Fragmented or Unified? The State of China’s Carbon Emission Trading Market

Abstract

China has adopted a gradualism principle in establishing its carbon emission trading system from the regional pilot markets to a national one. In view of the huge market potential and large differences across China, this paper applies the cointegration test and VECM (Vector Error Correction Model) to investigate the long-run trends and the price dynamics of regional pilot markets. The results show that the prices of the regional pilot markets form three long-run trends exhibit mean-reversion patterns. The launch of the national market marks the different performance of regional pilot markets. After the launch of the national market, the number of long-run trends reduces from three to one and the severity of the mean-reversion patterns is eased, indicating the efficiency improvement in China’s carbon market from a fragmented toward a unified market. The policy implication is that China should further develop its national market by incorporating the regional markets into the national one and encouraging more market participants for market transaction.

1. Introduction

As the second largest economy with the largest population in the world, China’s economic development has also made it one of the largest carbon-emitting countries [1,2]. China’s switch to low-carbon development is pivotal to the world’s climate targets. In order to fulfill the commitments to the 2016 Paris Agreement and to promote low-carbon economic development, China established a carbon emission trading system in 2013, implementing regional pilot markets one by one, followed by implementation of the national market in 2021. Carbon emission trading requires an effective monitoring, reporting, and verification system involving sufficient information and coordination [3]. As a financial market, the effective operation of the carbon emission market is the key to the realization of low-carbon development and carbon emission reduction. However, the efficiency of carbon emission trading markets in China has not yet been thoroughly explored. This paper attempts to shed some light on this topic by using the VECM (Vector Error Correction Model) method to investigate the price dynamics of the regional pilot markets and then assess the efficiency of the carbon emission trading market.

As the largest and the most mature carbon market, the carbon emission trading market in Europe was established much earlier than that in China, making it the primary subject of early research [4,5]. The research focuses on the efficiency of the markets first and then the interactions of carbon prices with other sectors. Regarding the efficiency of markets, Montagnoli and de Vries tested the efficient market hypothesis for the European Union Emissions Trading Scheme (EU ETS) using the variance ratio tests. They claim that the European carbon market is inefficient in the trial and learning phase and the efficiency restoration in phase II [6]. Charles et al. investigated the market efficiency of the EU carbon market in phase II by looking at the cointegration relationship between the futures contracts and spot prices. They show that the cost-of-carry model is rejected, signaling market inefficiency to some extent [7]. Tang et al. used the VECM method to investigate the pricing efficiency of futures contracts for the EU carbon market. They find that the futures contracts for the EU carbon market are efficient within one month [8]. Richstein et al. applied an agent-based simulation to study the effects of the backloading of EU emission allowances and the market stability reserve (MSR) on carbon price and volatility [9].

Regarding the interactions of carbon prices with other sectors, Ji et al. investigated the information linkages and spillover between carbon and energy markets in Europe. The carbon emission price is subject to the influence of Brent oil prices [5]. Wang et al. detected a cointegration relationship between the European and Chinese carbon markets. Feedback exists from the carbon market to other energy markets [10]. Zhou et al. studied the multidimensional risk spillover effects among carbon, energy, and nonferrous metals markets. They also evaluated the performance of portfolio diversification. Their research shows that there are dynamic risk spillover effects among the carbon, energy, and nonferrous metal markets, in addition to different risk spillovers in different dimensions [11]. Demiralay et al. explored the hedging function of carbon credit futures and claim that carbon futures demonstrate hedging and diversification benefits [12].

As an emerging market with huge market potential, China’s carbon market has become the frontier of research. Compared to mature financial markets, the pilot markets of China have a short history and suffer some shortcomings in terms of trading volume, market liquidity, and information transparency [13,14]. Zhao et al. summarized factors contributing to the shortcomings of the pilot markets. The low degree of participation of covered enterprises leads to poor market liquidity. The concentration of emission allowances is high, with the top 10 enterprises having 60–70% of total allowances in some pilot markets. The high concentration of emission allowances makes market liquidity poor and market efficiency low. Lastly, insufficient demand and supply lead to a limited trading volume and poor market liquidity [15]. Zhao et al, applied the ADF and run tests to evaluate the efficiency of the pilot markets in China. They concluded that China’s carbon trading market will improve its efficiency when the market is relatively mature, reaching weak-form efficiency [16]. In contrast, empirical analyses using advanced econometric methods reveal that China’s carbon markets are not effective, given that the prices could be impacted by historical prices [17,18]. Wang et al. also considered the Chinese carbon markets as inefficient, given that five out of six pilot markets follow a mean reversion process [19]. Wen et al. investigated the driving factors for the carbon price dynamics in China. Macro economy risk and uncertainty, energy, and environmental factors all impact the carbon price fluctuations and connectedness of the whole system [20]. Xia et al. reported that the carbon market tends to be the risk taker for the risk spillover from high-carbon-emitting industries [21].

