International Review of Financial Analysis 71 (2020) 101509

Contents lists available at ScienceDirect

International Review of Financial Analysis

journal homepage: www.elsevier.com/locate/irfa

The hedging effect of green bonds on carbon market risk T

a b b,⁎ c
Jiayu Jin , Liyan Han , Lei Wu , Hongchao Zeng
a
School of Accounting & Finance, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong
b
School of Economics and Management, Beihang University, Beijing 100191, China
c
College of Business, University of Nevada, Reno, NV 89557, United States of America

ARTICLE INFO ABSTRACT

Keywords:Carbon market riskGreen bondsHedging This paper explores effective hedging instruments for carbon market risk. Examining the relationship between
strategiesPortfolio managementMarket indices the carbon futures returns and the returns of four major market indices, i.e., the VIX index, the commodity index,
the energy index and the green bond index, we find that the connectedness between the carbon futures returns
and the green bond index returns is the highest and this connectedness is extremely pronounced during the
market's volatile period. Further, we develop and evaluate hedging strategies based on three dynamic hedge
ratio models (DCC-APGARCH, DCC-T-GARCH, and DCC-GJR-GARCH models) and the constant hedge ratio
model (OLS model). Empirical results show that among the four market indices the green bond index is the best
hedge for carbon futures and performs well even in the crisis period. The paper also provides evidence that the
dynamic hedge ratio models are superior to the OLS model in the volatile period as more sophisticated models
can capture the dynamic correlation and volatility spillover between the carbon futures and market index re-
turns.

1. Introduction levels, and fuel switching behavior play a central role in driving the
volatility of EUAs (e.g., Alberola, Chevallier, & Cheze, 2008; Chevallier,
The European Union Emissions Trading System (hereafter, EU ETS) 2009). Fourth, the trading activities of speculators who are more con-
is the cornerstone of the European Union's drive to reduce emissions of cerned with short-term operation than long-term price trends are also
man-made green house gases which are largely responsible for warming likely to influence the volatility levels in the EU ETS carbon market.
the planet and causing climate change. The implementation of the EU The high volatility of EUAs leads to great carbon market risk and
ETS in 2005 has created the world's largest carbon market that facil- lowers the risk-return trade-off of investments in the carbon market.3
itates the trading of European Union Allowances (hereafter, EUAs) To what extent carbon risk can be hedged by financial instruments has
among regulated firms and other investors.2 Over the last decade, the been of great interest to environmental policymakers, energy-intensive
EU ETS carbon market has continued to grow steadily in terms of firms, portfolio managers, as well as the emerging class of the so-called
trading volume, trading value, and number of market participants. “carbon” investors. Though there are a large number of studies doc-
However, the rapid growth of the market is accompanied by a drastic umenting the correlations between carbon emissions markets and other
increase in the price volatility of EUAs, which can be attributed to the important energy and financial markets (e.g., Balcılar, Demirer,
following factors. First, the cap-and-trade principle of the EU ETS re- Hammoudeh, & Nguyen, 2016; Luo & Wu, 2016; Zhang, Liu, & Yu,
sults in considerable uncertainty regarding the long-run price elasticity 2017), prior works on carbon risk hedging mainly examine hedging
of demand for carbon trading and hence leads to unstable carbon price instruments based on the links between the carbon spot and futures
behavior. Second, the continuous integration of the macroeconomic, markets. For instance, Fan, Rocan, and Akimov (2014) estimate optimal
financial, and commodity markets around the world unavoidably hedge ratios and evaluate the hedging effectiveness of futures contracts
transmits volatilities to carbon markets during global financial crises in the EU ETS carbon market. Balcılar et al. (2016) show that the
(Chevallier, 2011). Third, other energy price changes, institutional carbon spot and futures markets exhibit time-varying correlations and
events, and various market fundamentals such as weather, production volatile hedging effectiveness. Philip and Shi (2016) investigate optimal

Corresponding author at: School of Economics and Management, Beihang University, No.37 Xueyuan Road, Beijing 100191, China.
⁎

E-mail addresses: [email protected] (J. Jin), [email protected] (L. Han), [email protected] (L. Wu), [email protected] (H. Zeng).
2
The EU ETS remains the world's biggest emissions trading market, accounting for over three quarters of international carbon trading.
3
The term carbon risk is defined as a subset of environmental risks that include any corporate risk related to climate change or the use of fossil fuels (Hoffmann &
Busch, 2008; Jung, Herbohn, & Clarkson, 2018).

https://doi.org/10.1016/j.irfa.2020.101509
Received 19 September 2019; Received in revised form 7 March 2020; Accepted 28 April 2020
Available online 19 May 2020
1057-5219/ © 2020 Elsevier Inc. All rights reserved.
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

