Crude oil futures prices and foreign exchange markets

Abdur Rahman Forhadb, Mohammed Syedul Islamc and Joshua LawsondaDepartment of Finance and Economics, University of Tennessee at Chattanooga, Chattanooga, TN, USA; bDepartment of Humaniti

Full Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journalInformation?journalCode=raec20

Applied Economics

ISSN: (Print) (Online) Journal homepage: www.tandfonline.com/journals/raec20

Crude oil futures prices and foreign exchange

markets

Md Rafayet Alam, Md Abdur Rahman Forhad, Mohammed Syedul Islam & Joshua Lawson

To cite this article: Md Rafayet Alam, Md Abdur Rahman Forhad, Mohammed Syedul Islam &

Joshua Lawson (07 Dec 2023): Crude oil futures prices and foreign exchange markets, Applied Economics, DOI: 10.1080/00036846.2023.2288044

To link to this article: https://doi.org/10.1080/00036846.2023.2288044

Published online: 07 Dec 2023.

Submit your article to this journal

Article views: 106

View related articles

View Crossmark data

Citing articles: 1 View citing articles

Crude oil futures prices and foreign exchange markets

Md Rafayet Alam a , Md Abdur Rahman Forhad b , Mohammed Syedul Islam c and Joshua Lawson d

a Department of Finance and Economics, University of Tennessee at Chattanooga, Chattanooga, TN, USA; b Department of Humanities and Social Sciences, Dhaka University of Engineering and Technology, Gazipur, Bangladesh; c Division of Resource Economics & Management, West Virginia University, Morgantown, WV, USA; d Simon Business School, University of Rochester, Rochester, NY, USA

ABSTRACT

We apply a time-varying parameter VAR (TVP-VAR) extended joint connectedness approach, in

addition to the generalized connectedness approach, to understand the connectedness of crude

oil futures price and exchange rates of major oil-dependent countries We find time-varying nature

of pair-wise and total connectedness that are usually elevated during events such as COVID-19,

Brexit, European sovereign debt crisis and global financial crisis Both joint and generalized

connectedness approaches confirm that Japanese Yen and Russian Ruble are the leading net

receiver of the shocks, though the two approaches provide mixed results for some other

curren-cies Moreover, there is strong evidence of time-varying and bi-directional shock transmissions

between oil and foreign exchange markets We also show that oil price volatility and gold price

have predictive power on the connectedness Lastly, we analyse the policy and portfolio

implications.

KEYWORDS

Oil futures price; foreign exchange; dynamic connectedness; time-varying parameter VAR; hedge ratio

JEL CLASSIFICATION

Q40; F31

I Introduction

Connectedness measures the ‘systemic risk’ or the

risk of rapid spillover of crisis across firms,

indus-tries and markets and thus serves as an early

indi-cator of the severity of the contiguity of a crisis

Protection against risks becomes extremely difficult

in the presence of high connectedness among the

financial markets (Maggi, Torrente, and Uberti

2020) In fact, high connectedness is the main reason

the subprime crises of 2007–08 were so deep and

widespread (Andries and Galasan 2020) Therefore,

it is not surprising that understanding how markets

and their respective components transmit risks has

drawn special attention of researchers (Forhad, and

Alam 2022a, Alam, Forhad, and Sah 2022) In fact,

following the seminal works by Diebold and Yilmaz

(2009, 2012, 2014), a sizable proportion of the

lit-erature has been devoted to measuring these

spil-lover effects across financial and macroeconomic

variables through the analysis of connectedness

Exchange rates are important economic

indica-tors, and foreign exchange markets are among the

important markets for investors and policymakers

However, foreign exchange markets are also some

of the most volatile markets and are very

responsive to political, macroeconomic and finan-cial events On the other hand, crude oil is an essential commodity for the global economy and

is assumed to influence many macroeconomic and financial indicators For example, Reboredo (2012) shows that increased oil prices are responsible for economic recessions, trade deficits, high inflation, low values of stocks and bonds and high uncer-tainty for investment The market for crude oil is also extremely volatile For example, the price of WTI crude oil rose to $123 per barrel in

