-Based on China’s pilot emissions trading scheme

(1)

An empirical analysis of energy prices impact on carbon allowance prices in China

-Based on China’s pilot emissions trading scheme

09-07-2015

Study program: MSc International Financial Management Supervisor: Prof. Bert Scholtens

Student name: Weijia Li Student number: s2659239

(2)

Abstract

As fostering a low-carbon economy has become a globally common sense upon raised environmental concerns, China launched its pilot emissions trading scheme ¹ to better implement national carbon trading strategy in late October 2011. The aim of this is to reach the goal of 40-45% carbon intensity reductions by 2020². The paper investigates the relationship between carbon allowance prices and energy prices in China during the past year.

Empirical data is tested through a Vector Autoregressive Model (VAR). The results show that carbon allowance prices are significantly correlated to energy prices. The findings suggest changes in natural gas prices contribute the largest impact on the fluctuation of carbon allowance prices in the short run. However, in the long run, changes in energy prices have no statistically significant impact on the carbon allowance prices volatility.

Key words: Emissions trading scheme, Energy prices, Carbon allowance prices, Market fundamentals, ADF unit root test, Correlation, VAR model.

1 In October 2011, National Development and Reform Commission(NDRC) established a Notice on launching seven carbon emissions trading pilots in China.

2In November 2009, during the United Nations Framework Convention on Climate Change (UNFCCC)’ 15^thconference, the State Coucil published the carbon targets by 2020 to reduce CO2 per unit of GDP by 40% to 45% compared to 2005.

(4)

1. Introduction

China has become the largest CO2 emitter since 2006 due to its coal-dominated energy structure, booming exports and rapid economic growth, addressed by Lin et al. (2010). They point out CO2 emissions in China will continuously grow due to the main force of energy- related consumption and production. China is facing severe pressures in reductions of carbon intensity and energy intensity nationwide. With regard to factors which can possibly strengthen China’s low-carbon development, the ambitious demand for energy efficiency fulfillment and for market structure improvement is considered significantly important (Zhu et al., 2013). There is increased scholarly focusing on this area. For instance, Wang et al. (2014) demonstrate carbon pricing is helpful for CO2 reductions by adjusting electricity-energy prices. Zhang (2014b) also mentions that improving energy efficiency is considered the fastest and most effective way to emphasize environmental concerns.

Compared to China’s carbon trading market, the European Union (EU) emission trading scheme (ETS) possesses a relatively mature trading system. European Union emission allowance (EUA) is defined as the primary intermediary complying with certificated emissions reductions (CERs) and other secondary derivatives. Together with EU carbon market, China is also participating in international clean development mechanism (CDM) programs³. To achieve a bold national strategy and to gain international competitiveness, China launched its own pilot emission trading scheme, including seven pilots in the process (NDRC, 2011; Zhang, 2015). However, carbon emission trading (CET) is still in progress with some apparent bottlenecks. For example, bottlenecks are shown as fragmented trading markets, lack of motivated participants, immature pricing mechanism and weak competitiveness in international carbon trading markets (Liu et al., 2014; Zhou et al, 2011;

Zhang et al., 2010b). Therefore, research on carbon pricing could probably promote cost-

3In the EU ETS, EUA is regarded as the official unit traded that each allowance permits to emit one metric ton of CO2. Clean Development Mechanism (CDM) is developed by Emissions Trading Directive as the adoption of a provision linked with Joint Implementation (JI). CDM is to recognize verified project-based reductions in greenhouse gas (GHG) emissions in the developing countries. The Linking Directive confirms the Kyoto Protocol’s CDM as certified emissions reductions (CERs).

(5)

effective abatement regulator, deliver innovation incentives and potential participants incentives, as has been stated by Joseph et al. (2012).

There are existing discussions about what is driving the carbon prices (see, e.g., Christiansen et al., 2005; Benz and Trück, 2009; Daskalakis et al., 2009). Seifert et al. (2008) demonstrate that the carbon prices show a time-and price-dependent volatility structure, and market structure and institutional policies are regarded as main drivers for shaping the prices, as also discussed by Chevallier (2011a). Traditional energy-related prices, such as coal prices, oil prices and natural gas prices are highlighted as the three main drives of the carbon prices (Chevallier 2011b). Recent research papers show that energy-intensive entities are the largest producers of carbon dioxide emission, and thus majority of primary Carbon Emission Allowances are allocated into energy sectors and power sectors among all industries in global carbon markets. Christiansen et al. (2005) see it crucial to monitor development in fossil-fuel prices and to emphasize fuel-switching measurements, as Hintermann (2010) examines, further energy markets shocks affect carbon futures and the role of fuel prices should be highlighted besides with macroeconomic shocks. Among these are Mansanet-Bataller et al.

(2011), Hintermann (2010), andFalbo et al. (2012), Dalarue et al. (2010) have investigated that the carbon allowance prices are using the fuel-switching levels⁴.

The main contribution of this paper is to investigate the impact of energy prices on the carbon allowance prices in China, to seek for potential solutions in energy-efficiency promotion and carbon intensity reductions. Although previous literatures have shown a correlation between carbon prices and energy prices, the literatures mostly draws upon EU ETS, rather than the Chinese market. As the economic growth in China relies on the energy- intensive industries, it is a necessity for China to figure out how to promote energy efficiency and to achieve low carbon economy targets (Zhang, 2014b). This paper is with practical

4According to Christiansen et al. (2005), fuel switching from coal to gas in power and in heating sector is the single most important abatement measure in the short run.

BunnandFezzi(2008) say that the switiching from coal to gas allows a producer to reduce carbon emissions per MWh by 40% to 60%. Relevance of swithcing prices can be found in studies like Rickels et al., (2010); Delarue et al., (2008) and Denny and O’Malley (2009).