hedging in the EU ETS market and suggest that participants can benefit index has the strongest connectedness with the carbon market in terms
from using regime switching hedging strategies. To date, there lacks a of return and volatility. Second, the green bond index is an effective and
holistic investigation into the hedging instruments, hedging strategies, inexpensive hedging instrument for carbon risk. From an investor's
and hedging effectiveness for carbon risk based on the dynamic con- perspective, about 6% of the return variance of a one-dollar long po-
ditional correlations between the carbon market and other markets. sition in carbon futures can be hedged by shorting 3 cents of the green
The emergence of green bonds signals a new trend in recovering the bond index. During the crisis period, the green bond index still provides
world economy through developing low carbon technologies to reduce effective hedge against carbon risk while other market indices fail.
carbon emissions. Green bonds are defined as any type of bonds where Third, sophisticated models with time-varying hedge ratios such as the
“the proceeds will be exclusively used to finance or re-finance, in part DCC-APGARCH, DCC-T-GARCH, and DCC-GJR-GARCH models perform
or in full, new and/or existing eligible green projects” (ICMA, 2017: 2f), better than the simple OLS model when green bonds are used to hedge
that is, environmentally- or climate-friendly projects, such as renewable carbon risk. The advantage of dynamic hedge strategies over constant
energy, green buildings, clean transportation, sustainable waste man- hedge strategies becomes more obvious in the volatile period. This
agement, sustainable land use, biodiversity and clean water. With green finding contributes to the open question on the comparison between
bonds, the issuer obtains the capital to finance green projects, while the dynamic hedge ratio models and constant hedge ratio models (Lee,
investors receive fixed income in the form of interest. At maturity, the 2010; Lee & Yorder, 2007).
principal is repaid, unless the issuer goes bankrupt. Green bonds in this We contribute to the existing literature in several ways. First, we
way are the same as any corporate bond, but they are labeled “green” add to the debate on carbon risk hedging and portfolio risk manage-
because the issuer pledges to use the proceeds for environmentally- ment. Effective hedging strategies are particularly important in the
friendly or climate-focused projects in accordance with sustainability presence of significant volatility in carbon emissions markets as well as
standards. In January 2014, the International Capital Markets Asso- time-varying risk transmissions from energy markets to carbon markets
ciation published the Green Bond Principles to establish rules for a bond (Balcılar et al., 2016).Unlike previous studies, which generally focus on
to be labeled as green. The distinction between labeled and unlabeled the effectiveness of carbon derivatives markets in managing carbon
bonds laid out in the Green Bond Principles paves the way for the ex- risk, this paper explores different hedging instruments outside carbon
traordinary growth in the issuance of green bonds from 3bn USD in markets. Further, while a large body of literature documents evidence
2012 to 81bn USD in 2016 (Reboredo, 2018). The explosive growth of on the effects of economic factors such as energy prices, stock market
the green bond market has led Morgan Stanley to describe it as “green prices, and commodity prices on carbon prices, studies that examine the
bond boom” (Morgan Stanley, 2017). potential role of these factors in carbon risk hedging and portfolio risk
The characteristics of green bonds are not just limited to the “green” management are still limited. We provide comprehensive and robust
aspect which is directly linked to promoting and implementing green empirical investigations into this research question by performing a
development initiatives, but also include the government's responsi- detailed analysis of the hedging effectiveness of four major market in-
bility to accelerate climate actions and create geopolitical situations dices on carbon futures. As economic shocks have always been trans-
that are favorable to green deals. As emphasized by Agliardi and mitted to carbon markets, our findings that relevant market indices can
Agliardi (2019), the development of green bonds depends on building be used to hedge carbon risk would provide investors with flexibility in
investor confidence in the green bond market and enhancing the un- designing appropriate hedging strategies.
derstanding of green bonds' characteristics, which are closely associated Second, we provide strong evidence on the role of green bonds in
with regulatory changes and green policies. Because the environment carbon risk management. There is emerging empirical evidence on the
surrounding carbon trading is subject to significant uncertainty due to benefits of green bonds, particularly for firms in carbon intensive in-
the same regulatory changes and green policies, green bonds should dustries. Reboredo (2018) confirms that green bonds have sizeable di-
play an important role in the successful management of carbon market versification benefits for investors in stock and energy markets and that
risk. However, no studies have examined the market connectedness large price fluctuations in those markets have a negligible impact on
between carbon and green bond markets and the hedging role of green green bond prices. Gianfrate and Peri (2019) study the convenience of
bonds in the carbon market remains unknown. We aim to fill this gap in issuing green bonds for European issuers. They find that green bonds
the literature. are more financially convenient than non-green ones. The findings
Our study focuses on exploring the potential role of green bonds in support the view that green bonds can potentially play a major role in
hedging carbon market risk. Considering the continuous integration of greening the economy. However, as emphasized by Monasterolo and
the macroeconomic, financial, and commodity markets around the Raberto (2018), there is still high uncertainty in their design, due to
world, we address this issue using a five-variable system, which in- lack of consolidated knowledge of their direct and indirect effects on
cludes the carbon, stock, commodity, energy, and green bond markets. the real economy, financial markets, as well as carbon markets. In this
We use the S&P 500 Dynamic VIX Futures Index (hereafter, VIX index), study, we examine whether green bonds can be used to hedge carbon
the S&P Dynamic Commodity Futures Index (hereafter, commodity risk and provide a detailed analysis of the extent to which green bonds
index), the S&P Energy Index (hereafter, energy index), and the S&P can offer protection for carbon investments. Through documenting
Green Bond Index (hereafter, green bond index) to capture economic evidence on how green bonds impact carbon risk management, we
shocks from the stock, commodity, energy, and green bond markets. In make the first attempt at enhancing our knowledge regarding the ef-
the five-variable system, we compare the hedging performance of the fects of green bonds on the real economy.
green bond index to that of other market indices. Our study also derives Third, we contribute to the ongoing work that investigates the in-
hedging strategies for carbon risk based on suitable models. Constant terrelationship between carbon markets and other important markets in
hedge ratio models (e.g., OLS model) and dynamic hedge ratio models different phases of the EU ETS. This strand of literature has largely
(e.g., multivariate GARCH model) are popular models in the hedging focused on the links between carbon and energy markets. For example,
literature. We use the DCC-APGARCH, DCC-T-GARCH, and DCC-GJR- Ji, Zhang, and Geng (2018) study information linkages and dynamic
GARCH models to calculate optimal hedge ratios and construct hedged spillover effects between these two markets. They find that electricity
portfolios, based on which we develop and evaluate dynamic hedging prices are the biggest receiver of information from the carbon market
strategies. We have the following findings. First, we find evidence that and that Brent oil prices significantly impact carbon price changes and
the EU ETS carbon market is sensitive to changes in economic condi- risks. Tian, Akimov, Roca, and Wong (2016) examine the relationships
tions and has always been impacted by economic shocks. The carbon between returns of electricity stocks in Europe and EUA returns. They
market is generally a receiver rather than a transmitter of volatility in document significant positive conditional correlations between the
the economic system. Among the four market indices, the green bond volatility of EUAs and that of electricity stocks in Phase II. Extending