8 March 2022 from $ −36.98 in 20 April 2020,

a $160 increase in per barrel in less than 2 years Recent geopolitical and public health crises, stem-ming from the war and Covid-19, have also made the oil market extremely volatile and unpredict-able Such volatility and uncertainty of crude oil prices could have far-reaching effects on other financial and economic indicators including exchange rates In fact, the currencies of many oil- importing countries have already started to experi-ence sharp depreciation because of high oil prices

We analyse the total and directional connected-ness of crude oil and foreign exchange markets using two connectedness approaches One is TVP-

CONTACT Md Rafayet Alam mailtorafayet@gmail.com Department of Finance and Economics, University of Tennessee at Chattanooga, Chattanooga,

TN 37403, USA

https://doi.org/10.1080/00036846.2023.2288044

VAR extended joint connectedness approach as of

Balcilar et al (2021) and the other is the generalized

connectedness approach of Diebold and Yilmaz

(2012, 2014) (hereafter, DY (Diebold and Yilmaz

2012; Diebold and Yılmaz 2014)) The relationship

between commodity and foreign exchange markets

has a strong theoretical basis (Chen and Rogoff

2003) and changes in commodity prices should

have effect on the exchange rates, especially those

of commodity-dependent countries Oil is the

lar-gest-trading commodity in the world and oil prices

are believed to affect many economic indicators

including the balance of payments and exchange

rates of oil-dependent countries Foreign exchange

markets are connected within themselves and with

oil markets through various channels, such as

glo-bal economic conditions, disruptions in demand

and supply, monetary policy stance of major

economies and global inflation (Singh, Nishant,

and Kumar 2018) Business cycle convergence

and synchronization can also connect the financial

and foreign exchange markets (Wan and He 2021)

Other factors – such as international trade and

financial factors (Chow 2021), macroeconomic

announcements (Wen and Wang 2020) and policy

uncertainty (Huynh, Nasir, and Nguyen 2020) –

may also act as channels through which currencies

are connected within themselves and with oil

mar-ket Oil revenue received in US dollar (Petrodollar)

can also produce volatility in foreign exchange

pairs with US dollar (hereafter, USD) such as

EURUSD, USDJPY directly and between other

for-eign exchange pairs such as EURGBP and EURJPY

indirectly (Singh, Nishant, and Kumar 2018)

Common risk factors can also, at least partially,

explain the co-movement between oil and foreign

exchange markets (Forhad, and Alam 2022b)

However, the intensity and direction of causality

between the two markets depend on several factors

and may change over time For example, Singh

et al (2018) document a recent reversal of the

historical inverse relation between oil prices and

exchange rates

Theoretically, there are several channels through

which oil price shocks may affect the exchange

rates: the trade channel, the wealth effect channel

and the portfolio reallocation channel (Habib,

Butzer, and Stracca 2016; Singh, Nishant, and

Kumar 2018) In terms of the trade channel, an

increase in oil price generally depreciates (appreci-ates) the exchange rates of oil-importing (export-ing) countries by deteriorating (improv(export-ing) their trade balances (Backus and Crucini 2000) In the wealth effect approach, the increase in oil price causes wealth transfer from oil importers to oil exporters This wealth transfer then leads to a real depreciation (appreciation) of the exchange rates of oil-importing (exporting) countries, mainly in the short term through current account imbalances (Habib, Butzer, and Stracca 2016; Kilian 2009; Krugman 1983) Finally, in the portfolio realloca-tion channel, oil shocks affect the exchange rates, mainly in the mid and long term, depending on the nature of trade and preference for the USD by the country (Beckmann, Czudaj, and Arora 2020; Habib, Butzer, and Stracca 2016) In section II, we review the empirical studies on the relationship between oil and foreign exchange markets

Our article has several contributions to the related literature First, in our analysis, we utilize

a better methodology which has several benefits (as discussed below) and produces results that are dif-ferent (as discussed below and in section IV) from those using the DY (2012, 2014) approach on many occasions Second, we not only measure the con-nectedness but also test the predictive power of uncertainties and other variables on such connect-edness Third, our study also analyzes the portfolio implications by estimating hedge-ratio and hed-ging effectiveness Fourth, our analysis includes data from the period of COVID-19, which enables

us to shed light on and make a comparison with the findings from the unprecedented period of COVID-19 Fifth, our study utilizes futures prices,

as opposed to spot prices used in other papers, which are forward looking and capture more information

