Global energy markets connectedness: evidence from time–frequency domain

Abstract

We examine the presence of dependence across 51 energy markets classified into different regions from Jan 2007 to June 2021. In order to examine the presence of dependence across different energy markets, we apply standard and threshold dependence measures proposed by Diebold and Yilmaz, Int J Forecast 28:57-66, (2012) and Baruník and Křehlík, J Financ Econ 16(2):271-296, (2018). We highlight the presence of strong dependence between the energy markets at both regional level and across other regions. European and American energy markets are highly connected within the region over the long-run whereas Asia–Pacific and the African energy markets offer optimal diversification opportunities. Both short- and long-run dependence exists between Chinese and the Hong Kong energy markets and between the US and Canadian energy markets. We also witness substantial increase dependence across all the energy markets during different crisis periods.

Introduction

Since the last couple of decades, investment in energy markets highlight an important avenue in the modern investment pattern. This has become more appealing especially during the recent awareness of energy shortage all across the globe and the resultant escalating energy prices. According to the International Energy Agency (2021), investments in global energy are increasing consistently and reached up to $1.8 trillion till 2019 which are mostly contributed by the fuel production (32%), energy infrastructure (30%), and power generation (28%). Despite significant growth in 2019, the first quarter of 2020 resulted in an annual decline of 20% in the global energy investment due to COVID-19 pandemic. However, according to the International Energy Agency, annual energy investments were later set to scale up significantly, rebounding approximately up to 10% in 2021, thus bringing the total volume of energy investment up to $1.9 trillion. Such rise in energy investments in 2020 also set up a forecasting level of an increasing 4.6% in 2021. These estimations are already reflected in the form of increasing global oil prices, with oil prices increasing to $83 compared to $36 per barrel, gas prices reaching to $5 from $3, and gasoline prices plunging to $80 from $45 in October 2020.1 As a result, a considerable number of investors are now keen to allocate energy assets in their investment portfolios which resulted in high volume of investments in the global energy market (Tang & Xiong 2012; Lin & Li 2015). However, the presence of integration among different energy markets displays pattern of returns co-movement, which makes portfolio allocation choices more challenging for individual as well as institutional investors (Naeem et al. 2022a, b, c). For instance, Bencivenga et al. (2010) report long-run integration among the European energy markets whereas in the short-run, relationship between these markets2 are unstable. Similarly, Mensi et al. (2021a) and Zhang and Broadstock (2020) observe the presence of dynamic connectedness among different energy commodities which increased significantly during the different crisis periods (Alawi et al. 2022; Karim and Naeem 2022). Such increase in connectedness among energy markets limiting diversification opportunities make portfolio choices more challenging for investors.

A significant strand of literature documents relationship between different energy commodities (Mensi et al. 2021b; Gupta et al., 2018; Ji et al. 2018; Wang & Guo, 2018). Singh et al. (2019) examine spillover connectedness among MSCI Europe, North and Latin America, Asia–Pacific, and the African energy market indexes and highlight significant connectedness in returns of European and the North American energy markets.3 Rehman (2020) examine pairwise returns co-movement among developed and emerging energy markets and report diversification benefits between the developed and emerging energy markets. However, after each crisis period,4 energy markets exhibit high levels of integration. Similar results are also reported by Rehman and Vo (2020) that portfolio comprising of World Developed and European alternative energy markets together with emerging or BRIC energy markets provide optimal returns under investment horizons ranging from intra-week to monthly period.

There are studies which focus on returns integration between energy commodities during the global financial crisis period in 2008–09 (Albulescu et al., 2020) and the COVID-19 pandemic period (Abadie 2021; Bouri et al. 2021; Zhang et al. 2021). According to Singh et al. (2019), energy markets in the Asia–Pacific region exhibit high connectedness with the Russia market which act as a major transmitter of spillover. Naeem et al. (2020) analyze relationship between energy, electricity, carbon, and clean energy markets and report time-varying connectedness which further intensifies during the global financial crisis period (2008–09). Increase in returns connectedness among different energy markets during the global financial crisis is also documented by Mandacı et al. (2020) who report that the volatility connectedness among global energy markets is time varying and increases significantly during the 2008–09 global financial crisis period. Lin and Su (2021) report significant increase in the total connectedness among energy markets following the outbreak of COVID-19 pandemic; however, this increase only lasted for two month’s period after which the connectedness declined to its pre-crisis level (Karim et al. 2022a, b; Billah et al. 2022; Alawi et al. 2022).

According to Akyildirim et al. (2022), connectedness among the global energy markets is high during uncertain times, COVID-19, and low economic sentiments. Benlagha and Omari (2022) examine the impact of COVID-19 outbreak on the dynamic connectedness between oil, gold, and five leading stock markets and report an increasing connectedness. They also show that gold act as a receiver whereas oil appear as a transmitter of shocks towards these stock markets. Likewise, more recently, Luo et al. (2022) investigate connectedness between gray energy and natural gas. They find that most of the gray energy indexes possess an ability to predict natural gas returns. The authors also show that connectedness from the WTI crude oil performs better for out-of-sample forecasting.

Our main contribution in this paper is to examine the presence of dependence among energy markets of Africa, America, Asia–Pacific, and Europe. Existing literature5 mainly focuses on the relationship between different energy commodities; however, we find limited studies examining the presence of dependence among energy indexes based in different countries. Therefore, our work fills this gap by examining the presence of returns dependence among a wide array of energy markets based in African, American, Asia–Pacific, and the European region. Our second contribution is to investigate dependence between energy markets during the normal and crisis periods which mainly comprise of the global financial crisis (2008–09) and the COVID-19 periods, thereby highlighting an important avenue of investigation under the short- and long-run investment periods.

In order to examine the presence of returns dependence among energy markets, we apply network dependency measure proposed by Junior et al. (2015). This technique has several advantages over other competing methods for measuring return connectedness. First, it is quite useful in estimating the dynamic structure among a large set of variables without involving the specification of complex econometric models. Second, this approach is effective in analyzing interdependence within a system without restricting the number of variables. In this way, this technique not only helps in measuring returns integration within a dynamic structure but also in examining co-movements within a system comprising of large number of variables. Another advantage of this technique over other connectedness measures is its ability to estimate the dependency network based on partial correlation which provides simple interpretation for investment purposes.

Results of our work highlight energy market dependence within as well as across different regions. Among other markets, European and American energy markets are more connected within the region in the long run. Asia–Pacific (except Russia) and African (except South Africa) energy markets are more suitable for investments attributable to their low connectedness with other energy markets within and across other regions. Chinese and Hong Kong from the Asia–Pacific region whereas United States and Canada from the American energy markets are highly connected both in the short- and long-run period during tranquil market conditions. Our results also highlight substantial increase in return dependence across all energy markets during the global financial crisis (2008–09) and the COVID-19 periods. Such increase in return dependence between different energy markets under crisis periods limits investment benefits for investors. Implications generated from our work are based on the heterogeneous behavior of energy markets across different regions which calls for careful selection of these securities in a single portfolio. Moreover, dependence among these energy markets increase during turbulent times which reduces diversification benefits during bearish market conditions. “Data and methodology” expounds data sources and estimation techniques under “Methodology.” “Analysis and discussion” provides data analysis and interpretation of results. “Conclusion” presents conclusions based on our results with implications for investment in the energy markets during normal and financially distressed periods.