(6)

meaning for better implementing national and international strategies, forecasting models and formulating policy suggestions.

In this paper, Vector Autoregressive Model (VAR) modeling is conducted upon the analysis of the trading performance in seven trading pilots, where daily-trading prices will be averaged in time series. A correlation test will be examined first to gain an overall view of the relationship between carbon prices and energy prices in China’s trading market.

The reminder of the paper is as follows. The next section includes the literature overview and hypothesis development. Then, we will describe the overall design and trading performance in China’s seven pilot emissions trading schemes. The data, methodology and modeling will be described in Section 4. Next, we show the results and interpretations of the estimations in the model. The last section gives the conclusions and remarks of our research.

2. Literature overview

To conduct the research on carbon pricing, emissions trading scheme (ETS) should be investigated. In the EU’s carbon market, two broad policy instruments are introduced to encourage innovation for meeting reduction targets, respectively as carbon energy tax and emissions trading, according to Ellerman et al. (2010). They point out that the market-based instrument “emissions trading” enables the EU members to involve in a multinational-level trading scheme, which originally comes from United Stated and depends on a “learning by doing” basis. Regarding to the EU ETS experience, market-based instruments are with certain efficiency in conducting climate policies (see, e.g., Ellerman et al., (2007); Joseph et al., 2012). Joseph et al. (2012) demonstrate that an effective climate change policy would change the decision calculus for activities to promote efficient generation and use of energy, low carbon-intensity of energy and more carbon-lean economy, alternative design of instruments like carbon taxes, cap and trade and clean energy standards are matched also in the EU ETS.

With respect to China market, Zhang (2015) shows China relies mostly on administrative measures until market-orientated instruments have been introduced as main drivers of carbon trading market. In other literatures, main market instruments in China as fundamental are stated as environmental taxes, and emissions trading scheme, implemented by carbon linkage and compliance regulations, as addressed by Zhang (2014b, 2015). Market-based instruments

(7)

applied in climate change maps. Ellerman et al. (2007) discuss, “Action on market based instruments on tradable emissions permits fosters an informed and supportive culture”.

Considering previous research mentioned above on the importance of market-based instruments implications, especially the influence of emissions trading schemes, we believe that China’s ETS has given a priority for further development in carbon trading market.

When looking into the trading schemes, pricing mechanisms always stand out because of the updated market information, which are crucial in the decision making progress for potential participants and investors. Subsequently there are increasing researches drawing upon the relationship between carbon pricing and energy sectors (see, Section 1). Mansanet- Bataller et al. (2007) and Alberola et al. (2008) are the first to investigate the relations between energy markets and the carbon dioxide prices in the EU carbon trading market, which have also been mentioned by Chevallier (2011a), early work shown by Ellerman et al.

(2010). According to Christiansen et al. (2005), carbon prices are determined by market fundamentals such as policies, regulations, technical indicators and production levels.

Chevallier (2011b) discusses that carbon prices are influenced by institutional decisions, also by energy prices and weather conditions. However, there are few literatures in China about carbon pricing and the relevance of energy sectors. Zhen et al. (2011) state that there is no obvious phenomenon of nonlinear dynamics of the carbon prices fluctuation; carbon market self-construction and heterogeneous environment should be counted when pricing carbon emissions. Another literature in China shows that the carbon prices may also be affected by allowance allocation options and by marginal abatement costs difference among different sectors, e.g. energy sector (Cong et al., 2010).

Creti et al. (2012) state that energy variables are the most natural determinants of EUA prices and they denote a cointegrating relationship between carbon prices and energy fundamentals. Additionally, the relationship between energy prices and carbon prices are examined. Aatola et al. (2013) stress that there is a strong relationship among electricity prices, energy prices and the prices of EUA, and EUA forward prices depend on the market essentials, especially on the price of electricity as well as on the gas–coal difference, in a statistically significant way. Furthermore, Chevallier (2011b) evaluates the interaction between macroeconomic factors and energy spheres through Markov-switching VAR model.

Moreover, a nonlinear autoregressive distributed lags (NARDL) model is developed by Hammoudeh et al. (2014a,b) from the United States’ perspective that they emphasize oil

(8)

prices have a long-run negative impact on carbon prices, while falls on coal prices have stronger impact on carbon prices in the short run. Hammoudehet al. (2014c) also conduct a quantile regression method in United State, they demonstrate that an increase in the crude oil price generates a substantial drop in the carbon prices when the latter is very high, changes in the natural gas prices have an negative effect on the carbon prices when they are every low but have a positive effect when they are quite high, and the coal prices exert an negative effect on the carbon prices.

Empirical evidence has shown that there is a significant relationship between carbon prices and energy prices, while most of the literatures are drawing upon EU market or United States market (see, e.g., Christiansen et al., 20015; Delarue et al., 2010; Creti et al., 2012;

Hammoudeh et al. 2014c). As China has its own environment and market structure, the research will investigate how the energy prices impact on carbon prices in China, to see whether there is a strong relationship, and to formulate possible solutions for improving energy efficiency and pricing mechanism under China’s ETS. Based on previous studies on the EU ETS and United States’ carbon market, we develop our hypothesis on the relationship between carbon prices and energy prices in China. The hypothesis below will be tested and interpreted in details in Section 5:

H0: There is no significant relationship between carbon prices and energy prices.

H1: There is a significant relationship between carbon prices and energy prices.

Further, if changes in energy prices have impact on carbon price volatility in China market, then the hypothesis H1 is specified as (interpretations will be discussed separately in long run and in short run in Section 5):

H1a: Changes in energy prices have no impact on the carbon prices volatility.

H1b: Changes in energy prices have impact on the carbon prices volatility.