The institutional arrangement of regional pilot markets has led to research on how the regional prices interact among themselves. Zhao et al. examined nonlinear Granger causality and the time-varying effect in China’s pilot markets. It was found that bidirectional nonlinear Granger causalities exist among the pilot markets, besides the time-varying impulse responses [22]. Guo and Feng explored the spillover effects among the pilot markets and reported that there are return and volatility spillovers among the pilots, with the dynamics of price in each pilot driven mainly by its own factors [23]. So far, there is still a big gap between the Chinese market and other developed carbon markets in various respects, including the carbon price formation mechanism, legal and policy systems, etc. [1].

From the above literature, it can be seen that although carbon emission markets in China have become an important research topic in academia, there are still some research gaps. First, the existing literature mainly focuses on the price interactions among China’s regional pilot markets as well as the spillover effects with other financial markets, while the common market force underlying the regional pilot markets is insufficiently explored, i.e., within a nation, can regional pilot markets behave in a way that they constitute a unified market? Fan and Todorova’s research is an exception. They investigated the similarity and heterogeneity of price dynamics across China’s regional markets using the visibility graph network approach. Their result shows that the pilot markets form four clusters, indicating inequalities across the sub-market groups [17]. Nevertheless, their result still cannot quantitatively measure the extent of common market trends among the regional pilot markets. Second, research on the European market has revealed that market efficiency improves with the development of the market, with better efficiency after phase I. In contrast, no such research has been performed to investigate the market efficiency of China in different periods. China’s regional pilot markets continued to operate after the launch of the national market. It is still unclear whether the launch of the national market would impact the market efficiency of the regional pilot markets or not.

This paper aims to fill the research gaps by assessing the market efficiency of China’s regional pilot markets from the perspective of a unified market and then the progress in market efficiency in different periods. Existing literature usually splits the research into two components in investigating the efficiency of carbon markets: the long-run trends and short-run dynamics. The cointegration test is often adopted to check the existence and the number of long-run trends while VECM is applied to explore the short-run price dynamics and interactions [7,8,10,24]. Following the pipeline, we design our research as follows. Using the updated data covering the period of the operation of national carbon market, we adopt Johansen cointegration tests to assess the long-run trends among the regional pilot markets of China and then apply VECM regression to investigate the carbon price dynamics. Considering the change in market structure following the launch of the national carbon emission trading market, we study the cointegration relationships and the price dynamics in different subsamples. It is found that China’s carbon market is experiencing a transition from a fragmented toward a unified market state. An improvement in market efficiency is also detected.

The rest of the paper is organized as follows. In Section 2, we introduce the methodology and the data used in this paper. Section 3 reports the estimation results. Section 4 discusses the findings. Lastly, Section 5 concludes the paper.

2. Methodology and Data

2.1. Methodology

The cointegration test examines the integration of two or more non-stationary time series in a situation that the time series cannot deviate from the equilibrium in the long-run. The Engel–Granger test and Johansen test are the two main approaches for cointegration tests. The Engel–Granger test is only suitable for two single variables. When more than two variables are involved, the Johansen test is more applicable. Since this paper deals with more than two variables, we adopt the Johansen test in line with the literature [7,8].

