Carbon price prediction for China

Annals of Operations Research https://doi.org/10.1007/s10479-021-04392-7 ORIGINAL RESEARCH Carbon price prediction for China’s ETS pilots using variational mode decomposition and optimized extreme learning machine Shanglei Chai1 · Zixuan Zhang1 · Zhen Zhang2 Accepted: 29 October 2021 © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2021 Abstract With the national goal of “carbon peak by 2030 and carbon neutral by 2060 in China”, studies on carbon prices of China’s Emissions Trading System (ETS) pilots have shown growing interest in the related fields. Carbon price fluctuations reflect the scarcity of carbon resources, and accurate prediction can improve carbon asset management capabilities. Therefore, in order to clarify the dynamics of carbon markets and assign carbon emissions allocation rationally, we propose a hybrid feature-driven forecasting model with the framework of decomposition-reconstruction-prediction-ensemble. In this paper, the non-stationary, nonlinear and chaotic characteristics of carbon prices in China’s ETS pilots have been verified, and then the prediction model is built based on the tested features. Firstly, the original carbon price series are decomposed by Variational Mode Decomposition (VMD), and then reconstructed by Sample Entropy (SE). Next, Extreme Learning Machine (ELM) optimized by Particle Swarm Optimization (PSO) is conducted to predict the subsequences. Lastly, the forecasting series of every subseries are summed to obtain the final results. The empirical results based on carbon prices of China’s ETS pilots proved that the proposed model performs more efficiently than the current benchmark models. As carbon prices are expected to increase across all ETS during the post-COVID-19 recovery stage, the new prediction model will be useful for improving the guiding principles of the existing government policies including the likely introductions of Border Carbon Adjustment (BCA) in the EU and the US, and governing the large global public companies to deliver their “net zero” commitments. Keywords Carbon price forecasting · Emissions trading system (ETS) · Variational mode decomposition (VMD) · Particle swarm optimization (PSO) · Extreme learning machine (ELM) B Zhen Zhang 1 Business School, Shandong Normal University, Jinan 250358, China 2 Institute of Systems Engineering, Dalian University of Technology, Dalian 116024, China 123 Annals of Operations Research 1 Introduction Climate change is one of the global problems that is threatening the sustainable social development and safety of human lives, hence drawing profound attention of the communities. The academic community tends to proclaim a high degree of certainty about the causes of climate change (Macdonald et al., 2019), evolution trends (Westerhold et al., 2020) and the risks caused by it (Trisos et al., 2020). However, the main inducing factor of climate change is carbon emissions, which has been used as a significant reference point for the evaluation of future climate change in the prediction models (Tierney et al., 2020). In this connection, it can be perceived that reducing carbon emissions is the key to mitigating climate change and enactments of a variety of environmental regulation mechanisms, such as mandatory, voluntary, and market-oriented systems have been observed to reduce emissions (Chen et al., 2021). As a market-oriented policy instrument in particular, carbon emissions trading has been recognized to be an effective mechanism to achieve low-cost emissions reduction (Bauer et al., 2020; Cui et al., 2014; Daskalakis, 2013; Zhang & Wei, 2010). In the recent decades, China has been facing serious environmental problems. It is predicted that carbon intensity in the middle region will reach the highest in 2023 (Ma et al., 2020). The introduction of the Emissions Trading System (ETS) to tackle climate change has therefore been put forward on the policy agenda. Since 2013, China’s ETS pilots have been launched successively in Shenzhen, Shanghai, Beijing, Guangdong, Tianjin, Hubei, Chongqing and Fujian provinces. These pilots have covered nearly 3000 key emissions enterprises with a cumulative trading volume of 455 million tons, and a turnover of 10.55 billion yuan (approximately) till December 2020. The booming development of eight pilots thus promote the realization of “carbon peak by 2030 and carbon neutral by 2060 in China” (Liu & Sun, 2021; Qi et al., 2021; Xiao et al., 2021). However, the carbon price as a signal that transmits information in carbon markets, fluctuates violently due to a large number of potential factors such as macroeconomic events (Duc et al., 2021; Koch et al., 2014), energy prices (Adekoya, 2021; Balcilar et al., 2016), politically-driven policies (Duan et al., 2018), and overall situation of industry development from the macro and long-term perspectives as well as the supply and demand of carbon assets from the micro and short-term perspectives. Excessive price fluctuations would create risks for the market and weaken the effectiveness of ETS (Liu et al., 2020; Lyu et al., 2020; Song et al., 2019). Hence, it is necessary to forecast carbon prices accurately so as to enhance the enthusiasm of market participants. Evidence shows that carbon prices exhibit non-stationarity, nonlinearity, multi-scale and chaos (Fan et al., 2019; Huang et al., 2021; Manaf et al., 2016; Tian & Hao, 2020; Zhu et al., 2017; Zou & Zhang, 2020) and these make predictions difficult. Therefore, capturing features of carbon prices in order to maximise the forecasting precision is of great significance and it has become a challenging task for academic researchers. There are various carbon price prediction approaches that have emerged over the last few years. These approaches/models can be broadly classified into three categories: traditional econometric methods (Byun & Cho, 2013; Chang et al., 2017), artificial intelligence (AI) algorithms (Fan et al., 2015) and hybrid models. However, individual prediction models of econometrics or AI have shown lack of prediction robustness due to inherent shortcomings (Zhu & Wei, 2013; Zhu et al., 2015). To overcome the disadvantages above, decomposition and ensemble hybrid models have been introduced in carbon price forecasting. These models 123 Annals of Operations Research have grasped various characteristics and influencing factors of carbon price in a comprehensive manner and, therefore, been able to generate more accuracy in prediction outcomes (Li et al., 2021; Niu et al., 2019; Sun et al., 2021; Zhu et al., 2021). The core idea of hybrid models is to integrate the abilities of each algorithm, diversify the risk of single methods, and explore the complex internal information from various angles. Considering the high non-stationarity and nonlinearity features of carbon prices, decomposition by empirical mode decomposition (EMD) has been widely employed for maximi (...truncated)