3. Pilot emissions trading schemes

3.1 History to update of China ETS

In late October 2011, National Development and Reform Commission (NRDC)

(9)

scheduled to start trading between June 2013 and June 2014 (NDRC, 2011). National ETS is expected to commence in 2016. The seven pilots include two provinces (Guangdong and Hubei) and five cities (Beijing, Shanghai, Shenzhen, Tianjin and Chongqing). The cities included in the pilots system are under the National Central Authority and have the same level of independence as provinces. Since Hubei and Chongqing launched their first trading on 2^nd April 2014 and on 19^th June 2014, seven pilots have been officially begun to operate in whole (NDRC, 2014).

According to Zhang (2010 and 2011a,b), administrative measures are mostly used to achieve low-carbon economy targets until China decides to embrace market-based instruments since November 2013. Therefore, China has limited experience in applied instruments, where the energy-saving rate⁵ is assumed to be much higher than in any other countries (IEA, 2009). Together with China’s ETS, several emissions trading schemes around the world are established, e.g., the EU ETS, Regional Greenhouse Gases Initiatives (RGGI) in the United States, and Australia Carbon Pricing Mechanism (AUS CPM). Currently, quite a lot research has been conducted on the overall design and annual performance in China’s ETS.

Fundamental issues such as allowance allocation, price mechanism as well as state-owned key enterprises are discussed by Qi et al. (2013). In addition, trading principles and design options have been investigated among others as Jotzo (2013), Yang (2014) and Zhang et al. (2010b, 2012, 2014a, 2014b, 2015). According to China carbon market report 2014 (Zhong, 2014), ETS can encourage industrial entities to reduce CO2 emissions and also can emphasize cost shares to participants. Carbon pricing mechanism and trading structure in China’s ETS has also been discussed by Zhao (2014).

To foster a national-level carbon trading market, during the period 2014 to March 2015, several steps have been processed in seven trading pilots (IETA, NDRC, 2014, 2015):

5In May 2006, NDRC issued energy-saving responsibility agreements in 31 provinces in China to allocate overall energy-savingrate of top 1,000 enterprises, the program has been enlarged to 16,078 enterprises in December 2011. The program acheived total energy-savings of 170 milliontce over 2011 to 2012, which has completed 69% of the totalenergy-saving target in 12^th five-year plan period (NDRC, 2013). China pledged to have alternative energy sources to 15% of the national energy requirments by 2020 (Zhang, 2014a).

(10)

! April and June 2014. Hubei and Chongqing launched their own ETS.

! June 2014. First compliance investigation has employed in the former five trading pilots, trading volumes reached peak during June to August. 1.4% of all participants failed in first-year compliance.

!

August 2014. Guangdong Provincial Development and Reform Commission (GPDRC) released that mandatory auctioning for allowance trading has been cancelled, the floor price for auction has declined.

!

August 2014. Shenzhen pilot started to open for international investors through currency exchange regulators.

!

September 2014. First offset solution has been launched in Beijing pilot, according to Beijing Municipal Development and Reform Commission (BMDRC), including energy- saving programs and Forestry Carbon Sequestration.

!

November 2014.Joint Sino-US statement on climate change, carbon emission peak is expected in 2030.

!

November 2014. NDRC launched first 10 China Certificated Emission Reductions (CCERs) programs, which would enter into seven pilots market in 2015.

!

10^thDecember 2014.Provisional measures for the administration of carbon emission trading rights have been released by NDRC.

!

January 2015. National registry for voluntary emission trading has been announced.

!

February 2015. NDRC released a notice “Regarding the Fundamental conditions and operational thinking” to promote national market plan.

3.2 Structural framework and overall design

According to NDRC and Zhang (2014 a, b, 2015), the seven pilots have different characteristics in several aspects. Basically, overall design and guidelines for the seven pilots have been set up by NDRC. Variety is allowed in features such as: emission caps, coverage of sectors, allowance allocation methodology, monitoring, reporting and verification (MRV)

(11)

system, China Certificated Emission Reductions (CCERs) usage degree and others. Details are shown below in Table 1. Allowances issued are based on 2013-2014 level (IETA 2015).

Table 1. Summary of key features in seven pilots

Beijing Shenzhen Shanghai Guangdong Tianjin Hubei Chongqing Allowance

Issued

Not released

100 Mt CO2 in total for 2013 to 2015

150 Mt CO2

for 2013

388 Mt CO2

a year for 2014

Not released

324 Mt CO2 for 2014

Around 125 Mt CO₂, 4.13% of annual reduction Cap

Coverage

40% of city total emissions, 543 entities from power generation and others

38% of city total emissions, 832 entities from 26 sectors, participatio n open to financial institutions

57% of city total emissions, 190 entities listed

55% of province total energy consumptio n, 211 entities listed

60% of city total emissions, 114 entities listed

35% of province total carbon emissions, 138 entities listed

40% of total emissions, 242 entities from 6 sectors

Allocation Method

Free allocation, grandfathe ring for all covered sectors and benchmark ing for new entrants

Free allocation, benchmark ing for all covered sectors

Free allocation, grandfatheri ng and benchmarki ng

97% of free allocation in 2013, grandfatheri ng for non- electricity sectors, benchmarki ng for electricity sectors

Free allocation, grandfather ing for all covered sectors and benchmark ing for new entrants

Free allocation of 92%

before May, 8%

of annual allocation, grandfather ing

Free allocation, grandfather ing

Auction Small proportion s of allowance

Compleme ntary method for complianc e

Complemen tary method for

compliance

Complemen tary method for

compliance

Small proportions of

allowance

3% of reserve can be

auctioned No

Offsets 5% of annual allocations with CCERs, 50% of offsets generated within jurisdiction of the city

10% of annual allocation with CCERs

5% of annual allocations with CCERs

10% of annual allocations with CCERs, 70% from projects within the province

10% of annual allocations with CCERs

10% of annual allocations with CCERs, 100% from Hubei province

8% of annual allocations with CCERs, 100% from Chongqing

Sources: IETA 2015, NDRC 2014.