Data and methodology

Data source

Data of our paper comprises 51 energy markets from different regions. These regions are classified as Europe, America, Asia–Pacific, and Africa. For the representation of European emerging markets, we sample Austria, Belgium, Croatia, Cyprus, Denmark, Finland, France, Germany, Greece, Hungary, Ireland, Italy, the Netherlands, Norway, Poland, Portugal, Romania, Slovenia, Spain, Sweden, and Turkey. For the American region, country list include Argentina, Brazil, Canada, Chile, Columbia, Peru, and the USA. For Asia–Pacific, we sample Australia, China, Hong Kong, India, Israel, Jordan, Japan, Korea, Kuwait, Malaysia, New Zealand, Oman, Pakistan, Philippines, Russia, Singapore, Sri Lanka, Taiwan, and Thailand. Finally, for the African region, we select Egypt, Morocco, and South Africa. Daily data for all energy markets ranges from 2007 to 2020. Daily returns for all energy markets are calculated by taking natural log of the difference between two adjacent pricing levels. We extract data for all energy markets from Thomson Reuters DataStream.

Methodology

Our methodological framework comprises time-varying spillover measures proposed by Diebold and Yilmaz (2012) and Barunik and Krehlík (2018). The spillover approach by Diebold and Yilmaz (2012) is effective in measuring time-varying connectedness which is superior to the conventional static models. This method does not use Cholesky factor identification associated with the VAR model due to which results are independent to the order of the variables. The application of Diebold and Yilmaz (2012) also enables us to measure pairwise as well as system-wide connectedness in a coherent and consistent way. Following Diebold and Yilmaz (2012), we also apply the spillover index of Barunik and Krehlík (2018) because of its advantage in keeping track of time as well as frequency domain. In this way, we can measure spillover across short- and long-run periods across different financial markets.

• (i)Time-domain spillover framework of Diebold and Yilmaz (2012)
  The time-varying spillover approach of Diebold and Yilmaz (2012) is based on the generalized vector autoregressive (VAR) model used to compute the forecast error variance decomposition. To begin, we consider the time, $t=1,\cdots ,T$, whereas the structural VAR $(p)$ illustrates the n-variate process $x_{t,1},\cdots .,x_{t,n}$ as follows:

  $$\phi \left(L\right)x_t={\epsilon }_t$$ 1

  where $\phi \left(L\right)={\sum }_h{\phi }_h{\mathrm{L}}^h$ represents an $n\times n$ coefficient matrix $t$ with lag polynomial running into infinity. Accordingly, the forecast error variance decomposition, in line with DY (2012), is given as

  $${({\theta }_H)}_{j,k}=\frac{{\left({\sigma }_{\mathit{kk}}^{-1},{\sum }_{h=0}^H,{\mathrm{\Psi }}_h,\mathrm{\Sigma },\left({\mathrm{\Psi }}_h,\mathrm{\Sigma }\right),{\left({\mathrm{\Psi }}_h,\mathrm{\Sigma }\right)}_{j,k}\right)}^2}{{\sum }_{h=0}^H{\left({\mathrm{\Psi }}_h,\mathrm{\Sigma },{\mathrm{\Psi }}_h^{{}^{\prime }}\right)}_{j,j}}$$ 2

  where ${\sigma }_{\mathit{kk}}={(\mathrm{\Sigma })}_{k,k}$, and ${\mathrm{\Psi }}_h$ represents an $n\times n$ coefficient matrix having lag $h$.${({\theta }_H)}_{j,k}$ explains the shock contributed by $k$-th variable to the forecast error variance of another variable $j$. Computationally, the summation of individual row ${({\theta }_H)}_{j,k}$ is not equal to unity. Hence, the summed-up result of the row matrix is normalizing it which can be expressed as

  $${({\overline{\theta }}_H)}_{j,k}=\frac{{({\theta }_H)}_{j,k}}{{{\sum }_{k=1}^H({\theta }_H)}_{j,k}}$$ 3

  where ${{\sum }_{k=1}^H({\overline{\theta }}_H)}_{j,k}=1$ and ${{\sum }_{j,k=1}^n({\overline{\theta }}_H)}_{j,k}=N$.
  The spillover in the proportion of the cumulated elements in the off-diagonal represents an overall summed up matrix as

  $$C_H=\frac{{\sum }_{j\ne k}{({\overline{\theta }}_H)}_{j,k}}{\sum {({\overline{\theta }}_H)}_{j,k}}\times 100=\left(1,-,\frac{Tr\{{\overline{\theta }}_H\}}{\mathrm{\Sigma }{({\overline{\theta }}_H)}_{j,k}}\right)100$$ 4

  where $C_H$ denotes the overall spillover of the network and $Tr\{.\}$ is the trace operator.
  Thus, the directional spillover transmitted (received) by a variable $j$, to (from) variable $k$, in the network, estimated as

  $${{(C}_H)}_{j\to }=100\frac{1}{n}{\sum }_{j\ne k,k}{({\overline{\theta }}_H)}_{j,k}$$ 5

  $${{(C}_H)}_{j\leftarrow }=100\frac{1}{n}{\sum }_{j\ne k,k}{({\overline{\theta }}_H)}_{j,k}$$ 6

  where ${{(C}_H)}_{j\to }$ represents “to spillover” and ${{(C}_H)}_{j\leftarrow }$ denotes “from spillover,” respectively. Next, the net spillover is computed by the difference between spillover transmitted and received, as below.

  $${{(C}_H)}_{j,\mathrm{net}}={{(C}_H)}_{j\to }-{{(C}_H)}_{j\leftarrow }$$ 7

  The positive value of net spillover ${{(C}_H)}_{j,\mathrm{net}}$ indicate that the variable $j$ is a net transmitter of shock whereas the negative value imply that the variable $j$ is net recipient of shock.

• (ii)Frequency domain spillover framework of Baruník and Krehlík (2018)

In order to examine the variability of return connectedness between energy markets across different frequencies, we follow Baruník and Krehlík (2018). Based on the expression in Eq. (2) where the impulse function ${\mathrm{\Psi }}_h$ is assumed to be time-varying, we made another assumption on the impulse function term to reflect the frequency domain. The frequency response function derived from $\mathrm{\Psi }$ becomes $\mathrm{\Psi }\left({\mathrm{e}}^{-iw}\right)={\sum }_h{\mathrm{e}}^{-iw}{\mathrm{\Psi }}_h$, which captures the coefficients of the Fourier transform, with $i=\sqrt{-1}.$ The generalized causation spectrum is expressed as

$${(f\left(\omega \right))}_{j,k}\equiv \frac{{\sigma }_{\mathit{kk}}^{-1}{\left|\mathrm{\Psi },(e^{-iw}),\mathrm{\Sigma },{)}_{j,k}\right|}^2}{(\mathrm{\Psi }\left(e^{-iw}\right)\mathrm{\Sigma }{\mathrm{\Psi }}^{{}^{\prime }}\left(e^{-iw}\right){)}_{j,j}}$$ 8

where,${(f\left(\omega \right))}_{j,k}$ reflects proportion of the spectrum of the j-th variable at frequency $\mathrm{\omega }$, made by shocks in the k-th variable. $\mathrm{\Psi }(e^{-iw})$ represents Fourier transform of the impulse response $\mathrm{\Psi }$. We compute the generalized forecast error variance decomposition on a specific frequency band $d$ by following Baruník and Krehlík (2018) as

$${(\theta }_d{)}_{j,k}={\frac{1}{2\pi }}^d\int {\mathrm{\Gamma }}_j(\omega ){(f\left(\omega \right))}_{j,k}d\omega$$ 9

where ${\mathrm{\Gamma }}_j(\omega )$ represents the weighting function. We define frequency-based connectedness on the frequency band by considering the spectral representation of the generalized forecast error variance decomposition as

$$C_d^F=100\left(\frac{{\mathrm{\Sigma }}_{j\ne k}({\overline{\theta }}_d{)}_{j,k}}{\mathrm{\Sigma }({{\overline{\theta }}_{\infty })}_{j,k}},-,\frac{Tr\{{\overline{\theta }}_d\}}{\mathrm{\Sigma }({{\overline{\theta }}_{\infty })}_{j,k}}\right)$$ 10