2
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

these studies, we perform a comprehensive analysis to assess the links show that the volatilities of energy markets such as Brent oil, coal,
between the EU ETS carbon market and the stock, commodity, energy, natural gas, and electricity may be used to forecast the volatility of
and green bond markets in the EU ETS phases II and III. We find sig- carbon futures. Using the orthogonal GARCH model, Luo and Wu
nificant increases in the correlation coefficients of the carbon and green (2016) assess the time-varying correlations in EUAs, crude oil, and
bond pair and of the carbon and energy pair during 2008 to 2013. Such stock markets. They find that the correlations between EUAs and the
increases coincide with the time periods of the 2008 global financial equity markets are higher and more volatile in the US and Europe than
crisis, the 2010 European sovereign debt crisis, and the drastic drop in in China. Ji et al. (2018) provide further evidence that oil prices have a
the EUA price from 2012 to 2013. Our results can be reasonably ex- significant spillover impact on changes in carbon prices and risks.
plained by the financialization of carbon futures, which has emerged
with the penetration of interest-bearing capital in all economic activ-
ities and made the carbon futures market more integrated with other 3. Data and preliminary analyses
important markets, especially during volatile periods.
The remaining parts of this paper are organized as follows. Section 2 3.1. Data
presents literature review. Section 3 describes data and presents pre-
liminary analyses. Section 4 introduces methodology and reports esti- Our dataset consists of daily carbon futures prices and daily price
mation results. Section 5 discusses robustness tests. Finally, Section 6 series of four market indices, including the S&P 500 Dynamic VIX
concludes. Futures Index, the S&P Dynamic Commodity Futures Index, the S&P
Energy Index, and the S&P Green Bond Index.4 All data are obtained
2. Literature review from Wind Database. More details of these price series are discussed
below.
Over the last decade, there has been a growing body of literature on The EUA Futures Contract is a deliverable contract where each
carbon markets. The first strand of research has explored the price clearing member with a position open at cessation of trading for a
dynamics of EUAs. Benz and Trück (2009) examine the short-term spot contract month is obliged to make or take delivery of EUAs to or from
price behavior of EUAs using Markov switching and AR-GARCH the Union Registry. There are up to seven December, six quarterly, and
models. Conrad, Rittler, and Rotfuß (2012) establish that high-fre- two monthly contracts or as otherwise determined and announced by
quency EUAs returns not only obey conditional heteroscedasticity but the exchange from time to time. Last Monday of the contract month is
also are characterized by long memory. They find that the asymmetric the expiration date. As the near-December contracts convey a reliable
power GARCH process can better capture the high-frequency price mid-term price signal for market operators (Chevallier, 2009), the price
dynamics of EUAs. Sanin, Violante, and Bataller (2015) study the vo- series of successive futures contracts are spliced together so that ob-
latility dynamics before and after 2008. They show that the sharp in- servations shift to the next near-December contract on each expiration
creases in volume and policy announcements have led to greater vo- day. The last year's futures prices before the expiration of each De-
latility from spring 2005 to the end of 2007. Using ultra-high frequency cember contract therefore form a single time series of futures prices.
data, Bredin, Hyde, and Muckley (2014) assess the degree of develop- The VIX index is a real-time market index that represents the mar-
ment in the EU ETS futures market and find a negative relationship ket's expectation of 30-day forward-looking volatility. It is derived by
between volume and volatility. solving for the uncertain component of the standard option pricing
The second line of research examines the role of institutional in- formula using the market prices of S&P 500 index options as inputs.
vestors in the EU ETS carbon market. Hintermann (2017) investigates Investors, research analysts, and portfolio managers generally use the
market power in the EU ETS carbon market and finds evidence sup- VIX index to measure market risk and investors' sentiment. Chung, Tsai,
porting that both firm participants and institutional investors play in- Wang, and Weng (2011) provide evidence that the information content
dispensable roles in maintaining carbon market liquidity. Particularly, of the VIX options can be used to improve the prediction of returns,
Jaraite-Kažukausk and Kažukauskas (2015) show that smaller EU ETS volatility, and density in the S&P 500 index. We argue that carbon fu-
firms with less trading experience prefer to trade their permits in- tures returns are sensitive to the same risk-driving factors as stock
directly via third parties instead of trading directly by themselves. markets, and thus the VIX index may be used to explain the variation in
Those professional investors trade for their clients by relying on spec- the carbon futures returns.
ulative tools. Holt and Shobe (2016) implement a laboratory experi- The commodity index uses a quantitative methodology to track the
ment with financially motivated participants to compare alternative prices of a diversified portfolio of 24 commodity and financial futures
proposals for managing the time paths of EUA prices in the presence of contracts. Commodities are a crucial economic factor. Given the world's
random firm-specific and market-level structural shocks. They find that dependence on commodities, fluctuations in their prices have important
the implementation of a price collar which captures the auction reserve effects on the performance of national and global economies (Hamilton,
price and soft price cap effectively enhances market performance in 2003) and are used as predictors of economic downturns (Hamilton,
both the long-run and short-run horizons. Besides maintaining liquidity 2011). The likely influences of commodity markets on the carbon fu-
in carbon markets, making profit is also an important goal for firm tures prices are documented in Chevallier (2009). The influences also
participants. Makridou, Doumpos, and Galariotis (2019) investigate the appear to be bidirectional. As discussed in Carmona, Fehr, Hinz, and
financial performance of firms that participate in the EU ETS from 2006 Porchet (2010) and Kara et al. (2008), the cost of CO2 emissions al-
to 2014. They show that both economic and energy related variables lowances affects directly or indirectly almost all emission-related
significantly influence firms' profitability. commodities and economic activities. When the level of economic ac-
Another line of research investigates the linkage between carbon tivity is low, the industrials sector produces less in the face of de-
emissions and other important energy and financial markets. This creasing demand, which in turn decreases CO2 emissions as well as the
strand of research generally shows that carbon prices are significantly demand for allowances.
affected by economic aggregate variables. Chevallier (2009) argues that The energy index is designed to measure some of the largest publicly
carbon futures prices do not immediately respond to changes in mac- traded companies engaged in the exploration, production, marketing,
roeconomic conditions but are remotely connected to macroeconomic
variables with some time lag. Lutz, Pigorsch, and Rotfub (2013) ex- 4
Several investable products linked to these four market indices are avail-
amine the nonlinear relationship between the EUA price and its fun- able. For example, Fubon S&P 500 VIX Futures ETF, Horizons Beta Pro S&P500
damentals. They show that changes in stock prices and energy prices VIX Futures ETF, ishares S&P GSCI Commodity-Indexed Trust, ishares Global
are the most important drivers of EUA prices. Byun and Cho (2013) Energy ETF, and so on.