(12)

Besides with the key features difference shown in Table 1, governance and enforcement vary among pilots as well. Based on the regulations and policies published by NDRC, each pilot has its own governmental institutions to establish regional market information and trading notices. The governance systems help to arrange daily operations on each pilot. The governance framework of China’s ETS is stated as Table 2:

Table 2. Legal framework on pilot level

Pilots Highest order of legislature

Beijing Local Congress

Shenzhen Local Congress

Shanghai Shanghai Municipal Government (SMG)

Guangdong Guangdong Municipal Government

Tianjin Tianjin Municipal Government (TMG)

Hubei Hubei Provincial Government (HPG)

Chongqing Chongqing Municipal Government

Source: Zhang (2011a, b, 2015)

Furthermore, the method of allowance allocation has been regarded as a crucial factor influencing the carbon price and the market orientation as well. Grandfathering and benchmarking are the most common modes in China’s ETS. Qi et al. (2013) demonstrate that grandfathering allocates allowances based on a firm’s historic emission data while benchmarking encourages those participants with better performance. The difference is shown in Table 3 below:

Table 3. Difference among modes of allowance allocation

Grandfathering Benchmarking Hybrid mode

Encourage emission

reduction No Yes Partly

Consider financing

difficulties Yes No Partly

Consistency for new

entrants No No No

Requirements of data Simple Complex Complex

Complexity Simple Medium Complex

Source: Qi et al., (2013).

(13)

In a latest study on China’s national allocation plans (NAPs⁶), central government sets an overall emission cap for each scheme, and if the accumulated number of permits proposed by the provinces exceeds the national cap, it will be up to Beijing to make cuts in the provincial plans. Additionally, provinces will be ranked and sorted in groups according to the mitigation capacity in a bid to minimize the possibility of over-allocation. The level of flexibility for local governments in each pilot to make adjustments on the allocation plans will depend on the group they were assigned to.

Considering that items such as price uncertainty⁷, usage of offsets, coverage of sectors can contribute to different trading modes in seven pilots, further information should be expressed and deepened to pilot scale. However, due to previous practices in China’s ETS, monitoring system, pricing mechanism and enforcement process are still in progress, and international competitiveness is relatively weak in CDM programs, currently China is focusing on the carbon allowance trading and hope to build up its own national trading scheme. Therefore, allowance features are mainly discussed in this paper.

3.3 Carbon trading and performance during 2014 to March 2015

Since Hubei launched its first trading on 2^nd April 2014, Hubei has become the pilot with most outstanding trading volumes and turnover, compared to other six pilots. According to the annual carbon market in China, discussed by Zhong (2014), till October 2014, the seven pilots have realized 21 million tons CO2 trading volumes with 1 billion RMB trading amounts.

Among those, Hubei achieved5.28 million tons of trading volumes, with 125 million total trading amounts. This paper selects carbon allowance trading prices in six pilots, except for Chongqing pilot, in which the prices remain at 30.74 RMB per ton since the launch day, detailed information will be discussed in Section 4.

Figure 1 below shows overall price performance in the six trading pilots during the past year, the axes indicates the absolute trading prices of carbon allowances while the horizontal axis represents time duration from 2^nd April 2014 to 31^th March 2015. Additionally, detailed

6Each member in the EU ETS should have its own national allocation plan which determines the allocation of total EUAs nationalwide and the distribution of total subject to the commission’s approval and review (Ellerman et al., 2007).

7Price unvertainty is regarded as a major source of carbon price risk, according to Chevallier(2011a).

(14)

trading volumes and price fluctuations are expressed in Figure 2. Each pilot has two specific figures respectively shown as trading volumes and price trends. Real volumes are investigated in each pilot while price trends are shown as fluctuating percentage. Data in Figure 1 and Figure 2 is based on the observations of carbon allowance prices; observations are presented in Section 4. The data is also used for modeling VAR in latter sections.

Figure 1.Price performance in six carbon trading pilots

Source: Taipaifang, NDRC, BMDRC, TMDRC, SMDRC, GPDRC, SZMG, HPG. Own calculations.

Figure 1 demonstrates several interesting implications. The price performance shows a large volatility in all six trading pilots during the period from April to October in 2014, after that the trading prices show a relatively stationary trend. Prices in Hubei stay at the lowest in almost the entire year, while prices in Beijing become the highest since November 2014.

Carbon trading is shown as in a relatively efficient market when the price performance represents stable trends since October in 2014.

Figure 2. Pilots trading volume and price trends 2a. Beijing

0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 80.00 90.00

Overall price performance

BJ SZ SH GD TJ HB

(15)

-‐30.00%

-‐20.00%

-‐10.00%

0.00%

10.00%

20.00%

30.00%

Price trends-‐SZ

Source: NDRC, BMDRC. Own calculations.

Figure 2a shows there is a significant trading peak in Beijing during the period from June to July 2014 right after the compliance period. The carbon allowance trading prices express a continuing small volatility during the past year. The price trends show a large volatility from June to September in 2014, which there is a significant decline to -0.2% in August, but the prices express a relatively stable phase after that.