Let $y_t$ be the k-vector of carbon prices of regional pilot markets. The price series are often non-stationary I(1) variables. Consider a (p + 1)-order VAR

$$y_t=A_1{y}_{t-1}+\cdots +A_{p+1}{y}_{t-p-1}+e_t,$$

where $A_i$ is the k × k coefficient matrix. $e_t$ is a vector of white noises. The VAR can be transformed by first difference and rewriting it into the VECM, as shown below:

$$\Delta y_t=\Pi ·y_{t-1}+\sum _{i=1}^p{\Gamma }_i·\Delta y_{t-i}+e_t,$$

(1)

where $\Pi ={\sum }_{i=1}^{p+1}{A}_i-I$, ${\Gamma }_i=-{\sum }_{j=i+1}^{p+1}{A}_j$, I is the identity matrix.

According to Johansen (1991, 1995) [25,26], if $\Pi$ has reduced rank $r<k$, r would be the number of cointegration relationships. The cointegration relationship specifies the long-run trends the endogenous variables converge to. Meanwhile, there might exist short-run dynamics. In short-run dynamics, the price might deviate from the long-run trend. The deviation is called the error term or cointegration term. The error term could be corrected gradually according to short-run adjustment. The matrix $\Pi$ can be written in terms of the matrix of adjustment parameters $\alpha$ and the matrix of cointegrating vectors $\beta$, i.e., there exist $k\times r$ matrices $\alpha$ and $\beta$ such that $\Pi =\alpha {\beta }^{\prime }$ and ${\beta }^{\prime }y$ satisfy a property that ${\beta }^{\prime }y$ is stationary, i.e., I(0). ${\beta }^{\prime }y$ forms an r-vector of error terms. $\alpha$ captures the short-run adjustment.

Johansen [26] develops the likelihood ratio (LR) tests to determine the number of cointegration relationships based on the coefficient matrix $\Pi$. Two tests, the maximum eigenvalue test and the trace test, are involved in the Johansen tests.

For the maximum eigenvalue test, the null hypothesis is the number of cointegration relationships $r=r_0$, while the alternative hypothesis is $r=r_0+1$. The test statistic is

$$LR\left(r_0,r_0+1\right)=-Tln\left(1-{\lambda }_{r_0+1}\right),$$

where ${\lambda }_{r_0+1}$ is the $\left(r_0+1\right)$th largest eigenvalue.

For the trace test, the null hypothesis is the number of cointegration relationships $r=r_0$ while the alternative hypothesis is $r_0<r\le k$. The test statistic is

$$LR\left(r_0,k\right)=-T\sum _{i=r_0+1}^{k}ln\left(1-{\lambda }_i\right).$$

2.2. Data

This paper uses daily data to investigate the dynamics and interrelationships of carbon emission prices of different regional pilot markets in China. Currently, there are eight regional pilot carbon emission trading markets. Due to inactive trading and missing data, we exclude the Chongqing and Fujian pilots, following the practice of existing research [17]. We abstract the average transaction price of major exchangers (Beijing, Guangdong, Shenzhen, Shanghai, Hubei, and Tianjin) from Wind, a third-party database. The sample ranges from 5 January 2015 to 31 October 2022, with a total of 1964 observations.

All the data are transformed into the natural logarithm form. The first difference of the natural log is the return or the growth rate of carbon prices.

Table 1. Summary statistics.

	${\mathit{r}}^{\mathit{B}\mathit{e}\mathit{i}\mathit{j}\mathit{i}\mathit{n}\mathit{g}}$	${\mathit{r}}^{\mathit{G}\mathit{u}\mathit{a}\mathit{n}\mathit{g}\mathit{d}\mathit{o}\mathit{n}\mathit{g}}$	${\mathit{r}}^{\mathit{S}\mathit{h}\mathit{e}\mathit{n}\mathit{z}\mathit{h}\mathit{e}\mathit{n}}$
Average	0.000	0.000	0.000
Maximum	0.470	2.008	3.370
Minimum	−0.293	−2.397	−2.773
Standard dev.	0.069	0.176	0.309
Skewness	−0.448	−0.440	0.371
Kurtosis	8.104	43.602	25.697
Jarque-Bera	2198.840 ***	135,038.700 ***	42,224.140 ***
	$r^{Shanghai}$	$r^{Hubei}$	$r^{Tianjin}$
Average	0.000	0.000	0.000
Maximum	1.037	0.290	0.728
Minimum	−0.839	−0.282	−0.693
Standard dev.	0.057	0.039	0.042
Skewness	2.112	0.074	0.093
Kurtosis	91.351	9.562	143.343
Jarque-Bera	640,574.700 ***	3526.869 ***	1,612,635.000 ***

Notes: *** denotes significance at 1% levels, respectively.

reports the summary statistics for the return time series. The averages of all return time series are 0.000, indicating that there is no major monotonic increasing or decreasing trend over the sample. Please note that standard deviations are quite large, especially for Guangdong and Shenzhen, with values of 17.6% and 30.9%, respectively. In addition, all the return time series exhibit fat-tailed distributions with a kurtosis of much larger than 3.

