatriation from foreign markets to
almost half of US crude oil production coming from the accumulation the US market and hence contributed to convert foreign currencies
of US shale oil production in 2014 (see, for instance, Kilian, 2016). Fur- into US dollars. The recent strength in the dollar value can be explained
thermore, these patterns can be also attributed to the decreasing de- by the Federal Reserve (FED) decision to raise the US interest rates in
mand for oil due to a drop in global economic activity. The rising the end of 2015.7 In addition, a second explanation is attributed to the
pattern in oil inventories is observed until early-2017, when OPEC de- European Central Bank (ECB) decision to reduce the value of euro dur-
cided to cut oil production. ing the same year. Since euro accounts for the highest weight among
Turning to the VIX index, spikes are observed during periods of fi-
nancial stress, such as the period from mid-1990 (the Asian financial cri- 6
See, for example, Blecker (2003) who explains the benefits of a lower dollar and how a
sis of 1997–98) to mid-2000 (the internet bubble burst together with higher dollar has negatively affected US manufacturing producers.
the 9/11 terrorist attack). Since 2001 and until the GFC 2007, in the 7
The interested reader can find a relevant information at https://www.usnews.com

4
I. Chatziantoniou, M. Filippidis, G. Filis et al. Energy Economics 95 (2021) 105092

qffiffiffiffi
other currencies in the value of dollar, a weaker euro implies a stronger RVm 22
(whereby τ is the number of trading days per month
t with τ
dollar.8
Having explained the most important patterns of our indices, we and 22 is the average number of trading days per month) due to the fact
proceed to the descriptive statistics analysis of our variables which are that volatility is expected to be higher during a month which has more
reported in Table 1. Turning our attention to the kurtosis, a greater pos- trading days. A more technical information regarding this estimation
sibility of extreme movements is indicated by a value higher than 3, can be found to Xekalaki and Degiannakis (2010).
which practically implies that the leptokurtic distribution practically de-
scribes two of the determinants; that is, VIX and TED spread. What is 3.2. Model specification
more, the global crude oil inventories appear to be the only variable of
the study that is negatively skewed, implying that this variable is In order to capture the underlying dynamics, we employ the time-
characterised by speedy decrease and sluggish increase. For the rest of varying parameter vector autoregression (TVP-VAR) approach of Del
the variables that are positively skewed, this could be indicative of Negro and Primiceri (2015). This model can be regarded as a generalisa-
slow-moving declines and rapid-moving upturns. In addition, according tion of the standard constant parameter VAR (see Sims 1980). In effect,
to the Jarque-Bera test, three of the series are normally distributed; we assume that the volatility and the relationship across the relevant
namely, oil global production, inventories and the trade-weighted ex- series changes over time as a result of events such as economic and fi-
change rate index. nancial crises, technological innovations, as well as, political develop-
Finally, the Augmented Dickey-Fuller (ADF) introduced by Dickey ments. In this regard, we employ the TVP-VAR framework with
and Fuller (1981) unit root test guarantees that all variables are station- stochastic volatility (SV) as this version of the VAR model is not restric-
ary. In this regard, the global economic activity index is stationary by tive and does not assume constant volatility of the coefficients of the
construction reflecting the global business cycle index (see, Kilian and model across time.
Murphy 2014). Moreover, the realised oil price volatility, the volatility Furthermore, early work by Engle (1982) and Bollerslev (1986) has
VIX index and the TED spread are stationary in level and expressed in shown that volatility considerably varies over time, while, authors such
natural logarithms. On a final note, the natural logarithms of world as Hamilton (1989); Cogley and Sargent (2005) and Primiceri (2005)
crude oil production, global crude oil inventories, and trade-weighted have shown that linkages across variables may very well differ across
exchange rate index are transformed to stationary series by taking the different periods. Therefore, the Del Negro and Primiceri (2015) frame-
first differences. work of analysis is appropriate for the investigation of the very dynamic
Table 2 reports the unconditional correlation among the variables oil market, as it allows both volatilities and coefficients to change over
under consideration based on a linear relationship. We observe a time. The TVP-VAR model is derived from the basic constant parameter
weak negative unconditional correlation between our measure of oil VAR model, with the latter defined as follows:
price volatility and both the series of world crude oil production growth
yt ¼ B1t yt−1 þ . . . þ Bpt yt−p þ A−1
t Σ t εt ð2Þ
rate and global economic activity index. In addition, a weak positive un-
conditional correlation is evident between our measure of oil price vol-
atility and the series of global crude oil inventories growth rate and the where yt denotes an m × 1 vector of observed variables, and At, Bit, i = 1,
trade-weighted exchange growth rate. Finally, a relatively moderate …, p, t = p + 1, …, T denote m × m matrices of coefficients. The compos-
positive unconditional correlation between our measure of oil price vol- ite error term, A−1
t Σt εt , implies that the residual variance-covariance

atility and the series of the volatility VIX index and the TED spread is ob- matrix varies over time. More precisely, the parameters are adjusted
served. These preliminary findings motivate our decision to proceed over time as follows:
with a time-varying framework in which the parameters are allowed
to change over time. yt ¼ X 0t Bt þ A−1
t Σt εt ð3Þ

3. Methodology Bt ¼ Bt−1 þ uB,t ð4Þ

This Section is organised as follows: Section 3.1 designates the esti- αt ¼ αt−1 þ uα,t ð5Þ
mation of oil price volatility. Section 3.2 presents the time-varying pa-
rameter vector autoregression (TVP-VAR) model. log σt ¼ log σt−1 þ uσ ,t ð6Þ