2b. Shenzhen

Source: NDRC, SZMG

Figure 2b above indicates a relatively upward trend of trading volumes after the sharp decline behind the trading peak during the compliance period in July. The price trends demonstrate a relatively symmetrical but increasingly volatility, with an outstanding negative changing ratio as -0.3% and a positive changing ratio as 0.3% from November 2014 to

-‐30.00%

-‐20.00%

-‐10.00%

0.00%

10.00%

20.00%

2014/4/2 2014/5/2 2014/6/2 2014/7/2 2014/8/2 2014/9/2 2014/10/2 2014/11/2 2014/12/2 2015/1/2 2015/2/2 2015/3/2

Price trends-‐BJ

0 20,000 40,000 60,000 80,000 100,000 120,000

Trading Volume-‐BJ

0 20,000 40,000 60,000 80,000 100,000 120,000 140,000

Trading Volume-‐SZ

(16)

0 50,000 100,000 150,000 200,000 250,000 300,000 350,000

2014/4/2 2014/5/2 2014/6/2 2014/7/2 2014/8/2 2014/9/2 2014/10/2 2014/11/2 2014/12/2 2015/1/2 2015/2/2 2015/3/2

Trading Volume-‐SH

0 50,000 100,000 150,000 200,000

Trading Volume-‐GD

-‐30.00%

-‐20.00%

-‐10.00%

0.00%

10.00%

20.00%

30.00%

40.00%

Price trends-‐GD

January 2015.The prices in Shenzhen have experience more declines than increases during the past year upon the figures.

2c. Shanghai

Source: NDRC, SMDRC. Own calculations.

Shanghai pilot has relatively small trading volumes compared to other pilots, but it also shows an outstanding trading volume as nearly 300 thousand tons in June, which is far beyond other pilots’ trading performance in the same period. According to the figures, the prices maintain a continuing volatility except in October. There is an obvious decline in prices from July to September but increased in October. Overall, allowance trading prices in Shanghai experience two periods of increase, respectively as in June 2014 and in January 2015, both with larger than 0.2% growth.

2d. Guangdong

-‐30.00%

-‐20.00%

-‐10.00% 10.00% 20.00% 30.00% 0.00%

2014/4/2 2014/5/2 2014/6/2 2014/7/2 2014/8/2 2014/9/2 2014/10/2 2014/11/2 2014/12/2 2015/1/2 2015/2/2 2015/3/2

Price trends-‐SH

(17)

-‐15.00%

-‐10.00%

-‐5.00%

0.00%

5.00%

10.00%

15.00%

20.00%

Price trends-‐HB

-‐15.00%

-‐10.00%

-‐5.00%

0.00%

5.00%

10.00%

15.00%

20.00%

Price trends-‐TJ

Considering Guangdong pilot barely has trading volumes during the year, the observations here will be extended to the launch day on 19^th December in 2013. Trading volumes in Guangdong show three trading peaks respectively in June, in July 2014 and in March 2015. The maximum volumes reach nearly 200,000 tons in June. The price trend states that Guangdong experiences no fluctuation on trading prices from December 2013 to March 2014, however, the prices start floating after that, with most significant volatility in August.

2e. Hubei

Source: NDRC, HPG. Own calculations.

The trading volumes in Hubei show a continuing stable trend, the volumes traded are stated as the largest among these six pilots since it has been launched. The prices in Hubei also demonstrate a relatively stable trend, with only two to three visible fluctuations, separately in August 2014 and March 2015.

2f. Tianjin

Source: NDRC, TMDRC. Own calculations.

0 100,000 200,000 300,000 400,000 500,000 600,000

Trading Volume-‐HB

0 100,000 200,000 300,000 400,000 500,000 600,000 700,000

2014/4/2 2014/5/2 2014/6/2 2014/7/2 2014/8/2 2014/9/2 2014/10/2 2014/11/2 2014/12/2 2015/1/2 2015/2/2 2015/3/2

Trading Volume-‐TJ

(18)

Figure 2f above shows Tianjin pilot has its maximum trading volumes during the time from July to August 2014, where the volumes reach 600,000 tons. Except for the trading peak, the trading volumes stay small and stable during the whole year. On the other hand, the prices show an outstanding volatility before November 2014, especially from August to October.

The positions of two graphs here in Tianjin are reversed compared to in other pilots.

Besides of allowance-based carbon trading scheme, project-based trading system is also applied in China, particularly in Clean Development Mechanism (CDM) programs cooperated with international participants and institutions (Li et al., 2014). Since China has been facing specific problems in market structure and government intervention, also with trading supervision and pricing mechanism (Dong et al., 2014), the research on how factors influence carbon allowance prices shows practical meanings to achieve reduction targets in carbon- intensive industries. For instance, changes in energy prices influence the economy structure, energy efficiency, supply chain and end-product prices, which could possibly affect the carbon prices, as stated by He et al (2011).

4. Methodology and Data

4.1 Baseline model

To get a more adequate and robust view of the estimation of energy prices impact on carbon allowance prices, the paper combines previous literatures with empirical market fundamentals. The goal of this paper is to test whether the relationship between carbon prices and energy prices exists and how energy prices affect carbon prices, according to the hypothesis we developed above (refer to, Section 2). Four variables are included in this paper through VAR modeling.

4.1.1Background of applied methodology

Carbon allowance prices are basically determined by market supply and demand. Supply relies on the total amount of allowance allocation due to the National Allowance Plans (NAPs), the prices are influenced by factors relate to institutional policies and market forces.

The demand depends on several aspects influencing economic growth, for instance, firm’s marginal abatement cost, fuel prices, weather conditions, institutional measures and market structure (see among Christiansen et al., 2005; Paolella et al., 2008; Chevallier et al., 2011a, b;

(19)

model the relationship between energy price and carbon price upon the examination of carbon dynamics. Bunn and Fezzi (2009) use co-integrated vector error-correction model to quantify the interaction of prices as gas, electricity and carbon dynamics. Li et al. (2014) discuss that substitution effects among energy inputs have impact on the emissions reductions upon the State Information Center computable general equilibrium (SICGE) model. Others use time- varying parameter analysis (Nazifi, 2013), Dynamic Conditional Correlation (DCC) model (Kirat, 2011), and also Markov-switching VAR model drawn by Chevallier (2011a). An input-output analysis is assumed to evaluate the carbon dioxide emission in China market due to Lin et al (2010). Given the previous models developed in EU ETS to examine the relationship between carbon prices and energy prices, we suggest VAR modeling would be properly applied in China’s carbon market to deepen the pricing mechanisms. Detailed descriptions and motivations will be discussed in next paragraphs.