Table 2. Pairwise correlation matrix.

	${\mathit{r}}^{\mathit{B}\mathit{e}\mathit{i}\mathit{j}\mathit{i}\mathit{n}\mathit{g}}$	${\mathit{r}}^{\mathit{G}\mathit{u}\mathit{a}\mathit{n}\mathit{g}\mathit{d}\mathit{o}\mathit{n}\mathit{g}}$	${\mathit{r}}^{\mathit{S}\mathit{h}\mathit{e}\mathit{n}\mathit{z}\mathit{h}\mathit{e}\mathit{n}}$	${\mathit{r}}^{\mathit{S}\mathit{h}\mathit{a}\mathit{n}\mathit{g}\mathit{h}\mathit{a}\mathit{i}}$	${\mathit{r}}^{\mathit{H}\mathit{u}\mathit{b}\mathit{e}\mathit{i}}$	${\mathit{r}}^{\mathit{T}\mathit{i}\mathit{a}\mathit{n}\mathit{j}\mathit{i}\mathit{n}}$
$r^{Beijing}$	1.000
$r^{Guangdong}$	0.023	1.000
$r^{Shenzhen}$	−0.006	0.012	1.000
$r^{Shanghai}$	0.035	−0.003	0.009	1.000
$r^{Hubei}$	0.006	0.038	0.000	−0.012	1.000
$r^{Tianjin}$	0.027	−0.014	0.007	−0.009	0.004	1.000

reports the correlation between the variables. Overall, the correlations are very small, with magnitudes of less than 0.04. The largest correlation is observed between Guangdong and Hubei with a value of 0.038. The small correlations between the variables indicate a low chance of multicollinearity in the regressions.

We first check the unit root properties of all the time series. It is found that all the price time series are I(1) processes except that $p^{Beijing}$ and $p^{Shenzhen}$ are I(0) processes. Considering that all the prices are in level, we still include all the price time series in the cointegration test to investigate the existence of trends among the prices (Indeed, in the second subsample analysis, we find that both $p^{Beijing}$ and $p^{Shenzhen}$ become I(1) processes following the launch of the national carbon market.). We conduct the cointegration test using the technique by Johansen [25,26], and the result is reported in

Table 3. Johansen cointegration test.

r	Trace Statistic	5% Critical Values	Max-Eigen Statistic	5% Critical Values
$r\le 0$	272.568 ***	103.847	132.662 ***	40.957
$r\le 1$	139.906 ***	76.973	79.529 ***	34.806
$r\le 2$	60.377 **	54.079	33.150 **	28.588
$r\le 3$	27.227	35.193	18.217	22.300
$r\le 4$	9.009	20.262	7.418	15.892
$r\le 5$	1.591	9.165	1.591	9.165

Tests of the existence of long-run relationships among the prices. r is the number of cointegration relationships. ** and *** denote significance at levels of 5% and 1%, respectively.

. Both the trace statistic and maximum eigenvalue statistic indicate that there are three long-run relationships among the prices. As shown in the following section, deviations from the individual long-run relationship are denoted by CointEq1, CointEq2, and CointEq3, respectively.

3. Results

In this section, we present the VECM estimation result. For the specification of the VECM model, information criterion SIC is used to determine the lag order; the optimal lag order, p, is found to be 1. As reported in

Table 4. VECM estimation result for the whole sample.