3.1. Oil Price volatility estimates where yt is an m × 1 dimensional vector stacked at a given date, Xt′
= Im ⊗ (1, yt−1′, …, yt−p′), Bt stacks all Bit, At is a lower triangular matrix
Initially, it requires to estimate the realised volatility measures of oil with ones on the diagonal and all other elements stacked in the vector
price. It should be mentioned that the realised oil price volatility as well αt, Σt is a diagonal matrix with elements σt ¼ diag ðΣt Þ, and εt,uB,t,uα,t,
as the conditional oil price volatility which is used in a robustness check and uσ,t are all independent from each other. The TVP-VAR model is es-
are regarded as current-looking market volatility measures which im- timated using Markov-Chain Monte-Carlo (MCMC) methods with
plies that oil price volatility is estimated at the current time by using Bayesian inference, based on 30,000 draws after an initial burn-in of
the most recent available information. The annualised monthly realised 30,000 (i.e., we use a total of 60,000 iterations).
volatility is computed as the square root of the sum of the squared daily The generalised impulse response functions (GIRF) developed by
prices and is shown as: Koop et al. (1996) and Pesaran and Shin (1998) are based on the
time-varying coefficient and time-varying variance-covariance matrices
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
τ  2 retrieved from the TVP-VAR. For this reason the TVP-VAR has to be
m
RV t ¼ 100 12 ∑ logPt,j −logPt,j−1 ð1Þ transformed to its vector moving average (VMA) representation via
j¼1
the Wold representation theorem which can be illustrated as follows:
p ∞
where RVm t represents the annualised monthly realised volatility and
yt ¼ ∑i¼1 Bit yt−i þ εt ¼ ∑ j¼1 Λjt εt−j þ εt .
logPtj reflects the natural logarithm of the daily market price at day j of The GIRFs represent the responses of all variables following a shock
month t. Finally, we assume equal number of trading days by scaling in variable i. Since we do not have a structural model, we compute the
differences between a K-step-ahead forecast where once variable i is
8
For more information the interested reader should refer to https://www. shocked and once where variable i is not shocked. The difference can
theguardian.com be accounted to the shock in variable i, which can be calculated by

5
I. Chatziantoniou, M. Filippidis, G. Filis et al. Energy Economics 95 (2021) 105092

Table 1
Descriptive statistics.

ΔL(PRO) GEA ΔL(INV) VIX TED ΔL(TWE) RVOL

Mean 1.141 0.037 0.59 18.846 0.431 0.296 32.125

Median 1.17 −0.069 0.403 16.82 0.341 0.52 30.309
Maximum 7.35 1.892 10.325 72.67 3.619 19.949 168.61
Minimum −4.243 −1.616 −11.479 9.31 0.061 −15.003 9.314
Std.Dev. 2.114 0.627 3.796 8.192 0.369 6.999 15.268
Skewness 0.111 0.766*** −0.061 2.499*** 3.767*** 0.174 3.112***
(0.396) (0.000) (0.637) (0.000) (0.000) (0.184) (0.000)
Kurtosis −0.031 0.552* 0.271 10.191*** 21.084*** −0.077 20.126***
(0.948) (0.057) (0.261) (0.000) (0.000) (0.905) (0.000)
JB 0.709 37.653*** 1.260 1830.460*** 7122.843*** 1.804 6305.624***
(0.702) (0.000) (0.533) (0.000) (0.000) (0.406) (0.000)
ADF −5.216*** −3.368** −3.513*** −5.637*** −5.170*** −4.914*** −7.102***
(0.000) (0.001) (0.001) (0.000) (0.000) (0.000) (0.000)

Note: This Table summarises descriptive statistics (sample mean, median, maximum, minimum, standard deviation, skewness, kurtosis, the Jarque-Bera test statistic, and the p-value as-
sociated to the Jarque-Bera test statistic) of the natural logarithm of realised oil price volatility (RVOL), the first difference of the natural logarithm of world crude oil production (ΔL(PRO)),
the global economic activity index (GEA), the first difference of the natural logarithm of global crude oil inventories (ΔL(INV)), the natural logarithm of volatility VIX index (VIX), the nat-
ural logarithm of TED spread (TED) and the first difference of the natural logarithm of trade-weighted exchange rate index (ΔL(TWE)). Regarding the JB, asterisk * (**, ***) denotes the 10%
(5%, 1%) significance level. ADF denotes the Augmented Dickey-Fuller unit root test with 10%, 5% and 1% critical values of −2.571, −2.869 and − 3.448, respectively. The ADF tests the null
hypothesis that the series features a unit root. The sample period runs from January 1990 to May 2019.

Table 2
Unconditional correlations among variables under consideration.

ΔL(PRO) GEA ΔL(INV) VIX TED ΔL(TWE) RVOL

ΔL(PRO) 1.000 0.140 −0.035 −0.150 −0.107 0.025 −0.067

GEA 0.140 1.000 −0.079 −0.022 0.260 −0.542 −0.077
ΔL(INV) −0.035 −0.079 1.000 0.049 −0.125 0.224 0.232
VIX −0.150 −0.022 0.049 1.000 0.530 0.146 0.461
TED −0.107 0.260 −0.125 0.530 1.000 0.033 0.337
ΔL(TWE) 0.025 −0.542 0.224 0.146 0.033 1.000 0.158
RVOL −0.067 −0.077 0.232 0.461 0.337 0.158 1.000

Note: Unconditional correlations of the natural logarithm of realised oil price volatility
(RVOL), the first difference of the natural logarithm of world crude oil production (ΔL
(PRO)), the global economic activity index (GEA), the first difference of the natural loga-
rithm of global crude oil inventories (ΔL(INV)), the natural logarithm of volatility VIX
index (VIX), the natural logarithm of TED spread (TED) and the first difference of the nat-
ural logarithm of trade-weighted exchange rate index (ΔL(TWE)). The sample period runs
from January 1990 to May 2019.

