A Multiple Timescales Conditional Causal Analysis on the Carbon-Energy Relationship: Evidence from European and Emerging Markets

ABSTRACT

This study aims to investigate the nexus between European and emerging markets in terms of multiple-timescale conditional analysis of the carbon-energy relationship. The findings identified the price movements of fossil fuels, Granger-caused movements in the carbon price, and movements in the carbon price Granger-caused movements in the electricity price. Furthermore, it was determined that in the long term, the crude oil and gas markets may increase and the coal market may decrease their causal influence on the carbon market. Finally, the role of the carbon market in the conditional Granger-causal network was observed to weaken during Phase III of the European Union Emissions Trading Scheme. These findings imply asymmetric information spillover between the European and emerging markets, particularly in the long term.

KEYWORDS:

• Carbon market
• energy markets
• conditional Granger causality
• Fourier transform method

JEL:

• C32
• G15
• Q50

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Supplementary Material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/1540496X.2023.2192346.

Notes

1. The 55% energy consumption of Europe relies on the imports. Sources: Eurostat.

2. Since the COVID-19 pandemic, as an exogenous shock, hugely disturbs the markets, we do not consider it further.

3. We transform the units of coal price, Brent oil price, and GSCI to the Euro bases by employing the daily spot euro – dollar exchange rate.

4. We selected the optimal order, whose value was 1, for the VAR model according to AIC. These causalities are significant at the 5% level.

5. The Fourier transform method has been used in studies involving multiscale causality analysis based on wavelet analysis (Yang 2019; Yang, Cai, and Hamori 2017) and empirical mode decomposition (Yu et al. 2015; Zhu et al. 2017, 2019). This transform is suitable for decomposing nonstationary, nonlinear, and complex data without introducing a fixed bias (Huang et al. 1998).

6. We also employed 125- and 500-day rolling windows to reexamine the issues. The empirical results were similar in all cases.

7. We summarized all the edges and their degree of conditional G-causalities in the causal network. In particular, the degree of conditional G-causalities in the causal network was defined as the total causal connectedness of the network.

8. To obtain the cross-power spectral density, we employ a Fourier transform approach to decompose the autocovariance sequence. Thereafter, we use the cross-power spectral density to obtain the conditional spectral G-causality. For details, please refer to Barnett and Seth (2014).

9. We also employed the frequencies of 125 Hz (half a year) and 500 Hz (two years) to decompose the raw data. These frequencies provided similar results to those obtained with a frequency of 250 Hz. Moreover, we investigated the conditional spectral G-causalities for Phases II and III, separately. Although marginal differences were observed in the degree of the conditional spectral G-causalities, the main results still hold. The results are available upon request.

10. We selected the optimal order, whose value was 1, for the VAR model according to AIC. These causalities are significant at the 5% level.

Additional information

Funding

This research was supported by the Natural Science Foundation of Guangdong Province under Grant Number 2023A1515030221, Research Foundation of Shenzhen University Grant No.20200826152220001, and JSPS KAKENHI Grant Number 22K01424, 19K1373801 and 19K1373821.