Cointegrating Equation
Variables	CointEq1	CointEq2	CointEq3
$p^{Beijing}$	1	0	0
$p^{Guangdong}$	0	1	0
$p^{Shenzhen}$	0	0	1
$p^{Shanghai}$	−0.218 **	−0.364 ***	0.177
	(0.095)	(0.047)	(0.150)
$p^{Hubei}$	−0.295	−0.620 ***	0.948 ***
	(0.181)	(0.089)	(0.286)
$p^{Tianjin}$	0.046	−0.601 ***	−0.586 ***
	(0.139)	(0.069)	(0.221)
c	−2.521	1.873	−5.044
Error Correction
	$r^{Beijing}$	$r^{Guangdong}$	$r^{Shenzhen}$	$r^{Shanghai}$	$r^{Hubei}$	$r^{Tianjin}$
CointEq1${}_{t-1}$	−0.035 ***	0.012	0.048 **	0.012 **	0.005	0.000
	(0.006)	(0.014)	(0.025)	(0.005)	(0.003)	(0.004)
CointEq2${}_{t-1}$	0.001	−0.165 ***	0.003	0.009	0.007 *	0.008 *
	(0.006)	(0.015)	(0.026)	(0.005)	(0.004)	(0.004)
CointEq3${}_{t-1}$	0.004	−0.016 **	−0.101 ***	−0.004	-0.002	0.001
	(0.003)	(0.007)	(0.012)	(0.003)	(0.002)	(0.002)
$r_{t-1}^{Beijing}$	−0.005	−0.073	0.055	−0.033 *	0.000	0.008
	(0.023)	(0.053)	(0.090)	(0.019)	(0.013)	(0.014)
$r_{t-1}^{Guangdong}$	−0.012	−0.301 ***	−0.033	−0.005	−0.005	−0.001
	(0.009)	(0.022)	(0.037)	(0.008)	(0.005)	(0.006)
$r_{t-1}^{Shenzhen}$	−0.001	0.002	−0.396 ***	0.004	0.008 ***	0.002
	(0.005)	(0.012)	(0.021)	(0.004)	(0.003)	(0.003)
$r_{t-1}^{Shanghai}$	0.007	0.117 *	0.023	−0.051 **	0.003	−0.001
	(0.027)	(0.062)	(0.108)	(0.022)	(0.015)	(0.017)
$r_{t-1}^{Hubei}$	−0.019	−0.162 *	0.015	0.008	−0.244 ***	0.012
	(0.040)	(0.092)	(0.158)	(0.033)	(0.022)	(0.024)
$r_{t-1}^{Tianjin}$	−0.020	0.022	0.003	0.219 ***	−0.017	−0.124 ***
	(0.037)	(0.085)	(0.146)	(0.030)	(0.020)	(0.022)
c	0.000	0.001	0.000	0.000	0.000	0.000
	(0.002)	(0.004)	(0.006)	(0.001)	(0.001)	(0.001)
Log likelihood	13,090
AIC	−13.251
SIC	−13.029

Notes: standard error in parentheses, *, **, and *** denote significance at 10%, 5%, and 1% levels, respectively.

, there are three cointegration relationships among the carbon prices, indicating the coexistence of three trends among the carbon emission prices in different regions of China. Error-correction is observed among all the prices of different exchanges. $r^{Beijing}$, $r^{Guangdong}$, and $r^{Shenzhen}$ are subject to significant and negative effects from their corresponding error terms, CointEq1, CointEq2, and CointEq3, respectively. Furthermore, the lagged CointEq1 imposes a positive effect on $r^{Shanghai}$. The lagged CointEq2 also imposes a positive effect on both $r^{Hubei}$ and $r^{Tianjin}$. These effects eventually reduce the price deviations of the dependent variables from the long-run trend. The error-correction effects indicate that carbon emission markets of individual regional exchanges exhibit a certain level of efficiency. In addition, there are interactions arising from error terms of carbon price returns that are not from the same cointegration relationship. The lagged CointEq1 has a positive effect on $r^{Shenzhen}$. If the carbon price of the Beijing market were to positively deviate from its long-run trend, the price of the Shenzhen market would be subject to the positive spillover effect and subsequently increase. Negative spillover also arises with the lagged CointEq2 depressing $r^{Guangdong}$. For example, if the carbon price of Shenzhen positively deviates from its long-run trend, the price of the Guangdong market in the next transaction period will be subject to a depression effect.

Regarding the price dynamics, price returns of all markets face significant and negative effects from their own lagged variables except for Beijing, for which the effect is insignificant. Underlying these negative effects is the mean-reversion pattern in the price time series. An increase in prices tends to be followed by a negative one and vice versa; therefore, continual monotonic trends in prices are rare. This mean-reversion pattern is strongest in the Shenzhen market, followed by the Guangdong market.