      Fig. 2. Median impulse response of realised oil price volatility to a global oil supply shock.
GIRF t K, ιi,t , I t−1 ¼ E ytþK jιi,t , I t−1 −E ytþK jI t−1 ð7Þ Note: This Figure exhibits the evolution of the time-varying median impulse response of
realised oil price volatility to a global oil supply shock approximated by the world crude
−1 −1 1
oil production growth rate. The right front axis represents the time (years), the left front
Ψgi,t ðK Þ ¼ Σii,t2 ΛK,t Σt ei,t Σii,t2 ιi,t , ιi,t ¼ Σ2ii,t ð8Þ
axis denotes the impulse response function monthly horizon and the vertical axis
depicts the change. The vertical colour bar to the right side of the plot displays the
−1 colour scale and specifies the mapping of data values. The sample period runs from
Ψgi,t ðK Þ ¼ Σii,t2 ΛK,t Σt ei,t ð9Þ January 1990 to May 2019.

where K represents the forecast horizon, ei,t the selection vector with
one on the ith position and zero otherwise.
1990–2019 the initial response from the latter appears to be positive
4. Empirical results (i.e., on the plot, we note that, during the first months of the GIRF purple
colour progressively turns into green and in some cases into yellow). Oil
This Section presents the empirical findings associated with the im- price volatility is expected to respond positively to a positive global oil
pact of global potential determinants on realised oil price volatility. It supply shock. Notably, a positive global oil supply shock signifies an in-
should be noted that, given the time-varying character of the analysis, creasing world oil production related to technological advances to ex-
we illustrate generalised impulse response functions on 3D-plots tract oil or to OPEC's decision to increase oil quotas. This reduces the
(i.e., GIRFs vary over time). Furthermore, given that this is not a constant oil price and adds to higher levels of oil price volatility. What is more,
variance – constant parameter model, no confidence bands have been this positive response is rather more pronounced around years 1990
included on the graphs. Figs. 2–7 report the results of the TVP-VAR and 1999 (i.e., when GIRFs are marked by yellow).
model and show the median impulse response functions for horizons Our plot shows that a significant rise in realised oil price volatility is
up to 60 months at each point in time. caused during the period of 1990–1991 which is associated with Iraq's
invasion of Kuwait and the collapse of the former Soviet Union, one of
4.1. Oil supply the three largest oil producers in the world in 1991 as well as during
1998–1999 which coincides with oil production cuts by OPEC. Regard-
Starting with the impact of a positive shock in oil production on ing the period of 1990–1991, oil price fluctuations are attributed to
realised oil price volatility, it is evident in Fig. 2 that in between years flow supply shocks together with speculative demand shocks that

6
I. Chatziantoniou, M. Filippidis, G. Filis et al. Energy Economics 95 (2021) 105092

occurred at the same time as documented by Kilian and Murphy (2014), that their proxies for aggregate demand do not significantly affect oil
whereas Hamilton (2009) suggests that flow supply shocks itself con- price volatility.
tributed to oil price fluctuations. Turning to the higher oil price volatility
of 1998–1999, Kilian and Murphy (2014) show that this was not only
4.3. Global oil inventory
caused by OPEC's production cuts decisions (supply-side shock), but
also by higher flow demand (aggregate demand shock) and higher
As far as the global oil inventory is concerned, the initial impact of
speculative activity (speculative shocks). The discussion on the effects
this shock on realised oil price volatility is positive throughout the pe-
of the latter two shocks on oil price volatility is discussed in sections
riod of analysis. As we illustrate in Fig. 4, this persistent positive re-
4.2 and 4.3.
sponse is rather more pronounced between 1998 and 2005 (i.e., when
GIRFs assume a yellow colour). Oil price volatility is expected to respond
positively to a positive global inventory demand shock. Particularly, a
4.2. Global economic activity positive global inventory shock which originates from increasing oil
supply relative to oil demand, acts as a signal to market participants
Turning to the impact of a positive shock in global economic activity that this rising inventory levels could drive the oil price at lower levels
on realised oil price volatility, as we can see in Fig. 3, between years which in turn increases oil price volatility. In addition to this, a positive
1990–2019 the initial response from the latter appears to be negative global inventory shock could represent increasing speculative demand
(i.e., on the plot, we note that, during the first months of the GIRF due to changes in the expectations for the future availability of oil and
green and in some cases yellow colour progressively turns into purple). thus changes in the future oil price movements. This is associated with
Oil price volatility is expected to respond negatively to a positive global the sudden accumulation of inventories in the very short run which
aggregate demand shock. In particular, a positive global aggregate de- may cause a reduction in the availability of barrels of oil for current
mand shock suggests a higher global demand for industrial commodi- use in the market and consequently increasing levels of oil price volatil-
ties including oil. This reflects positive developments in global ity. With reference to the expectations for the future availability of oil,
macroeconomic activity despite the oil price increase and consequently these are related to changes in the accumulation of oil inventories and
reduces the levels of oil price volatility. Furthermore, this impact ap- as such they are captured by the speculative demand shock (or global
pears to be more pronounced prior to 2008 and after 2010 (i.e., when inventory shock), similarly to Kilian and Murphy (2014). In addition, ac-
GIRFs progressively turn into purple). cording to Alquist and Kilian (2010) speculative purchases may also be
Plausibly, the pattern prior to 2008 can be explained by the unex- precautionary in that they reflect increased uncertainty about future de-
pected economic growth of the main emerging Asian economies such mand or supply conditions.
as China and India. In this regard, Kilian and Hicks (2013) refer to the Following the Iranian oil crisis (1979–1980), and prior to mid-1990's
period between 2003 and mid 2008 that contributed to an intense the oil inventories level remained fairly high. However, the period from
surge in the price of oil. In particular, they not only report economic the mid-1990 through the end of 2000 is closely associated with the
growth surprises of the emerging economies but also lesser growth sur- sharp drop of oil inventories due to OPEC's production cuts with the ex-
prises that associated with OECD advanced economies during the afore- ception of a recovery between 1998 and 1999 due to OPEC's overpro-
mentioned period. Similarly, the period after 2010 represents the duction (Lynch 2002). In addition, the period from 2000 to 2005 is
recovery of global markets and hence a reduction in uncertainty regard- characterised by adjustments in oil production from OPEC's member
ing the macroeconomic activity. The increase in global demand was ac- countries in an attempt to prevent oil inventories to be accumulated
companied with an increase in oil demand and hence higher oil prices (Verleger Jr 2009). These episodes together with the weak global de-
were associated with lower oil price volatility. Overall, our findings are mand caused by the Asian financial crisis of 1997–98 and the
not in line with the findings by Robe and Wallen (2016) who conclude