4.1.2 VAR modeling

In a classic econometric model, exogenous variables and endogenous variables should be clearly specified. However, in certain circumstances, for instance, in China’s carbon trading market, it is difficult to define these possible correlated economic items such as energy prices and carbon prices as exogenous or endogenous variables. Therefore VAR model is of great comparative advantage due to VAR model contributes to a combination of endogenous variables to estimate changes in one variable are predetermined with respect to another.

Another reason to employ VAR modeling in this paper is due to variety in time series of our data. Our data includes four variables, respectively as carbon allowance prices, oil prices, natural gas prices and coal prices. All observations are collected on time-series levels (see, Section 4.2). Gao (2009) emphasizes that VAR modeling allows the existence of different characteristics of time series, that sequences can be stationary, partly stationary, cointegrated with same-order integration or with various orders’ integration. All time series are regarded as endogenous vectors. As long as there is a cointegrated relationship among the time series, effect of one variable on another variable can be tested. Even if cointegration is not testable due to different orders’ integration in time series, the mutual effects can also be diagnosed as the modeling is proved to be stable, where impulse response and variance decomposition will be applied for further investigations.

In general, the following vector auto-regressive order of p is applied (Kumar et al., 2012):

(20)

1 1

p

t i t t

t

y c

φ

y₋

ε

=

= +

∑

+

Wherey is a (n×1)_t ) vector of endogenous variables, while the intercept vector of the VAR is shown as c=

(

c1K c ʹ′n

)

,φ_i stands for the i th matrix of autoregressive coefficient for

1, 2,

i= K p. The last parameterεt =

(

ε1tK εnt

)

^ʹ′ is the generalization of a white noise process, which stands for a common time-series process upon stationary analysis.

In a VAR model, each variable relies on the lagged values (Gao, 2009; Brooks, 2014). In early days, a lag-augmented VAR testing procedure is proposed to a robust method for integration and co-integration properties of data, assumed by Toda et al. (1995), which has been proved to avoid pre-test bias. The selection of the appropriate lag length is of great importance (Kumar et al., 2012). If the lag length is too large in the sample, then the degree of freedom will decline, and the standard errors will be larger upon the estimated coefficients.

On the other hand, if the orders of integration in time series do not exceed the optimal lag lengths of the model, usual lag length selection could be applied.

VAR modeling in this paper is stated as following steps (Gao, 2009; Brooks, 2014):

-Stationary test for each time-series. An ADF unit root test will be applied to test whether the variables in time sequences are stationary or not, the orders of integration will be calculated to determine whether Granger causality and cointegration test should be further conducted.

-Diagnosis on lag orders in VAR model. Lag orders in the model represent the maximum lag lengths, which can be described in endogenous variables. In this paper, optimal lag orders will be determined in comparison with AIC principle or SC principle (Tang et al., 1991; Gao, 2009).

-Stability test. An AR root test will be tested to see whether the root of the model will be located at the unit circle. The test is conducted to confirm the stability of the model, not of the original time series.

-Impulse response test. After the conformation of the model’s stability, effects of unexpected changes in one variable on another variable and on the rest of the variables are

(21)

processed in an impulse response test. The test is for an absolute effect investigation. Results and interpretations will be shown in Section 5.5.

-Variance decomposition. The test is for a relative effect investigation. Expected variance of each endogenous variable will be decomposed to determine the contribution degrees on another variable within structural shocks.

According to Gao (2009) and Brooks (2014), if the time series show different order’s integration, then cointergration test and Granger causality test are not testable in a VAR model. Thus original data should be covert into data in difference forms, the mutual effect of each variable should be interpreted as volatility performance (see, Section 5).

4.2 Data selection

4.2.1 Fundamental data-carbon allowance prices

Considering China just started operating its CO2 emissions trading market, data is not available for a long time period’s inspection. In our study, daily carbon prices in the allowance trading market are selected from six trading pilots, out of seven, weekends and holidays are excluded. The reason for not choosing Chongqing pilot is that the pilot only had one-day trading volumes on 19^th June 2014 (see, Section 3), the day it launched its scheme.

Additionally, the carbon trading price in Chongqing never changed since the first day, the price stays at 30.74 RMB per ton carbon dioxide. For a fair and apparent view, data from six pilots, Beijing, Shenzhen, Shanghai, Guangdong, Tianjin, Hubei, is selected from 2^nd April 2014, the day when Hubei launched its trading scheme, to 31^st March 2015. Overall, the sample size could be shown below in Table 4. Mentioned in Section 3, the overall price performance and price trends shown in Figure 1 and Figure 2 are measured upon the observations in Table 4.

Table 4. Sample size

Pilots Time duration Observations

Beijing 02.04.2014-31.03.2015 186

Shenzhen 02.04.2014-31.03.2015 233

Shanghai 02.04.2014-31.03.2015 175

Guangdong 02.04.2014-31.03.2015 325

Tianjin 02.04.2014-31.03.2015 196

Hubei 02.04.2014-31.03.2015 239

Source: Tanpaifang.com, BMDRC, GPDRC, TMDRC, SMDRC, SZMG, HPM.