Therefore, the computation of the total spillover is estimated as

$$C_d^F=100\left(\frac{Tr\{{\overline{\theta }}_d\}}{\mathrm{\Sigma }({{\overline{\theta }}_d)}_{j,k}}\right)$$ 11

We can compute directional spillover at various frequencies like the time domain spillover framework. The “from”, “To,” and “net” spillovers can be calculated as

$${\left({\mathrm{C}}_d^F\right)}_{j\to }=100\times \left({\sum }_{j\ne k,k},(,{\overline{\theta }}_d,{)}_{j,k}\right)\frac{\sum ({\overline{\theta }}_d{)}_{j,k}}{\mathrm{\Sigma }({{\overline{\theta }}_{\infty })}_{j,k}}$$ 12

$${\left({\mathrm{C}}_d^F\right)}_{j\leftarrow }=100\times \left({\sum }_{j\ne k,k},(,{\overline{\theta }}_d,{)}_{j,k}\right)\frac{\sum ({\overline{\theta }}_d{)}_{j,k}}{\mathrm{\Sigma }({{\overline{\theta }}_{\infty })}_{j,k}}$$ 13

$${\left({\mathrm{C}}_d^F\right)}_{j,\mathrm{net}}={\left({\mathrm{C}}_d^F\right)}_{j\to }-{\left({\mathrm{C}}_d^F\right)}_{j\leftarrow }$$ 14

A positive value of ${\left({\mathrm{C}}_d^F\right)}_{j,\mathrm{net}}$ indicate that a specific energy market j is a net transmitter of shocks to other variables in the network whereas a negative value shows that the variable under consideration is net recipient of shocks from other variables.

Analysis and discussion

Table 1 presents descriptive statistics of all the energy markets which we sampled for our analysis and are clustered into four different regions. We also provide codes for each energy market for the ease of understanding. Among the all-equity markets, we see a mix of positive as well as negative daily average return values. The highest daily return value among the European energy markets is 3.9 percent for Denmark followed by Sweden (3.2 percent) and Turkey (2.6 percent). On the contrary, the highest mean loss of 3.6 percent is incurred by Ireland followed by a loss of 3.5 percent by the German energy market. Maximum standard deviation of 5.07 is exhibited by Ireland whereas Belgian energy market has the minimum deviation of 1.28 among other European energy markets. Among the American energy markets, Argentine provides maximum average returns of 0.05 percent while Peru exhibits maximum deviation of 3.32 percent. The energy market of Peru, Canada, and the USA exhibit negative average returns across the sampling period. For the Asia–Pacific region, we witness a mix of positive and negative daily returns where highest average daily return is for Kuwait (4.4 percent) followed by Sri Lanka (4.2 percent). However, in the African region, only the energy market in Morocco provides positive average returns. Jarque–Bera statistics suggest that the hypothesis of normal distribution is rejected for all the energy markets which is supported by the high kurtosis and negatively skewed values in most cases. Overall, based on our descriptive statistics, a mix of positive and negative average return values and the leptokurtic distribution with fat tails and negatively skewed values calls for a careful placement of energy securities in a portfolio.

Table 1.

Descriptive statistics

Region	Market	Symbol	Mean	Maximum	Minimum	Std. Dev	Skewness	Kurtosis	J-B	Obs
Europe	Austria	AUT	0.002	18.154	− 20.867	2.142	− 0.471	14.178	19,707.81a	3759
	Belgium	BEL	0.005	9.333	− 15.142	1.297	− 0.195	12.372	13,779.69a	3759
	Croatia	CRO	0.009	45.431	− 12.218	1.755	4.432	127.926	2,456,666.00a	3759
	Cyprus	CYP	0.000	16.737	− 10.534	2.108	0.315	8.941	5591.05a	3759
	Denmark	DEN	0.039	21.278	− 27.814	3.115	− 0.426	12.818	15,210.37a	3759
	Finland	FIN	0.051	21.270	− 12.747	2.213	0.195	9.377	6393.75a	3759
	France	FRA	− 0.012	13.519	− 18.094	1.710	− 0.282	15.538	24,669.86a	3759
	Germany	GER	− 0.035	17.244	− 18.772	2.313	− 0.403	8.931	5611.14a	3759
	Greece	GRE	− 0.012	14.101	− 18.540	2.112	− 0.141	8.979	5611.13a	3759
	Hungary	HUN	− 0.004	14.027	− 16.223	2.055	0.087	10.655	9182.48a	3759
	Ireland	IRE	− 0.036	49.247	− 47.084	5.069	0.114	18.522	37,742.93a	3759
	Italy	ITA	− 0.022	15.698	− 19.610	1.750	− 0.519	18.047	35,629.79 a	3759
	Netherlands	NET	− 0.017	18.751	− 20.917	2.251	− 0.464	13.060	15,984.69a	3759
	Norway	NOR	− 0.007	12.175	− 21.235	1.882	− 0.570	11.374	11,187.87a	3759
	Poland	POL	0.013	10.984	− 9.093	1.682	− 0.062	5.457	947.67a	3759
	Portugal	POR	0.010	22.069	− 18.054	2.108	0.342	13.372	16,922.66a	3759
	Romania	ROM	− 0.010	12.769	− 16.204	1.799	− 0.577	14.909	22,421.13a	3759
	Slovenia	SLO	− 0.007	12.480	− 9.484	1.536	0.123	10.475	8760.94a	3759
	Spain	SPN	− 0.013	12.387	− 13.807	1.728	− 0.437	10.006	7807.69a	3759
	Sweden	SWE	0.032	28.090	− 26.421	2.602	− 0.020	15.395	24,061.75a	3759
	Turkey	TUR	0.026	10.655	− 10.901	1.772	− 0.428	6.410	1935.80a	3759
America	Argentina	ARG	0.050	16.326	− 27.788	2.698	− 1.132	15.727	26,171.18a	3759
	Brazil	BRA	0.000	18.969	− 30.669	2.774	− 0.585	12.249	13,612.10a	3759
	Canada	CAN	− 0.013	14.156	− 22.034	1.746	− 1.181	24.011	70,020.39a	3759
	Chile	CHI	0.001	11.471	− 13.145	1.632	0.037	7.604	3321.15a	3759
	Columbia	COL	0.014	10.975	− 20.704	1.804	− 0.555	13.512	17,501.10a	3759
	Peru	PER	− 0.055	71.639	− 65.717	3.321	0.763	104.354	1,609,315.00a	3759
	USA	USA	− 0.005	17.327	− 23.605	1.866	− 0.921	21.081	51,737.96a	3759
Asia–Pacific	Australia	AUS	− 0.004	9.550	− 18.432	1.604	− 0.825	13.078	16,333.77a	3759
	China	CHN	− 0.002	17.239	− 17.153	2.093	0.260	10.846	9684.58a	3759
	Hong Kong	HKG	0.013	19.145	− 15.476	2.239	− 0.079	10.106	7911.66a	3759
	India	IND	0.029	16.464	− 16.462	1.658	− 0.564	15.515	24,730.38a	3759
	Indonesia	INS	0.025	15.968	− 31.535	2.289	− 0.488	16.879	30,320.46a	3759
	Israel	ISR	− 0.005	17.961	− 16.659	1.837	0.138	16.074	26,782.79a	3759
	Jordan	JOR	0.020	9.665	− 7.715	1.758	0.173	5.723	1180.55a	3759
	Japan	JPN	− 0.018	12.327	− 14.022	1.854	− 0.255	7.441	3129.56a	3759
	Korea	KOR	0.019	16.935	− 14.585	2.140	0.182	9.133	5911.27a	3759
	Kuwait	KUW	0.044	42.119	− 38.948	2.433	0.184	48.783	328,323.00a	3759
	Malaysia	MAL	0.016	9.266	− 10.110	1.137	− 0.358	12.664	14,708.50a	3759
	New Zealand	NZL	− 0.036	14.248	− 15.333	1.436	− 0.589	17.677	33,956.62a	3759
	Oman	OMN	− 0.012	9.859	− 12.622	1.332	− 0.496	16.474	28,590.75a	3759
	Pakistan	PAK	0.006	9.380	− 10.777	1.500	− 0.108	6.911	2403.29a	3759
	Philippines	PHI	− 0.030	17.191	− 21.574	1.910	0.122	18.930	39,752.79a	3759
	Russia	RUS	0.014	27.458	− 22.150	1.915	0.174	33.474	145,473.00a	3759
	Singapore	SIN	− 0.015	13.882	− 12.789	1.715	− 0.051	10.857	9671.58a	3759
	Sri Lanka	SRI	0.042	12.452	− 15.362	1.483	0.141	14.816	21,879.24a	3759
	Taiwan	TAI	0.008	9.443	− 11.246	1.650	− 0.061	6.752	2206.65a	3759
	Thailand	THL	0.012	13.268	− 30.036	1.901	− 1.037	24.835	75,345.36a	3759
Africa	Egypt	EGY	− 0.015	16.833	− 15.004	1.920	− 0.211	14.011	19,017.88a	3759
	Morocco	MRC	0.007	6.533	− 9.511	1.494	− 0.349	6.857	2406.37a	3759
	South Africa	SAF	− 0.009	21.974	− 43.094	2.310	− 1.625	42.224	242,623.20a	3759