Fig. 3. Median impulse response of realised oil price volatility to a global oil demand shock. Fig. 4. Median impulse response of realised oil price volatility to a global oil inventories
Note: This Figure exhibits the evolution of the time-varying median impulse response of shock. Note: This Figure exhibits the evolution of the time-varying median impulse
realised oil price volatility to a global oil demand shock approximated by the global response of realised oil price volatility to a global oil inventories shock approximated by
economic activity index. The right front axis represents the time (years), the left front the global crude oil inventories growth rate. The right front axis represents the time
axis denotes the impulse response function monthly horizon and the vertical axis (years), the left front axis denotes the impulse response function monthly horizon and
depicts the change. The vertical colour bar to the right side of the plot displays the the vertical axis depicts the change. The vertical colour bar to the right side of the plot
colour scale and specifies the mapping of data values. The sample period runs from displays the colour scale and specifies the mapping of data values. The sample period
January 1990 to May 2019. runs from January 1990 to May 2019.

7
I. Chatziantoniou, M. Filippidis, G. Filis et al. Energy Economics 95 (2021) 105092

uncertainty triggered by the terrorist attack crisis of September 2001, (2016) who signify the importance of the financial market uncertainty
resulted in a sharp decline in oil prices and further to higher levels of to drive oil price volatility.
oil price volatility. Such findings confirm the asymmetric volatility phe-
nomenon (Bekaert and Wu 2000; Wu 2001). Overall, this period 4.5. Interest rates
roughly matches the period from 1998 to 2005 in which our findings in-
dicate higher levels of realised oil price volatility. The GIRFs that correspond to a positive shock in the TED spread ap-
pear to follow those of a shock in the VIX. To put differently, GIRFs ap-
parently assume very large values during well-defined crisis periods,
4.4. Volatility in financial markets particularly during the crisis in the early-1990s and the GFC 2007.
GIRFs in connection with the TED spread are illustrated on Fig. 6. The
The initial impact of a shock in the volatility (VIX) index on realised only difference between the VIX GIRFs and the TED spread GIRFs is
oil price volatility as observed in Fig. 5, is rather positive throughout the that realised oil price volatility seems to be relatively more (less) re-
period of analysis; however, the effect is clearly more pronounced dur- sponsive to VIX (TED spread) shock during the two crisis periods. Oil
ing periods of recession that fuel global economic uncertainty. Oil price price volatility is expected to respond positively to a positive global in-
volatility is expected to respond positively to a positive global financial terest rate shock. Specifically, a positive global interest rate shock
market volatility shock. Precisely, a positive global financial market vol- which is a money market shock causes the cost of borrowing to increase
atility shock represents rising uncertainty about future stock price fluc- and discourages investment. This development generates uncertainty in
tuations and contributes to negative returns in global stock markets global markets and causes a reduction in aggregate demand. In turn, oil
including oil returns. Thus, a higher level of financial uncertainty is in- demand and consequently oil price are expected to fall which implies
dicative of worsening economic conditions and associated with higher increases in oil price volatility.
oil price volatility. We can clearly discern the positive response from We notice that realised oil price volatility exhibits two large spikes
the realised oil price volatility during the recessions in the early-1990s when financial health approaches higher levels of stress. Since TED
and the 2000s, during the months of the GFC 2007, as well as more re- spread is considered as a principal indicator of financial health, we ex-
cently, during the intervals that include the European sovereign debt pect that when the TED spread is widened (narrowed) it is a signal of
crisis (since 2009) and the oil price collapse of 2014. Apparently the pe- worsening (improving) financial stress and consequently leads to a
riod that marks the largest positive response from the realised oil price higher (lower) possibility of a credit crisis. In this regard, the observed
volatility is the GFC 2007. peaks in realised oil price volatility are associated with jumps in TED
Indeed, we observe the realised oil price volatility to show peaks spread in two periods of financial crisis. Indeed, we refer to the early-
when financial market volatility exhibits large spikes mainly during 1990 recession related to Iraq's invasion of Kuwait which was following
times of economic and financial turbulence. For instance, the Persian the wave of late-1987 (Black Monday) stock market crash as well as the
Gulf War (1990–1991), the early-2000 US recession, the GFC 2007, the persistent wave of GFC 2007.
European sovereign debt crisis (since 2009) and the oil price crash
(2014). Linking these global episodes of financial uncertainty with the 4.6. Exchange rates
oil market, we note that all of them are associated with a collapse or a
sudden significant fall of the global oil prices. The only exception was Finally, with regard to the TWE, the initial response from realised oil
the Persian Gulf War during which oil prices climbed speedily and re- price volatility to a shock in the TWE appears to be positive throughout
lated to geopolitical turmoil. In turn, such developments drive oil price the period of analysis, as evident in Fig. 7 (i.e., GIRFs show a purple col-
volatility to substantial spikes. Specifically, the finding related to the sig- our which progressively turns into green and in some cases into
nificance of VIX index has also been reported by Robe and Wallen

Fig. 5. Median impulse response of realised oil price volatility to a global financial market Fig. 6. Median impulse response of realised oil price volatility to a global financial
uncertainty shock. Note: This Figure exhibits the evolution of the time-varying median interbank stress shock. Note: This Figure exhibits the evolution of the time-varying
impulse response of realised oil price volatility to a global financial market uncertainty median impulse response of realised oil price volatility to a global financial interbank
shock approximated by the volatility VIX index. The right front axis represents the time stress shock approximated by the TED spread. The right front axis represents the time
(years), the left front axis denotes the impulse response function monthly horizon and (years), the left front axis denotes the impulse response function monthly horizon and
the vertical axis depicts the change. The vertical colour bar to the right side of the plot the vertical axis depicts the change. The vertical colour bar to the right side of the plot
displays the colour scale and specifies the mapping of data values. The sample period displays the colour scale and specifies the mapping of data values. The sample period
runs from January 1990 to May 2019. runs from January 1990 to May 2019.