3
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

refining and/or transportation of oil and gas products, coal and other and the returns of energy and green bond indices. The results that an
consumable fuels from around the world. The energy index has increase in the stock market volatility is accompanied by a decrease in
emerged as an industry benchmark. As discussed earlier, a large body of the energy and green bond index returns may indicate greater will-
literature on the linkage between carbon futures prices and energy ingness of investors to take on risk, which in turn fuels stock market
prices has documented that the carbon futures prices depend sig- volatility. We also find large and significantly positive correlations
nificantly on the prices of electricity, gas, and coal. In addition, carbon between the energy index returns and the green bond index returns,
emissions markets and energy markets are sensitive to the same state which suggests a low degree of global energy substitutability despite
variables such as climate change, market conditions, and geopolitical the rapid growth of green projects. More importantly, we find that the
situations. correlation between the carbon futures returns and the green bond
The green bond index is designed to track the global green bond index returns is the largest with a correlation coefficient of 0.25. This
market. This pioneering index maintains stringent standards in order to preliminary finding implies that the green bond index could potentially
include only those bonds whose proceeds are used to finance en- act as an effective hedge when the carbon market falls, since a long
vironmentally-friendly projects. Green bond prices are theoretically position in carbon futures can be hedged by a short position in the
linked to financial markets through the discount rate channel. Also, green bond index.
green bond prices can provide investors with the right signals and in- Though correlation-based analyses remain widespread, they mea-
centives to invest in sustainable, inclusive, and innovation-based sure only unconditional correlations and are constrained by linearity.
growth (Stern, 2016). Hence, it is a crucial economic and political in- We investigate whether the correlations are asymmetric in the sense
dicator. that positive and negative shocks behave differently, and whether the
Philip and Shi (2016) point out that the return dynamics of EUAs in connectedness is directional in the sense that some markets are trans-
the EU ETS phase I5 are substantially different from those in other mitters of volatility whereas others are receivers of volatility. We do not
phases due to regulatory and trading mechanism changes. Furthermore, impose any underlying economic structure such as the long-term and
the green bond index is not publicly available until November 28, 2008. stable relation required for cointegration that is common in many re-
We thus avoid the infancy stage of the EU ETS market and select a cent works, neither do we impose weak exogeneity on any variables.
sample period from December 1, 2008 to August 31, 2018, which Instead, we use a reduced-form vector autoregression (VAR) to conduct
covers EU ETS phase II and phase III. Fig. 1 shows the time series of the analysis, relying on a battery of causality tests and variance de-
daily prices for carbon futures and the four market indices. compositions to provide information on the asymmetric effect and to
The daily returns of carbon futures and the four market indices are conduct the directional connectedness analysis, respectively.
calculated as the difference between the logarithms of two consecutive
( )
P
prices, that is, Rt = ln P t × 100 , where Pt is the price at time t.
t 1
3.2.2. Asymmetric effect
Table 1 provides descriptive statistics for the daily returns. On average, To test the asymmetric effect of positive and negative shocks, we
carbon futures and the energy and green bond indices have positive employ the asymmetric causality tests proposed by Hatemi-J (2012).6
returns, while the VIX and commodity indices have negative returns. As The starting point of this analysis is to define the price (in levels) as a
for volatility, the carbon futures returns have a daily standard deviation random walk process:
of 3.34, which is about three times the average standard deviation of t 1
returns for the four market indices, indicating high investment risk in Pt = Pt 1 + t = P0 + t i.
the carbon market. Among the four market indices, the standard de- i=0 (1)
viation of returns is the highest for the energy index, followed by the
The constant P0 is the initial price of the underlying variable.
VIX and commodity indices, and the lowest for the green bond index.
Hatemi-J (2012) then proceeds to calculate cumulative positive and
Further, we find that the skewness statistics are either positive or ne-
negative sums of price shocks as
gative and all the kurtosis statistics are greater than three, indicating
that the distributions of all return series are skewed and leptokurtic. All t 1 t 1