A spillover effect is observed between the price returns. The price return on the Beijing markets imposes a negative effect on the Shanghai market, while the Shanghai price has a positive spillover effect on that of Guangdong. Overall, only the Guangdong price does not impose a spillover effect on prices of other markets while itself it receives two, which is the highest number among the markets, spillover effects from other markets. The Shanghai price also receives two spillover effects. The price of the Beijing market is not subject to any spillover effects from other markets. The lead–lag relationships between the prices of various markets suggest that both the Guangdong and Shanghai markets are sensitive to other regional markets.

Stages of Market Development

On 16 July 2021, the national carbon emission trading market of China launched online trading, while the regional markets continued to trade. A nationwide carbon emission market started to form, which should impact the price dynamics and interactions of the regional markets. Therefore, the price dynamics might not be the same before and after the launch of the national carbon trading market. In view of potential change in the market state, we divide the sample into two subsamples with respect to 16 July 2021. The estimation result based on the first subsample is reported in

Table 5. VECM estimation result for the first subsample.

Cointegrating Equation
Variables	CointEq1	CointEq2	CointEq3
$p^{Beijing}$	1	0	0
$p^{Guangdong}$	0	1	0
$p^{Shenzhen}$	0	0	1
$p^{Shanghai}$	−0.231 **	−0.298 ***	0.324 ***
	(0.092)	(0.037)	(0.098)
$p^{Hubei}$	−0.384 **	−0.415 ***	0.986 ***
	(0.184)	(0.075)	(0.196)
$p^{Tianjin}$	0.077	−0.497 ***	−0.298 **
	(0.137)	(0.056)	(0.147)
c	−2.300	0.766	−6.438
Error Correction
	$r^{Beijing}$	$r^{Guangdong}$	$r^{Shenzhen}$	$r^{Shanghai}$	$r^{Hubei}$	$r^{Tianjin}$
CointEq1${}_{t-1}$	−0.031 ***	0.032 **	0.019	0.016 ***	0.003	-0.001
	(0.007)	(0.016)	(0.025)	(0.006)	(0.004)	(0.005)
CointEq2${}_{t-1}$	−0.003	−0.312 ***	−0.076 **	0.000	0.004	0.011 *
	(0.009)	(0.021)	(0.033)	(0.008)	(0.005)	(0.006)
CointEq3${}_{t-1}$	−0.002	−0.070 ***	−0.148 ***	−0.007 *	−0.003	0.000
	(0.004)	(0.011)	(0.016)	(0.004)	(0.003)	(0.003)
$r_{t-1}^{Beijing}$	−0.022	−0.114 *	0.088	−0.048 **	−0.002	0.011
	(0.025)	(0.062)	(0.096)	(0.023)	(0.015)	(0.017)
$r_{t-1}^{Guangdong}$	−0.012	−0.224 ***	0.007	0.000	−0.004	−0.003
	(0.010)	(0.024)	(0.037)	(0.009)	(0.006)	(0.007)
$r_{t-1}^{Shenzhen}$	0.000	0.026 *	−0.386 ***	0.005	0.008 **	0.002
	(0.006)	(0.015)	(0.023)	(0.006)	(0.004)	(0.004)
$r_{t-1}^{Shanghai}$	0.013	0.080	0.017	−0.047 *	0.006	−0.002
	(0.026)	(0.065)	(0.101)	(0.025)	(0.016)	(0.018)
$r_{t-1}^{Hubei}$	−0.003	−0.153	0.097	0.006	−0.267 ***	0.015
	(0.040)	(0.100)	(0.153)	(0.037)	(0.024)	(0.028)
$r_{t-1}^{Tianjin}$	−0.023	−0.014	−0.018	0.220 ***	−0.023	−0.125 ***
	(0.036)	(0.088)	(0.136)	(0.033)	(0.021)	(0.025)
c	0.000	0.000	−0.001	0.000	0.000	0.000
	(0.002)	(0.004)	(0.006)	(0.002)	(0.001)	(0.001)
Log likelihood	10,629
AIC	−13.018
SIC	−12.759

Notes: standard error in parentheses, *, **, and *** denote significance at 10%, 5%, and 1% levels, respectively.