8
I. Chatziantoniou, M. Filippidis, G. Filis et al. Energy Economics 95 (2021) 105092

during the sample period. According to He et al. (2010) a 1% increase

in the US dollar index is associated with a 0.70% decrease in crude oil
prices, whereas Davig et al. (2015) find that a 1% appreciation of the
US dollar exerts a fall in oil prices by 2.3%. Evaluating the year 2014,
Davig et al. (2015) document the considerable appreciation of the US
dollar against other currencies caused a decline in oil demand for non-
US oil consumers and thus a reduction in oil price which triggered
higher levels of oil price volatility.

4.7. Summary of findings

Overall, we document that our set of potential determinants (at the

global level) appears to influence the realised oil price volatility and
therefore our study adds to the examination of oil price volatility
behaviour.9 We argue that fundamental and financial (non-fundamen-
tal) shocks are drivers of realised oil price volatility. It should be men-
tioned though, that if we consider the magnitude of these responses,
we document that there are certain differences among the GIRFs re-
sponses of oil price volatility to positive global shocks. Indeed, financial
shocks appear to exert a stronger impact on oil price volatility. We point
out that a common feature of our findings is associated with the impact
response of our potential global determinants which appears to be the
expected. Specifically, the initial impact takes approximately 10 months
Fig. 7. Median impulse response of realised oil price volatility to a global financial
ahead to progressively decay and then to level off. We could say that
currencies trend shock. Note: This Figure exhibits the evolution of the time-varying
median impulse response of realised oil price volatility to a global financial currencies short run effects can be associated with oil market fundamentals that
trend shock approximated by the trade-weighted exchange growth rate. The right front appear to exhibit such effects due to supply and demand imbalances.
axis represents the time (years), the left front axis denotes the impulse response In addition, financial indicators, incorporate information content
function monthly horizon and the vertical axis depicts the change. The vertical colour which drives the price of oil away from its fundamentals and conse-
bar to the right side of the plot displays the colour scale and specifies the mapping of
quently generates similar effects. However, the oil market has the
data values. The sample period runs from January 1990 to May 2019.
mechanism to adjust in new information and absorb the temporary im-
balances. In turn, this process reduces the uncertainty in oil price and
drives the volatility at very low levels in the long run.
yellow). Oil price volatility is expected to respond positively to a posi- On the whole, our findings support the evidence provided by Van
tive global exchange rate shock. Principally, a positive global exchange Robays (2016) who demonstrates that recessions and financial crises
rate shock is associated with an appreciation in the value of the US dol- trigger higher oil price uncertainty. On general principles, we suggest
lar which indicates a rise in the price of oil. However, global demand for that realised oil price volatility movements are attributed to changes
oil with the exception of the US market will be reduced and conse- in oil market fundamentals of oil supply and oil demand (including oil
quently causes oil prices to fall. This weakness in global economy results inventories) and also we suggest the importance of financial shocks
in higher oil price volatility. As has already been underscored in the that transmit information which has contributed to significant variation
analysis regarding the impact of a shock in either VIX or the TED spread, in oil price. We are in line with Kilian (2010) who discusses the origin
it is rather evident that in the case of a shock in exchange rates, both the and impact of oil price volatility and argues that expectations of
recessionary period of the early-1990s and the GFC 2007 also result in forward-looking traders, flow supply of oil and flow demand for oil
positive GIRFs of very large magnitude. are reflected in oil price changes. Our findings support the evidence pro-
On the basis of the above findings, changes in the trade-weighted ex- vided by Caldara et al. (2019) who emphasise the importance of oil sup-
change rate index appear to influence the realised oil price volatility in a ply shocks and global demand shocks in driving oil price fluctuations.
higher degree through the early-1990s and the GFC 2007 and in a lower Similarly, our suggestions appear to be in line with Beidas-Strom and
degree through 1995–2000 and 2010–2015. Regarding the early-1990s Pescatori (2014) who underline the short run impact of speculation
and specifically the Persian Gulf War (1990–1991), the general consen- on oil price fluctuations. By contrast, our findings do not appear to be
sus implies that wars trigger significant depreciation in the value of the in line with Van Robays (2016) who documents that speculation does
US dollar. Furthermore, the trade-weighted exchange rate index ex- not contribute to changes in oil price volatility. In particular, we show
hibits a persistent increasing trend and peaks during periods of financial that speculative shocks, as depicted by the global oil inventory shocks,
turmoil. For example, during 1995–2000, we focus on the Asian finan- trigger a material positive response of the oil price volatility, which is
cial crisis (devaluation in Southeast Asian currencies), during GFC relatively higher in amplitude compared to the responses of the latter
2007, we highlight the financial markets crash (capital repatriation to supply and aggregate demand shocks. As such, the accumulation of
from foreign markets to the US market which led to convert foreign cur- oil inventories arising from forward-looking market participants, in an-
rencies into US dollars) and during 2010–2015, we indicate the ticipation of rising oil prices in the future, tend to destabilise the market
European sovereign debt crisis (purchase of US dollars by investors). now and hence increase its uncertainty. To this end, Dvir and Rogoff
Clearly, realised oil price volatility appears to be influenced by periods (2009) also document that oil price volatility rises when the increase
of financial turmoil and unexpected wars that generate different trends in oil storage is subject to speculative demand.
in different currencies.
9
Following the previous findings for VIX and the TED spread, we no- We note that our findings regarding the impact from each potential global determi-
nant on conditional oil price volatility do not provide heterogeneous responses. Therefore,
tice that the impact magnitude of the TWE on realised oil price volatility we point out that our robustness check findings are qualitatively similar to those in the
is significantly larger and more persistent. This signifies that the TWE main analysis related to realised oil price volatility. For brevity we do not show the results
generates a relatively more sustainable impact on oil price volatility here, but they are available upon request.