return series are stationary as confirmed by the augmented Dickey- Pt+ = Pt+ i, Pt = Pt i,
(2)
Fuller (ADF) test. i=0 i=0

where ∆Pt+ = max (∆Pt, 0) and ∆Pt− = min (∆Pt, 0).

3.2. Preliminary analyses Asymmetric causality tests can now be performed between any two
variables. For example, to evaluate the impact of increases in the
In this subsection, we perform a number of preliminary analyses on carbon futures returns on the increases in the green bond index returns,
the connectedness between the carbon futures returns and the four we test the hypothesis of Granger causality in the context of a bivariate
market index returns. We proceed in three steps. First, in Section 3.2.1, VAR model that involves Pcarbon, t+ and Pgreenbond, t+. In Table 3, we
we use the full sample to calculate the unconditional correlation coef- calculate all the 32 combinations of asymmetric causality tests between
ficients of these variables. In Section 3.2.2, we perform a battery of carbon futures and the four market indices, to capture possible asym-
causality tests in the VAR framework to measure the asymmetric effect. metric effects among these variables.
In Section 3.2.3, we perform a directional connectedness analysis and a As reported in Table 3, the overall evidence supports strong asym-
rolling-sample analysis to “zoom in” on the directional connectedness metric effects in the carbon futures returns and the four market index
during the market's volatile period. These preliminary analyses con- returns, clearly suggesting that ignoring asymmetry in their relation-
tribute to the model choice for our estimation strategy. ships may lead to misleading conclusions. Specifically, the causality
tests reject the hypothesis of Non Granger causality from positive or
3.2.1. Unconditional correlations negative shocks in the commodity, energy, and green bond index re-
In Table 2, we report the unconditional correlations between the turns to negative shocks in the carbon futures returns. Currently, ex-
carbon futures returns and the four market index returns. We find large isting carbon prices generally fall below the carbon prices required to
and significantly negative correlations between the VIX index returns meet the EU emissions target. There is considerable uncertainty in the
long-run price elasticity of demand for fossil fuels and this uncertainty
5
EU ETS phase I is from January 1, 2005 to December 31, 2007; EU ETS
6
phase II is from January 1, 2008 to December 31, 2012; EU ETS phase III is from For more applications of this method, refer to Nguyen, Sousa, and Uddin
January 1, 2013 to December 31, 2020. (2015) and Batten, Ciner, and Lucey (2017).

4
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

Fig. 1. Time series of daily prices for carbon futures and four market indices.
Notes: the figure shows the time series of daily prices for the carbon futures, the VIX futures index, the commodity futures index, the energy index, and the green bond
index. The sample period is from December 1, 2008 to August 31, 2018.

partly explains why existing carbon prices are so low. Our finding Yilmaz (2014) and is closely related to the econometric notion of a
suggests that downturns in the carbon futures market are associated variance decomposition, in which the forecast error variance of variable
with fluctuations in the commodity, energy, and green bond markets. i is decomposed into parts attributed to the other variables in the VAR
We find evidence for significant causality from positive and nega- system. Following Diebold and Yilmaz (2014), we denote di, jH as the
tive shocks in the carbon futures returns to positive and negative shocks fraction of variable i's H-step forecast error variance due to shocks in
in both energy and green bond index returns. This finding suggests that variable j. Our analysis is based on the five-variable system of the
investors in energy and green bond markets are also sensitive to fluc- carbon futures returns and the four market index returns. We adopt a
tuations in the carbon market due likely to the fact that the carbon horizon of H = 20 days.
market can to some extent reflect climate change, uncertainty in green Table 4 is called a directional connectedness table. Its main upper-
public policies, and geopolitical situations. We also find that the caus- left 5 × 5 block contains variance decomposition results. The off-di-
ality from the VIX index returns to the carbon futures returns is sig- agonal entries measure pairwise directional connectedness, and there
nificant in all asymmetric patterns, while the causality from the other are 20 (52−5) separate pairwise directional connectedness values. The
way around is insignificant. That is, the VIX index returns do not re- directional connectedness table is simply augmented with a rightmost
spond to any changes in the carbon futures returns. column containing off-diagonal row sums and a bottom row containing
off-diagonal column sums. We label the rightmost column as “From
3.2.3. Directional connectedness others”, in which the total directional connectedness from other vari-
5
The directional connectedness approach is proposed by Diebold and ables to variable i is calculated as CiH = j = 1, j i diH
, j . Likewise, we label