J–B represents Jarque–Bera test of normality. ^aRejection of null hypothesis of normality at 1%

Figure 1 plots results of unconditional correlation among our sample energy markets. We cluster our sample period into (a) full sample period, (b) 2008–09 global financial crisis, and (c) COVID-19 crisis period. During our analysis of the complete sample period, we show that the energy markets are weakly correlated both within a region as well as across different regions. American and the Asia–Pacific energy markets are weakly correlated within themselves as well as across each other over the complete sampling period as opposed to the European energy markets where we witness relatively high regional correlation among few pairs. The correlation among all European energy markets except Belgium, Croatia, and Cyprus further intensifies during the global financial crisis period. Similarly, the magnitude of correlation among American energy market returns also increases during the global financial crisis. On the contrary, we witness weak return correlation among the Asia–Pacific energy markets during the GFC period except the Australian, Chinese, and Hong Kong energy markets (Naeem et al. 2022d; Karim et al. 2022c, d). It is also worth mentioning that correlation among European and American energy market returns also increases during the GFC period. During COVID-19 period, the correlation heat-map clearly highlights the presence of an increased returns correlation among the energy markets. This increased correlation is more pronounced among the European and American pairs which results in more interconnectedness among the energy markets across these regions. Similarly, Asia–Pacific energy markets highlight increasing pattern of returns integration with the European and American energy markets. Nevertheless, among all the sampled energy markets in our study, the markets of the Asia–Pacific region are less correlated within the region as well as across other markets suggesting the potential from diversification of assets during the crises periods (Naz et al. 2022; Pham et al. 2022; Naeem & Karim 2021). Overall, these results highlight that portfolio comprising of cross-regional energy securities provide optimal diversification opportunities both during normal and crises periods.

Fig. 1.

Correlation heat-map. (a) Full sample. (b) Global financial crisis (GFC). (c) COVID-19 crisis

We start our analysis by examining the pairwise directional dependency across our sample energy markets following Diebold and Yilmaz (2012). We classify energy markets under different regions and highlight in different colors. Red color represents European, blue indicate American, green is Asia–Pacific, and purple highlights African energy markets. The graphical depiction in of Fig. 2a highlights asymmetric relationship between pairs of energy markets indicating that the effect of energy market i on j is different from the effect of energy market j on i. The arrow pointing from one market to another highlights the direction of spillover. Our results highlight a high level of connectedness within European energy markets; however, results are not homogeneous for all markets. This adds to the findings by Veka et al. (2012) who also report significant relationship between European energy markets. For instance, AUT, SWE, SPN, POR, POL, NOR, NET, ITA, and ERA energy markets exhibit maximum returns coherence within region whereas SLO, ROM, IRE, HUNG, GRE, FIN, DEN, CYP, CRO, and BEL highlight least evidence of dependence within the region. Similar to European energy markets, our results highlight significant dependence among American energy markets. USA, CAN, COL, and BRA exhibit maximum regional coherence whereas PER, ARG, and CHI show low level of integration within the region. Among the Asia–Pacific energy markets, only RUS and SIN act as a recipient of spillover within the region, while on the other hand, we find no traces of regional integration among the African region. Therefore, compared with the European and American regions, Asia–Pacific and the African regions offer more diversification benefits for investments in the energy markets attributable to their low level of returns dependence with each other. Besides return connectedness between energy markets within different regions, significant integration exists between energy markets from different regions. For instance, AUT, SWE, SPN, POR, NOR, NET, ITA, and IRE among the European energy markets act as a major recipient of spillover from energy markets based in other regions whereas USA, CAN, and BRA receive major spillover of returns from European and the Asia–Pacific energy markets. Energy markets in Asia–Pacific act a major transmitter of spillover to energy markets in other regions except Russia which receives spillover from the European energy markets. One possible reason for such dependence between Russian and European energy markets might be because Russia is a major exporter of energy to the Europe (Rapaić and Novaković 2013). However, in the case of the African region, SAF energy market acts as a recipient of change from European energy markets where transmit changes towards the Asia–Pacific energy markets. Both EGY and MRC are neither connected within the region nor integration across the region. Therefore, diversification benefits lie for investments in the energy markets of Africa and Asia–Pacific though later is connected with the energy markets in other regions. Investments within the European and American energy markets need careful examination while developing a portfolio. Though we witness cases of strong dependence cross the regions, i.e., the American, Asia–Pacific, and the European regions, there still exist opportunities for investment within the region. For example, energy markets in Asia–Pacific and Africa provide optimal return opportunities for investors.

Fig. 2.