9
I. Chatziantoniou, M. Filippidis, G. Filis et al. Energy Economics 95 (2021) 105092

Finally, we argue that our findings could be associated with recent Furthermore, turning to a more detailed analysis of the importance
evidence in connection with the financialisation of the oil market (see, of our findings to investors, traders and portfolio managers who are in-
Büyükşahin and Robe 2014; Fattouh et al. 2013; Tang and Xiong 2012; terested in the oil market, we should highlight the fact that their invest-
Irwin and Sanders 2011). Oil prices are not only determined by oil mar- ment decisions should be dependent on the uncertainty that surrounds
ket imbalances of supply and demand but also driven by increasing the financial markets. Hence, those investors who are interested in oil
financialisation. Our findings provide support to the evidence provided market returns should intensify their hedging strategies or exit the oil
by Fratzscher et al. (2014) who report that oil responds instantaneously market when there is increased uncertainty in the financial world. Ob-
to news incorporated in other asset prices such as stock returns, interest viously, higher uncertainty in the latter is anticipated to lead to in-
rates and exchange rates, which, in turn, implies that oil behaves like creased oil price volatility and thus lower oil prices. On the other
any other financial asset. According to our findings, global financial mar- hand, investors who are interested in trading oil market volatility,
ket uncertainty, global financial interbank stress and global financial they should increase their long positions in the latter (via long straddles
trends in currencies are all expected to generate global economic uncer- using options for example) when the uncertainty in the financial mar-
tainty. In turn, economic activity and consequently oil demand will be kets is anticipated to increase.
reduced and oil price volatility will be increased. Finally, a promising area for future study could model additional var-
iables. Indeed, the examination of whether other key oil-specific and
global business cycle variables (oil futures spreads, producer price
5. Conclusion index for oil, gasoline price spreads, capacity utilisation rate, Baltic Dry
index), financial variables (aggregate stock prices, firm level oil stock
In this study we attempt to investigate whether unanticipated prices) and commodity variables (gold, natural gas) could provide ex-
changes in global crude oil-market specific fundamental factors and fi- planatory value to capture movements in oil price volatility. Further-
nancial indicators could be regarded as uncertainty transmission to more, future research may examine the leading role of oil price
the oil market and cause oil price volatility. We employ a time-varying volatility in influencing macroeconomic indicators, given the fact that
parameter vector autoregression model. We use monthly data over volatile oil prices lead to rising uncertainty in the economic activity.
the period from January 1990 to May 2019.
Our main findings can be summarised as follows. First, crude oil- Authors' Contributions
market specific fundamental factors and financial indicators (both at
the global level) appear to influence the evolution of realised oil price All authors of the study, have all contributed to the conceptualisa-
volatility. Second, the impact varies over time and it is evident in the tion, development, analysis and writing of the paper.
short run horizon. Third, the effect is mainly attributed to severe epi-
sodes of economic recessions and financial crises. Fourth, the greatest
influence is attributed to the financial indicators and thus we lend sup- Acknowledgment
port to the financialisation of the oil markets.
These findings could suggest important implications for policy The authors would like to thank the editor (R.S.J. Tol) and two anon-
makers and investors who are interested in receiving information re- ymous referees for their constructive comments on a previous version
lated to the extent to which oil price volatility has influenced by global of the paper. David Gabauer would like to further acknowledge that this
determinants. Increasing (decreasing) oil price volatility could be indic- research has been partly funded by BMK, BMDW and the Province of
ative of higher (lower) uncertainty and consequently generates an un- Upper Austria in the frame of the COMET programme managed by
stable (stable) financial and macroeconomic environment. For FFG. The usual disclaimer applies.
investors, increasing oil price volatility implies a reduction in planning
and investment decisions as well as less effective fiscal and monetary Appendix A. Supplementary data
tools for policy makers. This situation could lead to reallocation of re-
sources which in turn affects negatively market productivity and further Supplementary data to this article can be found online at https://doi.
economic growth. It should be mentioned that periods of persistently org/10.1016/j.eneco.2020.105092.
increasing oil price volatility could be possibly influence the use of re-
newable energy sources by policy makers which could lead to reduced References
levels of oil price volatility.
Having documented that there is an increased financialisation of oil Adams, Z., Glück, T., 2015. Financialization in commodity markets: a passing trend or the
new normal? J. Bank. Financ. 60, 93–111.
market and speculative activity in the market, our findings have impor- Adams, Z., Collot, S., Kartsakli, M., 2020. Have Commodities Become a Financial Asset? Ev-
tant policy implications. Before presenting these implications it is im- idence from Ten Years of Financialization. Energy Economics, p. 104769.
portant to mention that speculation could be beneficial for Alquist, R., Kilian, L., 2010. What do we learn from the price of crude oil futures? J. Appl.
commercial consumers and commercial producers given that it pro- Econ. 25 (4), 539–573.
Andersen, T.G., Bollerslev, T., 1998. Answering the skeptics: yes, standard volatility
vides liquidity to the market (see, Fattouh 2010). However, when spec- models do provide accurate forecasts. International Economic Review 885–905.
ulators attempt to dominate the market by holding an enormous share Andersen, T.G., Bollerslev, T., Diebold, F.X., Labys, P., 2003. Modeling and forecasting real-
of the futures market (excessive speculation), this results to the ized volatility. Econometrica 71 (2), 579–625.
destabilisation of the price discovery and consequently negatively af- Basak, S., Pavlova, A., 2016. A model of financialization of commodities. J. Financ. 71 (4),
1511–1556.
fects commercial users regarding their hedge capacity. Therefore, exces-
Basher, S.A., Haug, A.A., Sadorsky, P., 2012. Oil prices, exchange rates and emerging stock
sive speculation could be considered a rather important source of markets. Energy Econ. 34 (1), 227–240.
growing oil price volatility. In the presence of excessive speculation, pol- Baumeister, C., Kilian, L., 2012. Real-time forecasts of the real price of oil. J. Busi. Econ.
icy makers need to impose regulatory tools to restrict such activities in Statist. 30 (2), 326–336.
Baur, D.G., Dimpfl, T., 2018. The asymmetric return-volatility relationship of commodity
commodity futures markets and hence to control their market share,
prices. Energy Econ. 76, 378–387.
as also suggested by Fattouh (2010). Even more, regulators should in- Beidas-Strom, S., Pescatori, A., 2014. Oil price volatility and the role of speculation. Num-
corporate the developments in the oil market in both their micro and ber 14–218. International Monetary Fund.
macroprudential policy decisions, since the fact that the oil market be- Bekaert, G., Wu, G., 2000. Asymmetric volatility and risk in equity markets. Rev. Financ.
Stud. 13 (1), 1–42.
haves as any other financial asset (due to its financialisation), it makes
Bicchetti, D., Maystre, N., 2013. The synchronized and long-lasting structural change on
it prone to bubbles, which could then contribute to both systemic and commodity markets: evidence from high frequency data. Algorithmic Finance 2
non-systemic crises. (3–4), 233–239.