5
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

Table 1 the bottom row as “To others”, in which the total directional con-
Descriptive statistics of daily returns for carbon futures and four market indices. nectedness to other variables from variable j is calculated as
C H j = i = 1, i j diH, j . There are five “From others” and five “To others”
5
Carbon VIX Commodity Energy Green bond
values. We are also interested in net total effects. The difference be-
Mean 0.0090 −0.0320 −0.0190 0.0570 0.0130 tween the total directional connectedness to others and total directional
Median 0.0050 0.0000 0.0000 0.0000 0.0150
connectedness from others yields the net total directional connected-
Max 24.7430 12.0960 14.9990 17.4740 6.8320
Min −43.1070 −10.1440 −4.1280 −9.5280 −3.7770 ness to others, that is, CiH = C∙←iH − Ci←∙H. The bottommost row in
SD 3.3420 1.1480 0.7780 1.4210 0.5730 Table 4 presents five values of net total directional connectedness.
Skewness −0.753 1.089 2.882 0.814 1.087 We are especially interested in how the shocks in the carbon market
Kurtosis 18.350 26.419 60.870 15.942 20.104 is transmitted to other markets (C∙←carbonH), and how the carbon market
Jarque-Bera 24,680.6⁎ 57,391.7⁎ 350,902.2⁎ 17,652.3⁎ 30,842.2⁎
ADF −14.122 ⁎
−14.775 ⁎
−14.632 ⁎
−14.945 ⁎
−15.601⁎
receives shocks from other markets (Ccarbon←∙H). The highest observed
pairwise connectedness from other variables to the carbon futures re-
Notes: the table provides descriptive statistics for the daily returns of carbon turns is from the green bond index returns (Ccarbon←greenbondH = 5.75).
futures and the four market indices. The four market indices include the VIX Meanwhile, the highest observed pairwise connectedness from the
futures index, the commodity futures index, the energy index, and the green carbon futures returns to other variables is to the green bond index
bond index. The daily return is calculated as the difference between the loga- returns (Cgreen bond←carbonH = 5.52). In fact, the pairwise connectedness
rithms of two consecutive prices, that is, Rt = ln ( ) × 100, where P is the
Pt
Pt 1 t from the green bond index returns to the carbon futures returns is the
price at time t. largest off-diagonal element in the 5 × 5 block, and the pairwise
⁎
Indicates significance at the 5% level. connectedness from the carbon futures returns to the green bond index
returns is the second largest. These results show that the carbon futures
Table 2 returns have much higher connectedness with the green bond index
Unconditional correlations between carbon futures returns and four market returns than with any other market index returns.
index returns. In terms of the net total directional connectedness, the VIX index
Carbon VIX Commodity Energy Green bond
returns (0.61) lead the way, followed by the green bond index returns
(0.46), and the carbon futures returns have a negative value (−0.35).
Carbon 1.0000 These results imply that the carbon market is a receiver of volatility in
VIX −0.0841⁎ the economic system, while the green bond market is a transmitter of
(0.0000)
volatility. Further, though the connectedness is high from green bond to
Commodity −0.0392 0.0109
(0.0503) (0.5860) carbon (5.75) and from carbon to green bond (5.52), the net directional
Energy 0.0713⁎ −0.1542⁎ 0.0591⁎ connectedness is dominated by the connectedness from green bond to
(0.0004) (0.0000) (0.0032) carbon (5.75–5.52 = 0.23), which indicates that the green bond market
Green bond 0.2458⁎ −0.1444⁎ −0.0440⁎ 0.1715⁎ 1.0000
spreads its shocks to the carbon market in net terms. This result is
(0.0000) (0.0000) (0.0280) (0.0000)
consistent with the view that green bonds can provide investors with
Notes: this table presents the unconditional pairwise Pearson correlation coef- the signals to invest in sustainable, inclusive, and innovation-based
ficients for the carbon futures returns and the four market index returns. The growth. The volatility in the green bond market to some extent reflects
four market indices include the VIX futures index, the commodity futures index, the uncertainty in green policies, which in turn increases the un-
the energy index, and the green bond index. Probabilities are in parentheses. certainty in the carbon market.
⁎
Indicates significance at the 5% level.

Table 3
Asymmetric effects in carbon futures and four market index returns.
Null hypothesis: F-statistic Prob. Null hypothesis: F-statistic Prob.