Following Balcilar et al (2021), we use a time- varying parameter vector auto-regression (TVP- VAR) extended joint connectedness approach This method by Balcilar et al (2021) combines the approaches by Antonakakis et al (2020) and Lastrapes and Wiesen (2021) Balcilar et al (2021) mention several benefits of this approach, espe-cially over the approach of DY (2012, 2014) First, the time variation in the VAR coefficients is more accurately captured in this approach Second, this approach is affected less by outliers and adjusts

better to parameter changes Third, the arbitrary

choice of rolling-windows is no longer needed in

this approach Fourth, without losing observations,

this method accurately calculates the generalized

forecast error variance decompositions Fifth, this

method can more suitably be applied to low-

frequency datasets Sixth, it allows a theoretically

derived normalization technique as opposed to the

standard normalization technique applied in the

DY (2012, 2014) In fact, we show that on many

occasions the results derived from the TVP-VAR

extended joint connectedness approach are

differ-ent from those produced by the DY (2012, 2014)

generalized connectedness approach For an

exam-ple, whereas the DY (2012, 2014) generalized

con-nectedness approach finds Euro as the primary

transmitter of shocks, the TVP-VAR extended

joint connectedness approach shows that Euro is

instead a shock receiver.1

It is worth mentioning that the literature on the

connectedness between oil and exchange rates is

inconclusive, with many opposite and contrasting

findings Unlike some of the existing studies that

find a much stronger role of the oil market on the

foreign exchange markets, our study finds evidence

of bi-directional and time-varying causality that

runs between the two markets In addition, we

show that gold price and oil price volatility have

the predictive power on the connectedness of oil

and foreign exchange markets, and that oil could be

a cheap hedge for currencies though the values of

hedging effectiveness are small

The rest of the article is organized as follows

Section II reviews the related literature, section III

discusses data, summary statistics and

methodol-ogy, section IV presents and analyzes the results,

and finally, section V concludes with implications

II Literature review

Several studies apply the connectedness

approach to empirically examine the

connected-ness among exchange rates or between oil prices

and exchange rates.2 However, the findings of

these studies vary significantly based on the

sample period, countries under consideration

and methodology Wen and Wang (2020) ana-lyse the volatility connectedness among several currencies and find that the USD and EUR are major volatility transmitters, while other curren-cies, including the JPY and the GBP, are net volatility receivers They also show that total volatility connectedness increases during periods

of crisis Singh et al (2018) explore the dynamic and directional network connectedness between implied volatility measures of crude oil and the exchange rate of nine major currency pairs for

a sample period from May 2007 to December 2016 They find that volatility of crude oil affects the volatility of currencies in the first part of the sample, while the volatility

of currency affects the volatility of crude oil in the latter part of their sample Malik and Umar (2019) use the underlying shocks of the oil price changes and show that the connectedness of oil price shocks and exchange rates has significantly increased after the global financial crisis They also show that, though there is a significant volatility connectedness among the exchange rates, oil price shocks do not explain the varia-tion in exchange rate volatility Huynh et al (2020), analysing USD exchange rates of nine currencies, document asymmetric spillovers and connectedness among the exchange rates in the presence of trade policy uncertainty Wan and

He (2021) analysing the dynamic connectedness

of the G7 currencies show that the USD is always a transmitter, the British pound and Canadian dollar are always recipients, while the Euro and Japanese yen switch roles between transmitter and receiver of the shocks during the sample period (Wan and He 2021) Gomez- Gonzalez et al (2020) examine the connected-ness and causality between oil spot prices and exchange rates of a number of countries They find that the spillovers and causality are mainly from exchange rates to oil prices for oil- producing countries and from oil price to exchange rates for oil importers However, they find that oil prices are net receivers of shocks during most of the sample period Using daily data of BRICS countries, Tiwari, et al (2019) 1

We discuss the differences in the results between the two approaches in-detail in Section IV.