10
I. Chatziantoniou, M. Filippidis, G. Filis et al. Energy Economics 95 (2021) 105092

Blecker, R.A., 2003. The Benefits of a Lower Dollar: How the High Dollar Has Hurt US Hansen, P.R., Lunde, A., 2006. Consistent ranking of volatility models. J. Econ. 131 (1–2),
Manufacturing Producers and why the Dollar Still Nees to Fall Further (Economic Pol- 97–121.
icy Institute). He, Y., Wang, S., Lai, K.K., 2010. Global economic activity and crude oil prices: a
Bollerslev, T., 1986. Generalized autoregressive conditional heteroskedasticity. J. Econ. 31 cointegration analysis. Energy Econ. 32 (4), 868–876.
(3), 307–327. Henriques, I., Sadorsky, P., 2011. The effect of oil price volatility on strategic investment.
Bouri, E., Gupta, R., Hosseini, S., Lau, C.K.M., 2018a. Does global fear predict fear in BRICS Energy Econ. 33 (1), 79–87.
stock markets? Evidence from a Bayesian graphical structural VAR model. Emerg. Herrera, A.M., Hu, L., Pastor, D., 2018. Forecasting crude oil price volatility. Int. J. Forecast.
Mark. Rev. 34, 124–142. 34 (4), 622–635.
Bouri, E., Lien, D., Roubaud, D., Hussain Shahzad, S.J., 2018b. Fear linkages between the US Irwin, S.H., Sanders, D.R., 2011. Index funds, financialization, and commodity futures mar-
and BRICS stock markets: a frequency-domain causality. Int. J. Econ. Bus. 25 (3), kets. Applied Economic Perspectives and Policy 33 (1), 1–31.
441–454. Kilian, L., 2009. Not all oil price shocks are alike: disentangling demand and supply shocks
Bouri, E., Shahzad, S.J.H., Raza, N., Roubaud, D., 2018c. Oil volatility and sovereign risk of in the crude oil market. Am. Econ. Rev. 99 (3), 1053–1069.
BRICS. Energy Econ. 70, 258–269. Kilian, L., 2010. Oil Price Volatility: Origins and Effects. Technical report. World Trade Or-
Büyükşahin, B., Robe, M.A., 2014. Speculators, commodities and cross-market linkages. ganization (WTO), Economic Research and Statistics Division.
J. Int. Money Financ. 42, 38–70. Kilian, L., Hicks, B., 2013. Did unexpectedly strong economic growth cause the oil price
Büyüksahin, B., Haigh, M.S., Robe, M.A., 2009. Commodities and equities: ever a “market shock of 2003–2008? J. Forecast. 32 (5), 385–394.
of one”? J. Altern. Invest. 12 (3), 76–95. Kilian, L., Murphy, D.P., 2014. The role of inventories and speculative trading in the global
Cai, Y., Chou, R.Y., Li, D., 2009. Explaining international stock correlations with CPI fluctu- market for crude oil. J. Appl. Econ. 29 (3), 454–478.
ations and market volatility. J. Bank. Financ. 33 (11), 2026–2035. Kilian, L., Park, C., 2009. The impact of oil price shocks on the US stock market. Int. Econ.
Caldara, D., Cavallo, M., Iacoviello, M., 2019. Oil price elasticities and oil price fluctuations. Rev. 50 (4), 1267–1287.
J. Monet. Econ. 103, 1–20. Koop, G., Pesaran, M.H., Potter, S.M., 1996. Impulse response analysis in nonlinear multi-
Chatziantoniou, I., Degiannakis, S., Filis, G., 2019. Futures-based forecasts: how useful are variate models. J. Econ. 74 (1), 119–147.
they for oil price volatility forecasting? Energy Econ. 81, 639–649. Lynch, M.C., 2002. Causes of oil price volatility. Journal of Energy and Development 28 (1),
Cheng, I.-H., Kirilenko, A., Xiong, W., 2015. Convective risk flows in commodity futures 107–141.
markets. Review of Finance 19 (5), 1733–1781. Morana, C., 2013. Oil price dynamics, macro-finance interactions and the role of financial
Cogley, T., Sargent, T.J., 2005. Drifts and volatilities: monetary policies and outcomes in speculation. J. Bank. Financ. 37 (1), 206–226.
the post WWII US. Rev. Econ. Dyn. 8 (2), 262–302. Nguyen, Q.N., Aboura, S., Chevallier, J., Zhang, L., Zhu, B., 2020. Local gaussian correlations
Davig, T., Melek, N.Ç., Nie, J., Smith, A.L., Tüzemen, D., 2015. Evaluating a year of oil price in financial and commodity markets. Eur. J. Oper. Res. 285 (1), 306–323. https://doi.
volatility. Federal Reserve Bank of Kansas City, Economic Review 100 (3), 5–30. org/10.1016/j.ejor.2020.01.023.
Degiannakis, S., Filis, G., 2017. Forecasting oil price realized volatility using information Pesaran, H.H., Shin, Y., 1998. Generalized impulse response analysis in linear multivariate