VIX
VIX+ ≠ > Carbon+ 4.71⁎ 0.01 Carbon+ ≠ > VIX+ 0.41 0.67
VIX− ≠ > Carbon− 5.88⁎ 0.00 Carbon− ≠ > VIX− 0.79 0.46
VIX+ ≠ > Carbon− 8.15⁎ 0.00 Carbon− ≠ > VIX+ 0.99 0.37
VIX− ≠ > Carbon+ 4.89⁎ 0.01 Carbon+ ≠ > VIX− 0.78 0.46

Commodity
Commodity+ ≠ > Carbon+ 0.65 0.52 Carbon+ ≠ > Commodity+ 2.57 0.08
Commodity− ≠ > Carbon− 3.15⁎ 0.04 Carbon− ≠ > Commodity− 0.29 0.75
Commodity+ ≠ > Carbon− 1.63 0.20 Carbon− ≠ > Commodity+ 0.28 0.76
Commodity− ≠ > Carbon+ 1.63 0.20 Carbon+ ≠ > Commodity− 0.42 0.66

Energy
Energy+ ≠ > Carbon+ 1.17 0.31 Carbon+ ≠ > Energy+ 22.29⁎ 0.00
Energy− ≠ > Carbon− 1.92 0.15 Carbon− ≠ > Energy− 2.96⁎ 0.05
Energy+ ≠ > Carbon− 3.46⁎ 0.03 Carbon− ≠ > Energy+ 22.01⁎ 0.00
Energy− ≠ > Carbon+ 2.88 0.06 Carbon+ ≠ > Energy− 1.63 0.20

Green bond
Green bond+ ≠ > Carbon+ 0.98 0.37 Carbon+ ≠ > Green bond+ 10.07⁎ 0.00
Green bond− ≠ > Carbon− 3.22⁎ 0.04 Carbon− ≠ > Green bond− 10.68⁎ 0.00
Green bond+ ≠ > Carbon− 3.62⁎ 0.03 Carbon− ≠ > Green bond+ 11.03⁎ 0.00
Green bond− ≠ > Carbon+ 1.52 0.22 Carbon+ ≠ > Green bond− 9.81⁎ 0.00

Notes: the table examines the asymmetric effects in the carbon futures returns and the four market index returns. The four market indices include the VIX futures
index, the commodity futures index, the energy index, and the green bond index. To evaluate the impact of A on B, the asymmetric causality test is performed in the
context of a bivariate VAR model that involves A and B. A ≠> B means that variable A does not Granger cause variable B. + represents the positive shocks and −
represents the negative shocks. The optimal lag length for the VAR model is selected using the Akaike Information Criterion.
⁎
Indicates significance at the 5% level.

6
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

Table 4 In essence, both the T-GARCH and the GJR-GARCH models are
Directional connectedness between carbon futures returns and four market nested within the APGARCH model. We thus introduce the general form
index returns. of the multivariate DCC-APGARCH(1,1) model as follows.
Carbon VIX Commodity Energy Green bond From others rt = µ + art 1 + t, (3)
Carbon 92.77 0.67 0.15 0.66 5.75 7.23
VIX 0.69 94.69 0.12 2.35 2.16 5.31 t = + (| t 1| t 1) + t 1, (4)
Commodity 0.18 0.03 99.19 0.33 0.28 0.81
Energy 0.49 2.84 0.34 93.29 3.05 6.71
Ht = Dt Corrt Dt , (5)
Green bond 5.52 2.38 0.19 2.68 89.23 10.77
To others 6.88 5.92 0.79 6.01 11.23 Qt = (1 1 2)Q + 1 t 1 t 1 + 2 Qt 1. (6)
Net −0.35 0.61 −0.02 −0.7 0.46
In the mean Eq. (3), rt is the vector of index returns, μ is the vector of
Notes: the table reports the variance decomposition values based on the five- expected returns, and εt is the vector of zero-mean white noises. An AR
variable VAR system of the carbon futures returns and the four market index (1) term is included into the conditional mean to account for possible
returns. The four market indices include the VIX futures index, the commodity autocorrelation in the returns. In the volatility Eq. (4), ω is the constant
futures index, the energy index, and the green bond index. The optimal lag term, while γ represents the asymmetric effect. A positive (negative)
length for the VAR model is 1, which is selected using the Akaike Information value of γ implies that the past negative (positive) shocks have a larger
Criterion. Its main upper-left 5 × 5 block shows the estimated contribution to
impact on current conditional volatility than past positive (negative)
the 20-step ahead forecast error from variable j to variable i. The column “From
shocks. The ARCH and GARCH coefficients are α and β, respectively.
others” shows the total connectedness from other variables to variable i. The
row “To others” shows the total connectedness from variable j to other vari- The ARCH coefficient α measures short-term persistence and the
ables. The row “Net” shows the net total connectedness from variable j to all GARCH coefficient β measures long-term persistence or volatility clus-
other variables. tering. δ is the power term that captures volatility clustering by chan-
ging the influence of lagged residuals and past standard deviations. In
the APGARCH model, the power term is estimated within the model
The analysis of the full-sample connectedness provides a good rather than being imposed in advance. As emphasized by McKenzie,
characterization of the “average” directional connectedness among Mitchell, Brooks, and Faff (2001), the advantage of APARCH model is
variables, yet by construction it contains no information on the dy- that the APARCH model allows a power term to include any positive
namics of connectedness. We now perform a dynamic analysis using a value and so permits a virtually infinite range of transformations.
rolling estimation window of 200 observations. In Fig. 2, we plot the In Eq. (5), Dt = diag (σt). The Eq. (5) involves two-stage estimation
dynamic net pairwise connectedness from the four market index returns of the conditional covariance matrix Ht. In the first stage, a univariate
to the carbon futures returns. APGARCH model is fitted for each time series, and estimates of σii, tδ/2
The dynamic analysis of the directional connectedness provides us are obtained. In the second stage, time series residuals are transformed
with a better understanding of the strong connectedness between the by their estimated standard deviations from the first stage. That is, ηi,
carbon futures returns and the green bond index returns. Earlier in our δ/2
t = εi, t/σii, t , where ηi, t is then used to estimate the parameters of the
sample period, the 2008 global financial crisis influenced the behavior conditional correlation. In the DCC Eq. (6), Qt is the n × n time-varying
of the connectedness between these returns. This financial crisis led to a correlation matrix of residuals, and Q is the n × n time-invariant
positive value of net directional connectedness from the green bond covariance matrix of residuals. The non-negative scalar parameters φ1
index returns to the carbon futures returns, implying that the green and φ2 are used to construct the dynamic conditional correlations with
bond market spreads its volatility to the carbon market. With the burst a sum of less than unity. φ1 and φ2 capture the effects of the previous
of the 2010 European sovereign debt crisis, the net pairwise con- shocks and the previous conditional correlation on the current condi-
nectedness jumped from a low value of around 1 at the beginning of tional correlation, respectively.
2010 to nearly 5 in mid-2011 and stayed around 3 until the end of Since Qt does not have unit elements on the diagonal, the correla-
2012. Thus, it would be beneficial to design a hedging strategy for tion matrix Corrt is obtained by scaling it as follows:
carbon risk using the green bond index because the linkage between the
green bond market and the carbon futures market is extremely pro- Corrt = (diag (Qt )) 1/2Q
t (diag (Qt )) 1/2 .
(7)
nounced during the crisis period. We focus on this issue in the following
The off-diagonal element of Corrt has the form:
sections.
qij, t
= i , j = 1, 2…n, and i j.
4. Empirical models and estimation results ij, t
qii, t qjj, t (8)