2 There is a sizable literature on the relationship between oil and exchange rates using other econometric methods However, for the sake of brevity we concentrate on the literature that applies connectedness or similar approaches to investigate the relationship between oil and exchange rates.

find significant negative dependence between oil

prices and the currencies of Brazil, India and

South Africa in the long-run Wang et al

(2022) show that bitcoin is not a good safe-

haven for crude oil, but gold is a good safe-

haven for crude oil across various time horizons,

and both before and after the outbreak of the

COVID-19 pandemic Salisu et al (2022) show

that the association between oil tail risk and

USD tail risk is positive when USD/CAD and

USD/GBP are considered and negative when

USD/JPY is considered Dai et al (2020) provide

a systemic analysis of dependence and risk

con-tagion among oil, gold and the US dollar foreign

exchange markets and shows that the three

assets are better integrated in the medium-run

than the short-run

Fueki et al (2021) offer an SVAR model to

ana-lyse the factors of oil price dynamics and show that

future demand and supply shocks are as important

as the realized demand and supply shocks in

explaining historical oil price fluctuations They

also show that different types of oil shocks have

different effects on global output and suggest that

it is necessary to realize the reasons for oil price

changes in assessing their macroeconomic effects

Dai et al (2023) applying GARCH-S analysis on

cryptocurrencies and equity market indices for

advanced and emerging economies show that the

crashes originating in the cryptocurrency market

may cause equity market crashes For example, the

aftermath of cryptocurrency crashes, the occurrence

of co-crashes between cryptocurrency and equity

markets is in the 80% of the cases identified They

further show that cryptocurrency uncertainty

pos-sesses a much greater power for predicting co-

crashes than economic policy uncertainty They

also find that macroeconomic risk is a good

predic-tor of co-crash during short-run, while infectious

disease volatility is a good predictor of co-crash in

the long-run Khalifa et al (2015) show that

uncer-tainty measures have an impact on volatility of oil

and currencies but not on natural gas Sun et al

(2022) examine whether the relationships between

China’s exchange rate, domestic crude oil price and

the international crude oil price have a switch in the

period before and after China’s crude oil futures

launched by Shanghai International Energy

Exchange (INE) and show that since the launch of INE crude oil futures in the new regime, the fluctua-tions in the USD against the RMB (USD/CNY) exchange rate has had a significant positive effect

on China’s crude oil prices

Shang and Hamori (2021) report the significant roles of WTI crude oil in transmitting both return and volatility shocks to foreign exchange markets Nekhili et al (2021) examine the time-frequency return and volatility spillovers between commodity futures prices (copper, crude oil, gold, wheat) and currencies (British pound, Canadian dollar, Euro, Japanese yen, Swedish krona and Swiss franc) They find that the spillover between the commod-ity futures price and currency is time-varying, asymmetric, crisis-sensitive and currency- and commodity-specific Adekoya and Oliyide (2020) also find that the oil and US currency markets and other financial markets are strongly connected dur-ing the COVID-19 pandemic On the other hand, Asadi et al (2022) examine volatility connected-ness across crude oil, natural gas, coal, stock and currency markets in the US and China and suggest that total connectedness among energy, stock and currency markets is not high

Analysing high-frequency data, Ahmad et al (2020) argue that oil-jumps negatively impact the volatility of Chinese exchange rates Using 5-min-ute interval data, Alam et al (2019) show that the causal effects from currency markets to the crude oil market are stronger than that from crude oil market to currency market Fasanya et al (2021) also examine the effect of the US economic policy uncertainty on the connectedness across the oil and the most globally traded currency pairs They find

a strong connection between crude oil and cur-rency markets, with oil being the net receiver of shocks Albulescu and Ajmi (2021) examine the time- and frequency-domain spillover between oil and commodity currencies and argue that the financial markets are strongly connected, with oil being a net transmitter of shocks across all the frequency cycles However, as we have already mentioned in Section I, our study offers several contributions to the existing literature We expect this study to improve our understanding of this important but unsettled issue and the implications for various stakeholders