(22)

Specifically, data in the six trading pilots varies due to the regional conditions and different government regulations. Separately, carbon allowances traded have different definitions in the six pilots regarding of the daily trading information published in each official website. However, to obtain an accurate and transparent observation volumes, which will also be used to count for the average carbon price for the VAR model later, average daily prices is measured upon the opening prices, closing prices, maximum dealing prices as well as minimum dealing prices. The measurement refers to the regular carbon trading tendency collected on the Tanpaifang website. In detail, the data from Beijing is selected as the average prices of Beijing Emission Allowance (BEA). Average prices are measured in Tianjin and Shanghai with opening prices, closing prices, maximum dealing prices as well as minimum dealing prices. Prices in Hubei are selected from China Hubei Emission Exchange (CHEEX), the official trading platform in Hubei province. Guangdong Emission Allowance (GDEA) is counted as average prices in Guangdong province. Data from Shenzhen is quite different compared to other pilots. Prices on Shenzhen Allowance (SZA) -2013 are selected as the carbon price from 2^nd April 2014 to 6^th August 2014, while prices on SZA-2014 are measured as carbon prices from the launch day (6^th August 2014) till 31^st March 2015⁸. Derives to hypotheses mentioned before, the carbon allowance prices in each pilot demonstrate how well the trading schemes in each pilot operate, and provide a wholesale view for the trading market during the last year.

4.2.2 Data variables and data selection for VAR model

The carbon prices, regarded as the dependent variable, will be selected from the observations in six trading pilots during the past year. Traditional energy factors, oil prices, natural gas prices and coal prices will be collected as the independent variables. Renewable energy items are excluded in this paper, since traditional energy items are still the main drivers in the economic growth (Chevallier, 2011a; Zhang, 2014b). The four variables will be tested in the correlation analysis to identify the possible relationship, followed by VAR modeling.

8Shenzhen issues pre-allocated emissions quota each year in the first quarter, according to the provision released at 19^thMarch 2014, as《Interim Measures for the administration of carbon emissions trading in Shenzhen》. The annual pre-allocated quota “SZA-2014” was released at 31^stMarch 2014, “SZA-2014”replaced the 2013 annual pre-allocated quota “SZA-2013”to measure compliance rate since the release day (SZMG, 2014).

(23)

For the carbon price, 175 daily data is collected all over the six pilots because the price is available for one pilot but not for all other pilots in one specific day. Thus 175 daily prices are chosen upon the days when all pilots have available trading prices reported. Oil prices are based on the West Texas Intermediate (WTI) historical database, which is a common used index for oil prices⁹ (Zhang, 2014b, c). Furthermore, considering natural gas is represented as several trading forms, such as Liquid Natural Gas (LNG) and Condensed Natural Gas (CNG), prices on natural gas can be categorized based on the factors like export consumption, import consumption, original trading volume and end product procedures as well. Since China has a large amount of imported LNG (National Energy Network, 2014), thus the market-averaged LNG prices are measured upon the ex-factory prices reported from 11 representative regions in daily forms and unit conversion is calculated upon 1 ton equals 1390 cubic meters. As for the coal prices, since coal has various trading forms, for instance, thermal coal, coking coal, anthracite, fat coal, caking coal and non-caking coal, thus coal prices vary from different suppliers and regions in China. Upon the situation, daily averaged coal prices are calculated based on the daily announcements of market-based prices in different forms of coal reported by 50 biggest suppliers in the producing chain. The data is collected from the Cngold website and Taiyuan coal trading platform, the biggest official coal trading system in China.

In summary, all variables are selected from 2^nd April 2014 to 31^st March 2015, general holidays and weekends are excluded. Totally 175 carbon allowance prices, with 164 WTI oil prices presented, 173 coal prices calculated upon the daily reports, and 173 natural gas prices are collected. Considering data availability is counted on each day, 164 observations on each variable are tested in correlation analysis and VAR modeling. “Carbon prices” shown in the next sections stands for the carbon allowance prices.

5. Results

All four variables are computed in a VAR model together with the correlation analysis.

The goal of the tests is to investigate how the changes on energy-related prices impact on the

9China government implemented energy pricing mechanism since 2009 that domestic petroleum product prices will be adjusted upward if the moving average of international crude oil prices rise by more than 40% within 22 consecutive working days (Zhang, 2014b), see others, e.g., Zhang (2014a).

(24)

carbon prices in China’s pilot trading schemes. Detailed results and interpretations will be discussed on testing our hypothesis developed before in Section 2.

5.1 Correlation Analysis

Before applying to VAR model, correlation analysis is conducted. Each variable contains 164 averaged prices during the past year (see, section 4.2.2). Table 6 below shows the correlated indexes results. Indexes presented in the brackets show the coefficient probability.

Table5. Correlation Matrix

CARBON PRICE NATURAL GAS OIL COAL

CARBON PRICE 1

（---）

NATURAL GAS -0.26274 1

（0.0007）（---）

OIL 0.363066 -0.27468 1

（0.0000）（0.0004）（---）

COAL 0.673778 -0.15744 0.205837 1

（0.0000）（0.0441）（0.0082）（---）

Note: correlation indexes are computed upon real prices.

The table above shows several implications. Firstly, all correlation indexes appear to be statistically significant. Secondly, the coal prices and oil prices are positively related to carbon prices, while natural gas prices are negatively related to carbon prices. Finally, the coal prices show largest correlation with carbon prices. Regarding to our hypothesis developed above, the null hypothesis should be rejected while the H1 hypothesis should be reasonabl accepted, that there is a strong relationship between carbon prices and energy prices, with respect to real trading prices.