channels from other asset classes. J. Int. Money Financ. 76, 28–49. models. Econ. Lett. 58 (1), 17–29.
Del Negro, M., Primiceri, G.E., 2015. Time varying structural vector autoregressions and Phan, D.H.B., Sharma, S.S., Narayan, P.K., 2016. Intraday volatility interaction between the
monetary policy: a corrigendum. Review of Economic Studies 82 (4), 1342–1345. crude oil and equity markets. J. Int. Financ. Mark. Inst. Money 40, 1–13.
Diaz, E.M., Molero, J.C., de Gracia, F.P., 2016. Oil price volatility and stock returns in the G7 Pindyck, R.S., 2004. Volatility in natural gas and oil markets. Journal of Energy and Devel-
economies. Energy Econ. 54, 417–430. opment 30 (1), 1–19.
Dickey, D.A., Fuller, W.A., 1981. Likelihood ratio statistics for autoregressive time series Primiceri, G.E., 2005. Time varying structural vector autoregressions and monetary policy.
with a unit root. Econometrica, pages 1057–1072. Review of Economic Studies 72 (3), 821–852.
Dvir, E., Rogoff, K., 2009. Three epochs of oil. NBER Working Paper, (w14927). Prokopczuk, M., Symeonidis, L., Wese Simen, C., 2016. Do jumps matter for volatility fore-
Efimova, O., Serletis, A., 2014. Energy markets volatility modelling using GARCH. Energy casting? Evidence from energy markets. J. Futures Mark. 36 (8), 758–792.
Econ. 43, 264–273. Robe, M.A., Wallen, J., 2016. Fundamentals, derivatives market information and oil price
Elder, J., Serletis, A., 2010. Oil price uncertainty. J. Money Credit Bank. 42 (6), 1137–1159. volatility. J. Futur. Mark. 36 (4), 317–344.
Engle, R.F., 1982. Autoregressive conditional heteroscedasticity with estimates of the var- Sarwar, G., 2012. Is VIX an investor fear gauge in BRIC equity markets? J. Multinatl. Financ.
iance of United Kingdom inflation. Econometrica. pp. 987–1007. Manag. 22 (3), 55–65.
Fattouh, B., 2010. Oil Market Dynamics through the Lens of the 2002–2009 Price Cycle Silvennoinen, A., Thorp, S., 2013. Financialization, crisis and commodity correlation dy-
(Oxford Institute for Energy Studies). namics. J. Int. Financ. Mark. Inst. Money 24, 42–65.
Fattouh, B., 2012. The Financialization of Oil Markets: Potential Impacts and Evidence. Ox- Singleton, K.J., 2014. Investor flows and the 2008 boom/bust in oil prices. Manag. Sci. 60
ford Institute for Energy Studies, Oxford. (2), 300–318.
Fattouh, B., Kilian, L., Mahadeva, L., 2013. The role of speculation in oil markets: what have Tang, K., Xiong, W., 2012. Index investment and the financialization of commodities.
we learned so far? Energy Journal 34 (3). Financ. Anal. J. 68 (6), 54–74.
Ferderer, J.P., 1996. Oil price volatility and the macroeconomy. J. Macroecon. 18 (1), 1–26. Van Robays, I., 2016. Macroeconomic uncertainty and oil price volatility. Oxf. Bull. Econ.
Fratzscher, M., 2009. What explains global exchange rate movements during the financial Stat. 78 (5), 671–693.
crisis? J. Int. Money Financ. 28 (8), 1390–1407. Verleger Jr., P.K., 2009. Anatomy of the 10-Year Cycle in Crude Oil Prices. University of Cal-
Fratzscher, M., Schneider, D., Van Robays, I., 2014. Oil Prices, Exchange Rates and Asset gary, Calgary, Alberta, Canada.
Prices. Wang, Y., Liu, L., Wu, C., 2017. Forecasting the real prices of crude oil using forecast com-
Haase, M., Zimmermann, Y.S., Zimmermann, H., 2016. The impact of speculation on com- binations over time-varying parameter models. Energy Econ. 66, 337–348.
modity futures markets – a review of the findings of 100 empirical studies. Wu, G., 2001. The determinants of asymmetric volatility. Rev. Financ. Stud. 14 (3),
J. Commod. Mark. 3 (1), 1–15. 837–859.
Hamilton, J.D., 1989. A new approach to the economic analysis of nonstationary time se- Xekalaki, E., Degiannakis, S., 2010. ARCH Models for Financial Applications. John Wiley &
ries and the business cycle. Econometrica 357–384. Sons.
Hamilton, J.D., 2009. Causes and Consequences of the Oil Shock of 2007–08. Technical re- Zhang, Y.-J., Chevallier, J., Guesmi, K., 2017. “De-financialization”of commodities? Evi-
port, National Bureau of Economic Research. dence from stock, crude oil and natural gas markets. Energy Econ. 68, 228–239.
Hamilton, J.D., Wu, J.C., 2014. Risk premia in crude oil futures prices. J. Int. Money Financ.
42, 9–37.

11