III Data, summary statistics and methodology

We use the USD exchange rates of five major oil-

exporting countries: Brazil, Canada, Mexico,

Norway and Russia, and six major oil-importing

countries/region: India, Japan, South Korea, China,

the United Kingdom and the European Union.3

The nominal exchange rates are retrieved from

Federal Reserve Economic Data (FRED), St Louis

Fed.4 The monthly exchange rates are the average

of the daily exchange rates which are measured in

terms of the domestic currency of each country per

USD.5 In addition, we employ the S&P GSCI Crude

oil index, representing the price of the ‘First Nearby

Contract Expirations’ on each business day, which

is retrieved from Datastream (GSCLSPT) To check

the predictive power on the connectedness, we

have also used gold price (generic 1st future),

Citigroup’s Citi Macro Risk Index (a measure of

global risk aversion), economic policy uncertainty

and oil price volatility Gold price (GC1) and

Macro Risk Index (MRICITI) are retrieved from

Bloomberg terminal Uncertainty series is from

Baker et al (2016),6 and oil price volatility (ovx)

is from Chicago Board Options Exchange (CBOE)

Following the extant literature, and to avoid the

unit-root problem, the exchange rates and crude oil

price are used in the log-return form The sample

period, 1 March 2000 to 1 March 2022, is dictated

by the availability of data

Table 1 shows the summary statistics of oil price

and the exchange rates The mean returns of the

exchange rates of all the countries except Canada,

China and the European Union are positive

Usually, the variances of crude oil futures contracts

are higher than those of exchange rates We use

D’Agostino (1970) for the skewness test Table 1

also shows that the returns series for crude oil,

Canada, the European Union, Japan and South

Korea are negatively and significantly skewed In

contrast, the return series for China, Mexico,

Norway and Russia are positively and significantly

skewed Negative skewness implies that means are

smaller than the medians, while positive skewness implies that medians are smaller than means For the kurtosis test, we follow Anscombe and Glynn (1983) approach Table 1 shows that all series are signifi-cantly leptokurtic implying fat tail distribution We use Jarque and Bera (1980) test to check normality of the distributions The skewness and kurtosis tests and the Jarque and Bera (1980) normality test con-firm that all the return series are significantly non- normally distributed Moreover, the Fisher and Gallagher (2012) weighted portmanteau test con-firms that returns and squared returns are autocor-related (Table 1) All these characteristics of the series justify modelling the interconnectedness using a TVP-VAR approach with a time-varying variance-covariance structure Finally, Stock et al (1996) unit root test (ERS) confirms that all the return series are stationary at the conventional levels

of significance

We estimate the generalized connectedness fol-lowing the Diebold and Yilmaz (2012, 2014) approach.7 We estimate the TVP-VAR extended joint connectedness following Balcilar et al (2021) Balcilar et al (2021) combine Antonakakis

et al (2020) and Lastrapes and Wiesen (2021) methods and provide a TVP-VAR extended joint connectedness framework Antonakakis et al (2020) offer a TVP-VAR connectedness method that more accurately measures the changes in the parameters and overcomes some of the shortcom-ings of the conventional method by Diebold and Yilmaz (2012, 2014) Lastrapes and Wiesen (2021) formulate a joint spillover index that normalizes the model based on the goodness-of-fit matrix, which is more effectual than the normalization technique used in the conventional method, and allows an easier interpretation of the connected-ness and spillover process

A TVP-VAR model can be transformed into

TVP-VMA model, and from there the H-step

fore-cast error can be written as8

3

We do not consider Middle Eastern countries as they maintain managed exchange rate regimes.

4 https://fred.stlouisfed.org/categories/95.

5

To perform the hedge-ratio analysis in section IV, we measure the exchange rates as USD per foreign currency as it makes the explanation more intuitive.

6 https://www.policyuncertainty.com.

7

For details of the methodology we refer readers to the original papers of Diebold and Yilmaz (2012, 2014).