5.2 Stationary test-ADF unit root test

In general, the test model is presented as (Wang Q and Wang S, 2014):

( )

1

1 ⁿ

t t t z j t j t

j

y α β p y₋ δ y₋ ε

=

Δ = + + − +

∑

Δ +

(25)

In the equation,Δ is a first order difference; α and βy_t ^t interpret the intercept and trend item, ρ is the test statistic; while n is the order of lags, and is determined by AIC (Akaike) guidelines or other methods;δ_j is function expression of n arguments; ε_trepresents the error sequence. The null hypothesis in the unit root test is that the time series includes unit root, which implies that the series is not stationary. When the absolute value of statistical measurement is smaller than the absolute value of threshold upon specific confidence levels, for instance, 1%, then the null hypothesis should be rejected strictly, and the time-series is regarded as strictly stationary.

According to Brooks (2014), if an original time-series is not stationary, d orders of difference are applied to adjust the series to be stationary. If the series becomes stationary in 1 order of difference, the series will be denoted as I(1), which means 1-order integration. If the series becomes stationary in d orders of difference, I(d) will be denoted as d-order integration.

Later on, if all time series become stationary in same orders of integration, then a cointegration test should be applied. In the case that time series show different orders of integration, then variables in difference forms should be tested in remaining modeling process.

The stationary test in this paper is based on the four variables to a deeper examination on the relationship. Each variable represents time-series prices. The test result is shown below in Table 7. All variables are in logarithmic forms (Brooks, 2014).

Table 6. Unit root test on Logarithmic variables

Variables Test Form P-value

Lncarbonprice (C,T,0) 0.1026 Non-stationary

Dlncarbonprice (C,N,0) 0.0000 stationary***

Lnnaturalgas (C,N,2) 0.2807 Non-stationary

Dlnnaturalgas (N,N,1) 0.0000 stationary***

Lnoil (C,T,0) 0.9462 Non-stationary

Dlnoil (C,T,0) 0.0000 stationary***

Lncoal (C,T,0) 0.0000 stationary***

Note: Variables started with letter “Ln” means logarithmic prices, while variables started with letter “Dln”

means prices in difference forms. Test Form (C,T,N) represents including constant term, trend items and lag orders. * means 10% confidence level, **stands for 5% confidence level, *** shows 1% confidence level.

Table 7 demonstrate that coal prices are originally stationary while prices in other three variables (carbon prices, oil prices, natural gas prices) are not originally stationary. However, with 1-order difference, the three time-series variables are stationary, all test statistics are below the 1% confidence level. Thus, Lnoil can be denoted as I(0), other three can be denoted

(26)

as I(1). We define the four time-series variables as different order’s integration, thus cointegration test and Granger causality test cannot be fostered. Variables in 1-order difference form will be applied in next tests, in another word, the influence of volatility in energy prices on carbon prices will be investigated.

5.3 Diagnosis of lag orders

In a VAR model, the diagnosis of the lag orders is crucial. Lag orders show the lag intervals for endogenous variables. Basically, optimal lag orders should be selected upon the criteria that residuals can achieve whitenoise. If the lag orders are too large compared to the sample size, then the needed parameters in the model will exceed the regularones, thus freedom degree would decline. According to Gao (2009) and Brooks (2014), optimal lag orders are determined when AIC and SC guidlines meet minimum lag orders. Tang et al.

(1991) discuss that AIC is more suitable for small-size samples while SC is suitable for a larger sample size investigation. Lag orders are computed in comparison with five test criterions in this paper. The test result is shown in Table 8:

Table7. Lag orders selection

Lags LL LR test FPE test AIC test SC test HQ test

0 293.521 NA 2.80E-07 -3.73575 -3.65721 -3.70385

1 336.8687 83.89884 1.97E-07 -4.08863 -3.695928* -3.929122*

2 356.2108 36.43798 1.89e-07* -4.131752* -3.42489 -3.84464

3 366.5842 19.00679 2.03E-07 -4.05915 -3.03813 -3.64444

4 375.1182 15.19612 2.24E-07 -3.96282 -2.62764 -3.4205

5 385.6871 18.27384 2.41E-07 -3.89274 -2.2434 -3.22281

6 400.335 24.57074 2.47E-07 -3.87529 -1.91179 -3.07776

7 413.1519 20.83784 2.59E-07 -3.83422 -1.55656 -2.90908

8 438.355 39.67442* 2.32E-07 -3.95297 -1.36115 -2.90023

Note: *stands for the selected lag order refer to different tests. Descriptions of tests: LR: sequentia lmodified LR test statistic (each test at 5% level); FPE: Final prediction error; AIC: Akaike information criterion; SC: Schwarz information criterion; HQ: Hanna-Quinn information criterion.

Refer to Table 8, 1 lag order is selected as the optimal order in SC test and HQ test.

However, 2 lag orders is seen as the optimal one in AIC test and FPE test. Considering we

(27)

have our sample size as 164 observations, which is rather small, then2 lag orders are selected based on AIC priciple.

5.3 VAR model-stability test

The stability test is based on the lagged variables with 2 lag orders to examine whether the model is stable or not. In a VAR model, stability is tested upon the AR root criterion, if the reciprocal roots of AR features locate outside the unit circle, the model is not stable, then further impluse response test and variance decomposition test cannot be conducted. The descriptive statistics are shown in Appendix A. The unit circle result is stated below in Figure 3:

Figure 3. Stability Test

The reciprocal roots of AR features are alllocated within the circle, the biggest reciprocal root shown is 0.8899, which is less than 1. We can describe the model is stable.

5.4 Impulse response test

Impulse response test reflects an impact from random disturbance effects on endogenous variables within a specific lagged period. In an impulse response test, changes in one explanatory variable can be diagnosed if one unit impulse is added on the variable within t

-Based on China’s pilot emissions trading scheme

An empirical analysis of energy prices impact on carbon allowance prices in China