The network of return connectedness using Diebold and Yilmaz (2012). (a) Without thresholding. (b) With thresholding. Note: this shows the connectedness among 51 sampled energy markets, classified by regions. In (b), we only keep the values larger than the average of the 100 largest individual pairwise connectedness

In order to get a clear view of returns dependence between different energy markets, we simplify the network structure with a restricted number of edges by capturing the values which are larger than the average of the top 100 individual risk spillover. This simplified arrangement of dependency network only captures the most important pairwise relationship. Results in Fig. 2b confirm our findings that majority of the European energy markets are dependent within the region and FRA act as a net recipient of information from other European markets within the regions and from USA and CAN energy markets across other regions. FRA, ITA, SPN, and NOR also receive changes from other European energy markets whereas NET act as a transmitter of spillover towards other European energy markets. Among the American energy markets, we witness the transmission of information is greater in magnitude between USA and CAN compared with other American energy markets. Both USA and CAN remain net recipient of information from the Asia–Pacific energy markets where USA receives spillover from AUS and JPN energy markets whereas CAN act as a net recipient of spillover from the AUS energy market. In terms of dependence within the Asia–Pacific energy market, only CHN and HKG energy markets are integrated with each other. No other market neither transmit nor receives any change within the Asia–Pacific region. Notably, African energy markets are fairly segmented within and from all other regions. Overall, within region dependence among European and the American energy markets is high highlighting few investment opportunities which are similar to the findings by Lin and Li (2015) who show that increasing number of investors allocate energy assets in their portfolios which results in high integration among this asset class. On the contrary, Asia–Pacific and the African energy markets provide optimal diversification opportunities due to their low level of dependence.

Figure 3a plots network of return connectedness between energy markets classified by regions for short-run period. We follow network connectedness approach proposed by Baruník and Křehlík (2018) to examine short-run dependency among the European, American, Asia–Pacific, and African energy markets. Our results highlight that energy markets of AUT, SWE, SPN, POR, NOR, NET, ITA, and FRA appear as net recipients of transmission from other energy markets both within and across other regions, particularly from the America and the Asia–Pacific.

Fig. 3.

The network of return connectedness using Barunik and Krehlik (2018)—short run. (a) Without thresholding. (b) With thresholding. Note: this shows the connectedness among 51 sampled energy markets, classified by regions. In (b), we only keep the values larger than the average of the 100 largest individual pairwise connectedness

These results are in line with the previous results of Singh et al. (2019) who also report significant connectedness of returns among European and North American energy markets. USA, BRA, and CAN among the American energy markets appear as net recipient of spillover from both the regional and cross-regional energy markets with strong spillover in magnitude between the USA and the CAN energy markets. It is worth mentioning that the US energy market acts as a net recipient of spillover more than it transmitting change to other energy markets, thereby highlighting the sensitivity of US energy markets during the short-run period. The RUS energy market from Asia–Pacific act a net recipient of information from the energy markets of other regions whereas CHN and HKG are appear as active participants within region. On the other hand, only SAF energy market among the African energy markets receives as well as transmits spillover towards the energy markets of other regions; however, these African energy markets are not integrated within the region. Overall, we show that energy markets of Europe and America exhibit high level of dependence within the region and with the energy markets of other regions, thereby suggesting a careful placement of energy securities under a short-term investment horizon. Our results support a recent work by Rehman (2020) that the presence of returns co-movement has increased between the global energy markets which requires careful placement of these assets in a portfolio.

To get a clear view of the short-run network dependency among our sample energy markets, we simplify network with a restricted number of edges and only capture values which are larger than the average of top 100 individual pairwise connectedness. Our results in of Fig. 3b support earlier findings that the European and the American energy markets are more integrated within region compared to the cross-regional dependence. The FRA, ITA, NOR, and SPN act as a net recipient of spillover from other European energy markets whereas FRA energy market remains a net recipient of spillover from both European as well as the American energy markets. For American energy markets, highest dependency is witnessed between USA and CAN whereas BRA energy market transmits change both towards the USA and the CAN. On other hand, ARG, CHI, and PER energy markets neither receives nor transmits information in the short-run. Among the Asia–Pacific region, CHN and HKG exhibit high level of dependence with each other over the short-run investment horizon. It worth mentioning that only FRA from the Europe and USA among rest of the American energy markets act as a net recipient of change from other regions. These results suggest that significant diversification opportunities exit for investment under short-run period in the global energy markets. These results support the findings by Rehman and Vo (2020) that portfolio comprising of developed and the emerging energy securities provide maximum diversification benefits for investors.

Figure 4 presents network connectedness for all the energy markets using Baruník and Křehlík (2018) over the long-run period. Overall, we witness quite similar results as presented in Fig. 2. However, we note traces of information transmission towards BRA and RUA energy markets as shown in Fig. 4a. Regarding the integration level between energy markets within the European region, our results highlight that the SPN, NOR, ITA, and FRA energy markets exhibit maximum returns dependence whereas all other energy markets highlight low level of spillover. Beside regional dependence, significant integration exists among the cross-region energy markets. For instance, energy markets of AUT, SWE, SPN, POR, NOR, ITA, and FRA appear as net recipients of spillover from the American and Asia–Pacific energy markets. Among the American energy markets, USA acts as a major recipient of information followed by the BRA energy market. Notably, the CAN energy market acts as net transmitter toward the US and the CHI energy markets. On the other hand, the PER energy market shows no traces of dependence, neither within a region nor across other regions. For the Asia–Pacific region, the RUS energy market is the net recipient of spillover within the region as well as from its European counterparts whereas THI, AUS, and JPN appear as major transmitters of spillover towards the US and the European energy markets. Our findings are consistent with the results of Lin and Li (2013) who report significant correlation between the energy markets in Europe and Japan. However, energy markets within Asia–pacific highlight low dependence except only one case of high level of dependence between the CHI and the HKG energy markets. However, the SAF energy market exercise high returns dependence with the energy markets in other regions.

Fig. 4.

The network of return connectedness using Barunik and Krehlik (2018)—long run. (a) Without thresholding. (b) With thresholding. Note: this shows the connectedness among 51 sampled energy markets, classified by regions. In (b), we only keep the values larger than the average of the 100 largest individual pairwise connectedness

Figure 4b highlights a simplified view of long-run dependency network by restricting the number of edges and only keeping values larger than the average of the top 100 individual pairwise connectedness. We find traces of dependence within regions as well as across other regions. For example, the NOR and the FRA from the European energy markets appear as major recipients of spillover within region as well across other regions; however, POR and AUT act as net transmitters of spillover within Europe as well as towards the US energy market. Likewise, the US market appears as a net recipient of information from the energy markets in other regions whereas the CAN energy market act as a major recipient of spillover from the SAF, AUS, and JPN energy markets. Among the Asia–Pacific region, THL, AUS, and JPN appear as major transmitters of spillover towards the NOR, FRA, US, and CAN energy markets. However, energy markets of CHN and HKG are highly integrated with each other in the long run. The SAF energy market is a major transmitter of spillover towards NOR, FRA, US, and CAN energy markets. Notably, all the energy markets in Asia–Pacific and Africa appear as net transmitters of spillover with not a single market acting as a recipient of spillover during the long-run investment horizon. These results support the findings by Shen et al. (2018) that Asian energy markets highlight more integration in terms of risk transmission compared to the US and the European markets.