. Overall, the result is consistent with the one based on the whole sample. There are three cointegration relationships and the coefficients for the individual cointegration relationships are similar. In terms of the effects of error terms, the results are comparable. Regarding interactions among the price returns, the result is consistent, except that the effect of $r_{t-1}^{Beijing}$ on $r^{Guangdong}$ is significant while the effects of $r_{t-1}^{Shanghai}$ and $r_{t-1}^{Hubei}$ on $r^{Guangdong}$ are insignificant in the first subsample although the signs of the three coefficients are the same.

Table 6. VECM estimation result for the second subsample.

Cointegrating Equation
Variables	CointEq1
$p^{Beijing}$	1
$p^{Guangdong}$	0.860 ***
	(0.214)
$p^{Shenzhen}$	−0.244 ***
	(0.053)
$p^{Shanghai}$	−0.444
	(0.325)
$p^{Hubei}$	−2.146 ***
	(0.365)
$p^{Tianjin}$	0.029
	(0.394)
c	2.634
Error Correction
	$r^{Beijing}$	$r^{Guangdong}$	$r^{Shenzhen}$	$r^{Shanghai}$	$r^{Hubei}$	$r^{Tianjin}$
CointEq1${}_{t-1}$	−0.039 **	−0.040	0.250 ***	−0.006	0.037 ***	−0.005
	(0.020)	(0.025)	(0.087)	(0.005)	(0.007)	(0.004)
$r_{t-1}^{Beijing}$	0.052	−0.024	−0.214	0.019	−0.020	0.005
	(0.056)	(0.071)	(0.246)	(0.015)	(0.019)	(0.012)
$r_{t-1}^{Guangdong}$	0.023	−0.441 ***	−0.161	−0.004	−0.015	0.003
	(0.039)	(0.050)	(0.171)	(0.010)	(0.013)	(0.008)
$r_{t-1}^{Shenzhen}$	0.000	0.001	−0.382 ***	−0.001	0.015 ***	0.002
	(0.011)	(0.015)	(0.050)	(0.003)	(0.004)	(0.002)
$r_{t-1}^{Shanghai}$	−0.214	0.228	0.246	−0.287 ***	−0.079	0.070 *
	(0.199)	(0.252)	(0.872)	(0.053)	(0.067)	(0.042)
$r_{t-1}^{Hubei}$	−0.165	0.043	−0.682	0.037	−0.020	−0.019
	(0.153)	(0.194)	(0.672)	(0.040)	(0.051)	(0.032)
$r_{t-1}^{Tianjin}$	−0.071	−0.138	0.011	0.007	0.200 **	−0.064
	(0.260)	(0.330)	(1.142)	(0.069)	(0.088)	(0.055)
c	0.002	0.002	0.009	0.001	0.001	0.001
	(0.004)	(0.006)	(0.019)	(0.001)	(0.001)	(0.001)
Log likelihood	3063
AIC	−17.549
SIC	−16.945

Notes: standard error in parentheses, *, **, and *** denote significance at 10%, 5%, and 1% levels, respectively.

reports the estimation result based on the second subsample, the one after the national carbon emission trading market was launched. It is noticed that the number of cointegration relationships changes from three to one. Regarding the response of the carbon price of individual markets to their price deviations from the long-run trend, all the individual prices tend to correct its deviation in the previous period. In terms of price dynamics, price returns of individual markets still face negative effects from their own lagged variables. Note that the effects of the lagged return in $r^{Hubei}$ and $r^{Tianjin}$ are insignificant, indicating that information is better processed in the two regional markets so that historical transaction prices do not impact market transactions. Market interactions become less apparent in the second subsample. Only prices of Hubei and Tianjin receive positive spillover effects from other regional prices. $r^{Hubei}$ is subject to the positive spillover from price returns of Shenzhen and Tianjin while the price return of Tianjin, $r^{Tianjin}$, in turn, is subject to the spillover effect of the price return of Shanghai. Overall, the interactions among the regional carbon market reduced after the national carbon market was launched.