8 For details of the methodology we refer readers to the original paper of Balcilar et al (2021).The estimation code was retrieved from the website of Dr David Gabauer https://github.com/GabauerDavid/ConnectednessApproach.:

Table

ζ tð Þ ¼H y tþH E y tþH jy t ; y t 1;

�

H 1

h¼0

A h;t ε tþH h (1)

where y t is the vector of variables The covariance

matrix of forecast error is represented by

E ζ tð ÞH ζ0tð ÞH

¼A h;t

X

t

The corresponding joint connectedness model can

be expressed as:

S jnt; from i ;t

¼E ζ2i;tð ÞH

E ζ�i;tð ÞH Eζ i;tð ÞH�j2 " �i; t þ 1; ; 2" �i; t þ H

i2

E ζ2i;tð ÞH

¼

PH 1

h o e0i A kt

P

t M i M0tPt M t0� 1M0tPt A0kt e t

PH 1

k 0 e0i A kt

P

t A0kt e i

(3)

where matrix M i is equal to the identity matrix

without the ith column For all variables except

variable i; ε"�i; t þ 1 indicates the K-1

dimen-sional vector of shocks at time t þ 1 In this case,

unlike the original connectedness approach, no

normalization is required to confirm the spillovers

to be within zero and unity

The joint total connectedness index is then

expressed by:

K

XK

i¼1

S jnt; from i �;t (4)

Then, following Balcilar et al (2021), the total

directional connectedness and the net total

direc-tional connectedness metrics are obtained from the

following equations:

S jnt;to i�!;t¼ X

K

j¼1;i�j

S jnt;net j;t ¼S jnt;to i!�;t S jnt; from i �;t (6)

We also regress the dynamic total connectedness

index on uncertainties and a set of other regressors

to check their predictability on the connectedness

(Equation 7) For uncertainties, we use economic

policy uncertainty and oil price volatility

Following Dai et al (2023), we also include gold

price and Citigroup’s Citi Macro Risk Index (MRI),

a measure of risk aversion often used when com-paring global financial markets, in the regression

Connect t¼β0þβ1Gold tþβ2MRI tþβ3Uncertainty t

þβ4Oilvolatility tþe t ;

(7)

Further, we calculate optimal hedge-ratio employ-ing VAR-DCC-MGARCH and rollemploy-ing window methods The dynamic hedge ratio between

a long position in the exchange rates for

a country and a short position in the oil prices is:

γ tjI t 1 ¼h ex;o;t

h o;t

where h ex;o;tis the conditional covariance between

exchange rate and oil price returns at period t, and

h o;t is the conditional variance of the oil price returns Then, the hedging effectiveness index (HE) is constructed as:

HE ¼ Var unhedged Var hedged

Var unhedged

IV Results and discussions

We first estimate average dynamic connectedness to get an idea about the average connectedness of the variables (section 4.1) Then, we calculate dynamic total connectedness to show the evolution of total connectedness over time and how the connected-ness is related to various events (section 4.2) We then calculate net total and pairwise directional con-nectedness to determine whether a variable is a net transmitter or receiver and to have an idea about the specific bilateral interactions (section 4.3) In addi-tion, we compare our results with the findings obtained from the commonly used generalized con-nectedness approach of Diebold and Yilmaz (2012,

2014) In section 4.4 we analyse the predictive power

of uncertainties and other variables on the connect-edness In section 4.5 we analyse the results from COVID-19 sub-sample and in section 4.6 we analyse hedge-ratio and hedging effectiveness