Figure 5 presents total time-varying connectedness among all energy markets following Diebold and Yilmaz (2012). The purpose of such analysis is based on the fact that the sampling period does not follow a smooth timeline rather consists of financially turbulent periods including the Global financial crisis (2008–09), the ESDC period (2012–14), and the recent COVID-19 pandemic. Therefore, the interdependency among our sampled energy markets tends to vary considerably following a dynamic correlation patterns. Our findings highlight significant increase in connectedness among energy markets during all the crises periods. For instance, the escalated value during 2008–09 is attributed to the global financial crisis (GFC) period which marks high level of dependency among the energy markets. Later, this increase is followed by a decline during the post crisis period until the beginning of 2012–14 ESDC. The total dependency again increased in the beginning of the ESDC in 2012; however, the magnitude of increase in relatively low compared to the 2008–09 GFC. Later, we witness that the total dependency decreases substantially between 2013 and 2015 indicating diversification opportunities for investors. This decrease in dependency is followed again by increasing level of connectedness when the global economy was faced by one the largest decline in oil prices in modern history, thereby limiting benefits from diversification. We again witness high dependency among the energy markets at the end of the sample period mainly attributed to the global COVID-19 pandemic. Overall, the dependency network suggests turbulent patterns across the entire sample period with escalated dependence levels during economic and financial distressed periods. These results are consistent with the findings by Albulescu et al. (2020) that co-movement among different energy markets increases during extreme market conditions. Such volatility in the total dependence correspond with the pattern in global business cycle. The graph of total dependency increases in value when global market experience financial and economic crisis, whereas decreases significantly during the periods of economic recovery and expansion.

Fig. 5.

Total time-varying connectedness using Diebold and Yilmaz (2012). Note: this figure shows the rolling-window version of total connectedness. The rolling-window size 260 days

Figure 6 presents total time-varying connectedness among our sample energy markets following Baruník and Křehlík (2018). The red color in graph presents long-run dependency whereas green color shows short-run dependency among the energy markets. Our results highlight an increase in the energy market dependency during different crisis periods like GFC (2008–09), ESDC (2012–14), and the recent COVID-19 pandemic both during the short- and long-run periods. However, the magnitude of dependency is relatively strong in the short run. These results support the earlier findings by Lin and Su (2021) that connectedness among the global energy markets increased significantly during the COVID-19 pandemic. These results suggest that the presence of turbulent market conditions affect dependence level between different energy markets more during the short-run and,therefore, carry implications for the short-run investments. Since the interdependency across energy markets remains consistent under normal market conditions, it increases significantly under periods when the global market experiences financial and economic crises. Therefore, the overall network dependency behaves in a cyclical pattern following the global business cycle. These results highlight limited diversification benefits during the crisis periods both during the short- and long-run periods; however, diversification opportunities increase during periods of expansion and recovery.

Fig. 6.

Total time-varying connectedness using Barunik and Krehlik (2018). Note: this shows the rolling-window version of short run (green) and long run (red) connectedness. The rolling-window size 260 days

Figure 7 presents network of return connectedness for our sample energy markets during the 2008–09 Global financial crisis period. Our results presented in Fig. 7a highlight strong dependency between energy markets during the GFC period. All the energy markets are highly connected within region and across different regions. Such increase in interdependency highlight that during the financially turbulent period, investing in energy markets provides limited diversification benefits, thus following contagion and therefore requires a careful placement of energy securities in a portfolio during such periods. Albulescu et al. (2020) report similar findings that returns coherence among different energy markets increases during crisis periods. In a next step, to get a simplified view of the network dependency among our sample energy markets during the GFC (2008–09) crisis period, we simplify the dependence network with a restricted number of edges by only capturing values which are larger than the average of top 100 individual pairwise connectedness. Our results (Fig. 7b) support the earlier findings that energy markets are connected both within and across other region’s energy markets. For instance, the energy market of SLO acts as a recipient of spillover from not only from the GRE (regional) but also from the energy markets in the Asia–Pacific region. Similarly, NOR energy market receives information from ITA (regional) as well as from the USA (cross-region) energy markets and, furthermore, transmits changes toward the COL (cross-region) energy market. The DEN energy market acts as a recipient of information from the TUR (regional), SIN, and PAK (cross-regional) energy markets whereas GER receive information from IND during the global financial crisis. In case of American Energy markets, BRA and COL act as recipient of changes from the European energy markets of SPN and NET, respectively. From the perspective of American energy markets, CAN transmits spillover towards SLO and US towards the NOR (European) and SAF (African) energy markets. It is worth mentioning that, American energy markets are not connected within the region during the 2008–09 global financial crisis which is quite from the Asia–Pacific energy markets where we witness both regional and cross-regional dependence during the global financial crisis. For example, energy markets of SRI, SIN, HKD, IND, and MAL act as recipient of change from their European counterparts whereas, SIN and SRI receive information from both regional as well as European energy markets. Such increase in interconnectedness during the crisis period within as well as across different regions limit opportunities for diversification. Therefore, investors need careful examination before placing energy securities in a portfolio during crises periods.

Fig.7.

The network of return connectedness using Diebold and Yilmaz (2012)—global financial crisis. (a) Without thresholding. (b) With thresholding. Note: this shows the connectedness among 51 sampled energy markets, classified by regions. In (b), we only keep the values larger than the average of the 100 largest individual pairwise connectedness

Figure 8a presents network of return connectedness under the short-run period during the 2008–09 Global financial crisis period following Baruník and Křehlík (2018). Our results highlight strong short-run interconnectedness between energy markets within and across different regions during the GFC period which limits the diversification opportunities for investors. In Fig. 8b, we simplify network with a restricted number of edges and only capture values which are larger than the average of top 100 individual pairwise connectedness. Our results support earlier findings regarding the regional and cross-regional connectedness between energy markets. Among the European Energy markets, SLO act as recipient of information from IRE, HUN, and GER (regional) along with the CAN (American) and PHI (Asia–Pacific) energy markets. The NOR energy market receives spillover from the FRA and GER (regional) as well as from USA and CAN (American) energy markets. Both GER and FRA (European markets) receive changes from the IND and AUS (Asia–Pacific markets), respectively. In the case of American energy markets, we only witness COL as spillover recipient from the NET (European), OMN, and AUS (Asia–Pacific) energy markets. However, the energy markets of US and CAN act as net transmitters of change towards other regions. Energy markets in Asia–Pacific are also integrated in the short-run with regional as well as with cross-regional energy markets. For example, energy markets in SRI, TAI, HKG, IND, MAL, NZL, and OMN receive spillover from their European counterparts whereas RUS and KOR act as recipient of change from CAN (American) and SAF (African) energy markets, respectively. Notably, the energy market of EGY act as recipient of information from the European energy markets whereas SAF receives changes from both USA and the European energy markets. On the other hand, MRC transmits change towards IRE (European) whereas SAF spillover towards the KOR (Asia–Pacific) energy market during the short-run period. Overall, a careful investigation about energy assets placements in a portfolio is required at regional as well as at international level under the short-run investment horizon.

Fig.8.