4. Discussion

Before the launch of national carbon market, there were three long-run trends among the regional pilot markets, indicating a certain degree of fragmentation in China’s carbon market, i.e., the Chinese carbon market was characterized by three regional market combinations. No unified market was formed within the whole market. This finding is consistent with the fact that these regional markets form fragmented clusters as the management of these markets is handled by the local authorities [23]. Moreover, five out of six regional pilot markets witnessed that their prices were subject to the impact of historical price movements, exhibiting a mean-reversion pattern. Our result for the impact of historical prices is consistent with the finding of Wen et al. [20]. The mean-reversion pattern indicates that market prices either overreact or cannot incorporate market information into prices in a timely manner. The regional pilot markets were unable to meet the requirements of weak-form efficiency, which requires that the historical price, trading volume, and profit data should not affect the asset price or predict the future trend of the asset.

After the launch of the national carbon market with the regional pilot markets continuing to operate, the number of long-run trends among the regional pilot markets reduced from three to one. A single common trend means that the price of individual regional pilot markets takes into account the rest of the regional pilot markets, or the overall situation of China’s carbon market, instead of the regional market alone. In addition, each individual market displays its capability of self-correction to reduce its price deviation from the long-run trend, signaling a certain market efficiency in pricing. Another improvement in market efficiency comes from the impact of historical prices. The number of regional pilot markets impacted by historical prices decreased from five to three following the launch of the national carbon market. Wang et al. investigated the performance of China’s carbon market without considering the different development stages of the overall market of China [19]. Compared to their conclusion of poor market efficiency for China’s carbon market, our results confirm that the efficiency of China’s carbon market is not high enough, but we further find that the efficiency of China’s carbon market is improving.

The existing literature has discovered that the efficiency of the European carbon market improved with market development [6]. Our finding based on the regional pilot markets indicates that China’s carbon market is undergoing a similar course of efficiency improvement. In this course, the setup of a national carbon market is an important step as it has promoted the development of China’s carbon market from regional segmentation toward a unified market.

5. Conclusions

The carbon emission trading market is an important institutional innovation, which uses market mechanisms to control and reduce greenhouse gas emissions and promote the green and low-carbon transformation of economic development. In accordance with the principle of “gradually establishing a carbon emission trading market”, regional pilot carbon markets in China have been launched for trading since 2013. The national carbon emission trading market started online trading in 2021, while the existing regional markets continued to trade. This paper examines the pricing efficiency and price dynamics of the regional pilot markets by taking into account market development in terms of the setup of the national carbon market.

Before the national carbon emission trading market was launched, the regional markets formed three long-run trends. In contrast, after the national carbon emission trading market was launched, the number of long-run trends reduced to one. The single long-run trend means that all of the regional markets stick to one common trend. From the perspective of market operation, all of the regional markets are connected through a common trend; thus, a unified national market has begun to take shape. In terms of market efficiency, the regional markets tend to reduce their corresponding price deviations in the lagged period so that the regional market prices converge to the long-run trend. The ability to self-correct implies a certain market efficiency of the regional markets.

Regarding price dynamics of the regional markets, several regional markets still exhibit mean-reversion price patterns after the launch of the national market. Please note that the number of regional markets exhibiting mean-reversion decreases after the launch of the national market signals an improvement in the market operation.

From the results, we can see that the setup of China’s national carbon market has promoted the development of China’s carbon market from segmentation toward a unified market, in addition to the improvement in terms of historical price impact. We are able to offer some policy implications. China should further develop its national carbon emission market. The existing regional pilot markets should also be gradually incorporated into the national market to increase the market size. At present, the trading entities of the national market are limited to emission control enterprises while the qualification of institutional and individual investors has not been specified for participation in the market, resulting in relatively insufficient liquidity in the market. Since liquidity is important to market efficiency, institutional and individual investors should be encouraged to participate in market transactions to improve market liquidity. With the further development of the national carbon market, market efficiency should be further improved in the short term. As the setup of the carbon market is to promote green and low-carbon transformation, this transformation should be better achieved in the long term.

Carbon emissions are directly related to energy use and energy structure. Therefore, carbon price is correlated with various factors, such as the level of economic development, technological development, energy consumption, and environmental protection [20]. Future research should explore the determinants of carbon prices, such as government policies and industrial sector development. Moreover, the carbon market is essentially a financial market. Its interaction with other financial markets could be another research direction.