Average dynamic connectedness

In Table 2, we present the average connectedness

results based on TVP-VAR extended joint

connect-edness approach The rows in Table 2 correspond

to the contribution of each variable to the forecast

error variance of a variable in the system In

con-trast, columns correspond to the effect one variable

has on all other variables separately For example,

the second cell of the first column shows that oil

contributes 2.67% to the forecast error variance of

the Brazilian exchange rates, whereas the second

cell of the first row shows that Brazilian exchange

rates contribute 3.05% to the forecast error

var-iance of oil As we see from Table 2, crude oil

contributes most to the Canadian, Russian and

Norwegian exchange rates On the other hand,

Canadian and Norwegian exchange rates

contri-bute most to the variation in crude oil prices

The Row ‘To’ in Table 2, the second row from

the bottom, shows the total contributions of one

variable to all the others For example, the first cell

of the row ‘To’ implies that oil contributes

a combined 27.41% shock to the variations in all

the exchange rates in the network The Column

‘From’ (far right) in Table 2 shows total

contribu-tions from other variables to a variable For

exam-ple, the first cell of column ‘From’ indicates that

crude oil receives 28.3% of total contributions from

all the other variables in the network

The row ‘Net’ in Table 2, at the bottom, is the

difference between ‘To’ and ‘From’, a positive

(negative) number indicating that a variable is

a net transmitter (receiver) Table 2 shows that the

Canadian dollar is the largest net transmitter

(+5.09), and the Japanese Yen (−9.05) and Russian Ruble (−5.15) are the largest net receivers of shocks

On the other hand, the net value of crude oil is

−1.12 meaning crude oil, on average, is marginally

a net receiver of shocks Although it is difficult to pinpoint any specific characteristics, Table 2 shows that the exchange rates of oil-importing countries, except China, are usually net receivers of shocks The average value of the total connectedness index (TCI), the bottom-right cell in Table 2, is 44.51% which implies that this network of variables can explain the developments within the network reasonably well and there is a considerable co- movement among the variables of this network In other words, 44.51% of the forecast error variance within this network of variables can be regarded as the product of cross-market innovations

We now compare the results from TVP-VAR extended joint connectedness model reported in Table 2 with that obtained by using the Diebold and Yilmaz (2012, 2014) generalized connected approach (Appendix 1) Both approaches attest that Russian Rubel and Japanese Yen are the main net receiver of shocks; however, there are notable differences in the findings between the two approaches According to the generalized connectedness approach, Norwegian Krone and Euro are the largest net transmitter of shocks, whereas according to the joint connectedness approach, the Euro is a net receiver, and the Norwegian Krone is not among the largest net transmitters According to the joint connected-ness approach, the Canadian dollar is the largest net transmitter of shock

Table 2 Average connectedness.

Crude Oil Brazil Canada China Euro India Japan Mexico Norway Russia S Korea UK FROM

Results are based on a TVP-VAR model with lag length of order one (BIC) and a 20-step-ahead generalized forecast error variance decomposition.

Dynamic total connectedness

The average results presented in the previous

sec-tion are unable to show the evolusec-tion of the

con-nectedness over time They are also unable to

examine if connectedness is influenced by any

major economic or political event To overcome

these shortcomings, we need a dynamic framework

of analysis A dynamic framework not only

con-siders the evolution of the total connectedness

index (hereafter, TCI) over time but also

demon-strates how the role of specific variables within the

network of study may change over time (for

exam-ple, from net transmitter to net receiver or vice

versa) Figure 1 shows the TCI, where the black

shaded area is from the TVP-VAR-based joint

con-nected approach, and the red line is from the

Diebold and Yilmaz (2012, 2014) generalized

con-nectedness approach The black shade in Figure 1

shows that TCI varies considerably over time,

where a high TCI value implies greater spillovers

between the variables TCI is very high just before

and during the initial phase of the global financial

crisis during 2006–08, where the highest peak is

well above 80% After that, the TCI is again

relatively high in 2011–2014, coinciding with the European sovereign debt crisis Then with the col-lapse of crude oil prices in 2014, the level of con-nectedness fell sharply Again during 2016–17, connectedness was relatively high, indicating the era of Brexit Connectedness is again high during the COVID-19 crisis Therefore, TCI is indeed responsive to major economic events and connect-edness rises with the higher levels of uncertainty that usually coincide with such events The Diebold and Yilmaz (2012, 2014) generalized connected-ness approach (the red line) has also confirmed all the relevant peaks and troughs mentioned above Our findings are consistent with the find-ings of some other studies that find that the con-nectedness among financial assets increases during the period of crisis and uncertainty (Balli et al

2019; Mokni et al 2020)

Net total and pairwise directional connectedness

We next investigate the net total and pairwise directional connectedness Unlike the average con-nectedness in Section 4.1, the dynamic approach of

Figure 1 Dynamic Total Connectedness Index (TCI) The black shaded area is based on joint connectedness approach, while the red line is based on the generalized connectedness approach.