The network of return connectedness using Barunik and Krehlik (2018)—short run–global financial crisis. (a) Without thresholding. (b) With thresholding. Note: this shows the connectedness among 51 sampled energy markets, classified by regions. In (b), we only keep the values larger than the average of the 100 largest individual pairwise connectedness

Figure 9 presents network of return connectedness between energy markets during GFC under long-run period. Our results presented in Fig. 9a highlight strong dependence between energy markets in the long-run during the GFC period. The energy markets in all four regions are highly interconnected not only within the region but also across different regions. In addition, each energy market acts as a recipient as well as transmitter of spillover over the long-run. Such increase in coherence at regional and international level suggests limited diversification opportunities for investors under distressed financial situations. Our findings are consistent with the results by Rehman (2020) that integration among energy markets increased significantly during the GFC period (2008–09). Figure 8b again presents long-run network connectedness after simplifying the network with a restricted number of edges and only capturing values which are larger than the average of top 100 individual pairwise connectedness. Our results highlight that European energy markets receive changes from cross-regional energy markets more than transmitting changes towards other regions. For instance, SLO, POR, IRE, GER, GRE, DEN, and CRO from the Europe receive spillover from the Asia–Pacific energy markets whereas, IRE, GER, and DEN receive changes from the American energy markets. AUT energy market receives spillover from both EGY (African) and ARG (American) energy markets during the GFC period. Notably, energy markets in the American region do not exhibit dependence with each other; however, CHI and PER act as recipient of information from the Asia–Pacific energy markets. Similarly, SRI and IND (Asia–Pacific) receive changes from TUR and CRO (European) energy markets, respectively. JOR energy market is a net recipient of information from the ARG (American) energy market. On the contrary, African energy markets are neither connected within the region nor receive any spillover from any other energy market during the GFC. Nonetheless, EGY energy market acts as net transmitter of change towards SLO and AUT (European) energy markets in the long run. These results highlight regional diversification opportunities for investors holding portfolio of African energy markets.

Fig.9.

The network of return connectedness using Barunik and Krehlik (2018)—long run–global financial crisis. (a) Without thresholding. (b) With thresholding. Note: this shows the connectedness among 51 sampled energy markets, classified by regions. In (b), we only keep the values larger than the average of the 100 largest individual pairwise connectedness

Figure 10a plots dependency network of Diebold and Yilmaz (2012) for the sampled energy markets during the COVID-19 pandemic period. Our findings highlight strong presence of dependence between the energy markets within region as well as across other regions. Our results of such strong connectedness between different energy markets during the COVID-19 pandemic is consistent with the finding by Lin and Su (2021) who report a dramatic increase in total connectedness among energy markets in the beginning of COVID-19 crisis. We witness that each single market exhibits bidirectional spillover with other energy markets during the COVID-19 pandemic. This increase in dependency suggest limited diversification opportunities for investors during the pandemic. In order to get more clear view of network connectedness, Fig. 10b simplifies the network with a restricted number of edges and only capture values which are larger than the average of top 100 individual pairwise connectedness. Our results highlight that among the European energy markets, SLO acts as a net recipient of spillover from American (COL and CHI), African (SAF), and Asia–Pacific (RUS, NZL, JOR, and JPN) energy markets. These results are different from the ones reported by Singh et al. (2019) that the energy markets within the Asia–Pacific region are connected with the Russian energy markets which act as a major transmitter of spillover. In addition, NOR appears as the major recipient of information from the TUR and SWE energy markets during COVID-19 pandemic. On the other hand, American energy markets act as a net transmitter rather than the recipient of change from other energy markets. We witness that COL, CHI, and ARG from the American energy markets transmit spillover towards SLO (European) and PAK (Asia–Pacific) energy markets during the COVID-19 pandemic. In the case of Asia–Pacific energy markets, SRI acts as a recipient of change from European and energy markets within the region whereas PAK receives spillover from ARG during the pandemic. Among the African energy markets, only SAF transmit changes towards SLO (European) energy markets. Despite both regional and cross-regional interconnectedness among our sample energy markets, there exist few opportunities for diversification for investors both at regional as well as at international level even during the pandemic.

Fig. 10.

The network of return connectedness using Diebold and Yilmaz (2012)—COVID19. (a) Without thresholding. (b) With thresholding. Note: this shows the connectedness among 51 sampled energy markets, classified by regions. In panel (b), we only keep the values larger than the average of the 100 largest individual pairwise connectedness

Next, short-run dependency network across our sample energy markets during the COVID-19 pandemic is depicted in Fig. 11 following Baruník and Křehlík (2018). Figure 11a shows short-run dependency network among energy markets during the COVID-19 pandemic. Our results reveal that energy markets are highly dependent within and across other regions. Such increase in network dependency suggests limited opportunities for investors for portfolio diversification in the short run during the pandemic. To get a clearer view of the dependency network, we simplify network with a restricted number of edges and only capture values which are larger than the average of top 100 individual pairwise connectedness in Fig. 11b. Our results highlight that, among the European Energy markets, only CRO and ROM energy markets act as recipient of spillover from Asia–Pacific energy markets whereas BEL energy market is the major transmitter of change towards the SRI (Asia–Pacific) energy market. In case of American energy markets, we witness that in the short run, ARG energy market receives spillover from the CRO (European) and KUW (Asia–Pacific) energy markets during the COVID-19 pandemic whereas CHI receives change from the DEN (European) and CAN energy market (within the region). On the other hand, USA, BRA, and PER energy markets transmit change towards the energy market in other regions. Among the Asia–Pacific energy markets, SRI act as major recipient of spillover from the European and the American energy markets. PHI (Asia–Pacific) receives spillover from IRE (European) and BRA (American) energy markets. Likewise, PAK energy market acts as recipient of change from the ARG (American) energy market during the COVID-19 pandemic in the short run. Notably, majority of the Asia–Pacific energy markets are connected within a region implying limited benefits from diversification by investing only in the energy sector of Asia–Pacific region. These results are supported by the recent work of Akyildirim et al. (2022) that connectedness among the global energy markets increases significantly during the COVID-19 period. The case of African energy markets is similar to Asia–Pacific markets in terms of receiving spillover of change from both the European and the American energy markets. For instance, SAF energy market act as recipient of information from the IND and the PHI (Asia–Pacific) energy markets together with the IRE (European) energy market during the COVID-19 pandemic, whereas MRC receives information from NET (European) as well as from US and CAN (American) energy markets during the pandemic implying limited diversification benefits in the short run.

Fig.11.

The network of return connectedness using Barunik and Krehlik (2018)—short run–global financial crisis. (a) Without thresholding. (b) With thresholding. Note: this shows the connectedness among 51 sampled energy markets, classified by regions. In (b), we only keep the values larger than the average of the 100 largest individual pairwise connectedness

Long-run network of return connectedness across our sample energy markets during the COVID-19 pandemic is presented in Fig. 12 following Baruník and Křehlík (2018). Our results depicted in the Fig. 12a highlights strong long-run dependence between the energy markets during the COVID-19 pandemic. We witness that each single market exhibits bidirectional spillover in the long run during COVID-19 pandemic. Among the European energy markets, SLO, POL, and NOR act as major recipient of spillover from the energy markets in Asia–Pacific and the American region, while CAN among the American energy markets acts as a major recipient of spillover from the European and Asia–Pacific energy markets. Similarly, COL acts as a major transmitter of change towards the European and the Asia–Pacific energy markets in the long run during COVID-19 pandemic. From the perspective of Asia–Pacific energy markets, THI, SRI, SIN, PAK, JOR, JPN, and INS appear as a major recipient of information from the European energy markets during the pandemic. On the other hand, SAF and EGY exhibit bidirectional spillover towards the energy markets in other regions. In Fig. 12b, we present long-run network connectedness after simplifying the network with a restricted number of edges and only capturing values which are larger than the average of top 100 individual pairwise connectedness. Our results highlight that European energy markets receive spillover from the energy markets in other regions whereas transmits information only towards the SIN (Asia–Pacific) energy market. The SLO energy market acts a net recipient of spillover from many cross-region energy markets like Asia–Pacific, African, and American regions. Furthermore, the SLO energy market also receives spillover within region during the COVID-19 in the long run. Notably, all energy markets in the American region are sensitive to regional as well as cross-regional spillover except CAN which receive information from the JPN (Asia–Pacific). The COL energy market transmits information towards SLO and NOR (European) and SIN (Asia–Pacific) energy markets during COVID-19 period. Similarly, SIN, JPN, and NZL among the Asia–Pacific energy markets are connected with cross-regional energy markets in the long run whereas all other markets remain insensitive to changes in other energy markets during the COVID-19 pandemic. On the other hand, EGY among the African energy markets acts as net transmitter of information towards SLO (European) whereas SAF energy market transmits information towards SLO and POL (European) as well as towards the SIN (Asia–Pacific) energy market. Overall, though we find few traces of regional and cross-regional connectedness, our findings suggest diversification opportunities for within and cross-region-based energy portfolio during the COVID-19 pandemic in the long run.

Fig. 12.

The network of return connectedness using Barunik and Krehlik (2018)—long run–global financial crisis. (a) Without thresholding. (b) With thresholding. Note: this shows the connectedness among 51 sampled energy markets, classified by regions. In (b), we only keep the values larger than the average of the 100 largest individual pairwise connectedness

To ensure that our results discussed above are not sensitive to the choice of forecasting horizon ($h)$ or the window length $\left(W\right)$, we change our baseline rolling window and forecast horizon and select all possible combinations from the following parameter sets: $w$ $\in$ {200; 260; 300} and $h$ $\in$ {75; 100; 125}. Figure 13 plots results from the nine possible combinations where all the forecasting horizons ($h)$ or the window length $\left(W\right)$ have comparable paths and, therefore, overlap most of the time. Furthermore, all the choices lead towards several boost points, including the crisis period of global financial crisis in 2008–09, Eurozone crisis in 2010–12, global oil crash in 2015, and the COVID-19 pandemic in 2019–20. Hence, our findings provide evidence that the results are not sensitive to the choice of forecasting horizon ($h)$ or the window length $\left(W\right)$.

Fig. 13.

Robustness to the choice of rolling window and forecast horizon. Notes: this shows the results for each other combination of window-length $w$ $\in$ {200, 260, 300} and forecast-horizon $h$ $\in$ {75, 100, 125}

Conclusion

We examine the presence of dependence across different energy markets classified into regions, i.e., Africa, America, Asia–Pacific, and the Europe, covering period from Jan 2007 to June 2021. Our analysis period covers significant time period encompassing various financial and economic crisis periods, including the global financial crisis (2008–09) and the recent COVID-19 pandemic. To examine the presence of dependence across different energy markets, we apply normal and threshold dependence measures proposed by Diebold and Yilmaz (2012) and Baruník and Křehlík (2018). Our results highlight strong dependence between our sampled energy markets at both regional as well as across the region. We find that European and American energy markets are highly connected within the region over the long-run. Asia–Pacific (except RUS) and the African energy markets (except SAF) offer optimal diversification opportunities by highlighting low connectedness of energy returns within the region as well as across other regions. We also find that CHI and HKG among the Asia–Pacific and USA and CAN among the American energy markets are highly connected both in the short and long run. The magnitude of such dependence further intensifies during different crisis periods. We witness substantial increase in returns dependence across all the energy markets both within region and across the energy markets in other regions. Such increase in the dependence between different energy markets during the crisis periods limits chances of optimal portfolio returns using the diversification strategies. Results of threshold dependence measures support the presence of short- and long-run diversification opportunities which not only exist within region but also across other regions.

Our results carry important implications for the investment community. We examine the presence of dependence between energy markets individually within and across certain regions. Knowledge of dependence between the energy markets on country basis can help individual as well as institutional investors for investment purposes. However, presence of dependence on the country level can not only benefit investors but also policy makers. There are studies which examine the effectiveness of inter-regional against within region diversification (Ahmad et al. 2022; Narayan & Rehman 2021); so, our work also contributes in this aspect. Policy makers can analyze and assess market conditions across different regions in terms of efficiency to lure investment from international investors. Our findings imply a careful selection of energy securities in a portfolio for optimal returns. We witness a heterogeneous behavior of different energy markets with other energy securities in a single portfolio. There is no consistent behavior for any one single security together with other securities in a portfolio. One security may yield optimal returns in one combination but may not be appropriate in terms of portfolio returns with another combination. Therefore, the possibility of returns exists but the selection of portfolio should be based on the dependence structure between different securities. Another implication arising from this work is the behavior of energy securities structured within different regions. We witness many securities offering diversification benefits with the energy securities in one region but no benefits in another region. Likewise, the presence of heterogeneous dependence structure among securities within a single region rules out the possibility of regional diversification benefits. These benefit though may exist for some energy assets but a careful examination of their underlying dependence behavior is important. Finally, the contagion phenomena seem quite obvious during the crises periods, i.e., GFC and the COVID-19 periods. However, few markets offer diversification opportunities under these financially and economically stressed periods. We recommend directions for future research by expanding the scope of this study to include other asset classes along with the green energy market, for example, green energy bonds, conventional equities, socially responsible stocks. In this way, more implications can be generated for diversification in portfolio comprising of different assets. Because of the difference in asset classes, such combinations can result in optimal performance of portfolios. Another avenue of research could be to examine the energy markets on the regional level as well. Such an analysis can add robustness to results of this paper which sampled individual energy markets within a certain region.

Appendix

Appendix Table 2

Table 2.

Analysis-oriented attribute table

Authors	Sampling	Method
Ahmad et al. (2022)	Stocks markets of BRICS	Panel cointegration and panel regression
Akyildirim et al. (2022)	Global energy equity indexes	Spillover approach of Antonakakis et al. (2020)
Benlagha and Omari (2022)	Gold, oil and five leading stock markets	Gabauer's (2020) DCC-GARCH connectedness approach
Luo et al. (2022)	Grey energy market to the natural gas market	Connectedness framework of Antonakakis et al. (2020)
Narayan and Rehman (2021)	Stock markets Asia, Central and Eastern Europe (CEE), Latin America, or the Middle East and North Africa (MENA)	Panel cointegration tests and vector autoregressive error-correction (VECM) models
Abadie (2021)	Spanish electricity and natural gas prices	Stochastic model with deterministic and stochastic parts
Albulescu et al. (2020)	Energy, agriculture, and metal commodity markets	Copula-based local Kendall’s tau approach
Baruník and Křehlík (2018)	11 major financial firms representing the financial sector of the U.S. economy	Barunik and Krehlik (2018) spillover approach
Algieri and Leccadito (2017)	Energy, food, and metal commodity markets	Delta conditional value-at-risk (ΔCoVaR) approach based on quantile regression

Author contribution

Mobeen Ur Rehman: idea generation, initial write up, and discussion of results. Naeem Abubakr: data curation, software, and formal analysis. Nasir Ahmad: initial write up and discussion. Xuan Vinh Vo: idea, supervision, and editing.

Funding

This research is partly funded by the University of Economics Ho Chi Minh City, Vietnam.

Availability of data and materials

Data will be made available for request.

Declarations

Ethical approval

The paper does not have any ethical concern and does not contain any primary data.

Consent to participate

The paper is based on secondary data and therefore does not involve participation of any respondent.

Consent to publish

The authors give ESPR a right to publish this paper.

Competing interests

There are no competing results to declare.