4.1. Dynamic hedge ratio models In the empirical literature, the DCC-T-GARCH and DCC-GJR-GARCH
models are also commonly used to capture the asymmetric effect and
ARCH and generalized ARCH (GARCH) models have become time-varying connectedness. The DCC-T-GARCH model involves a
common tools for dealing with time series heteroskedastic features, and power term of one, which implies that the conditional standard de-
these models are widely used in risk analysis, portfolio management, viation of a series is related to lagged absolute residuals and past ab-
and hedging strategy selection. In this paper, we employ three GARCH solute standard deviation, while the DCC-GJR-GARCH model involves a
models, i.e., the Asymmetric Power GARCH (APGARCH) model pro- power term of two, which implies that the conditional standard de-
posed by Ding, Granger, and Engle (1993), the T-GARCH model pro- viation of a series is related to lagged squared residuals and past squared
posed by Zakoian (1994), and the GJR-GARCH model proposed by standard deviation. We consider all the three DCC-GARCH models in
Glosten, Jagannathan, and Runkle (1993). We further include the dy- the empirical analysis.
namic conditional correlation (DCC) model (Engle, 2002) into the
GARCH framework for the maximum flexibility and accommodation to 4.2. Estimation results from dynamic hedge ratio models
the data with time-varying connectedness. Thus, the full specifications
for the three GARCH models are the DCC-APGARCH, the DCC-T- Table 5 presents the estimated results of the DCC-APGARCH, DCC-
GARCH, and the DCC-GJR-GARCH. T-GARCH, and DCC-GJR-GARCH models for the carbon futures returns

7
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

Fig. 2. Dynamic net pairwise connectedness from market index returns to carbon futures returns.
Notes: the figure plots the dynamic net pairwise connectedness from the four market index returns to the carbon futures returns. The four market indices include the VIX futures
index, the commodity futures index, the energy index, and the green bond index. The difference between the directional connectedness from the market index to the carbon
futures and the directional connectedness from the carbon futures to the market index yields the net pairwise connectedness from the market index to the carbon futures.

and the four market index returns. We focus on the asymmetrical term the VIX index returns. As the VIX index measures the future expectation
γ, the ARCH term α, and the GARCH term β in the conditional variance of market volatility, a positive shock in the VIX index implies an in-
equation, and the coefficients φ1 and φ2 in the conditional correlation crease in the level of expected volatility. Given the nature of the VIX
equation. For the DCC-APGARCH model, the power term δ is another index, the positive asymmetric term γ provides evidence of volatility
focal point. clustering since a positive γ represents long-term persistence of volati-
As reported in Table 5, the coefficients α and β are generally sta- lity.
tistically significant, indicating that the price return volatilities of In Table 5, power terms that are larger than 1 but smaller than 2 are
carbon futures and the four market indices are influenced by both their reported as the optimal transformation for the price return series of
own past shocks and volatilities. As the estimate of β is larger than that carbon futures and the VIX, energy, and green bond indices, while a
of α for each price return series, the price returns are more sensitive to smaller than 1 power term is reported as the optimal transformation for
past volatilities than to past shocks. In the DCC equation, as the sum of the commodity index. The parameter λ reported in Table 5 is the Shape
φ1 and φ2 is close to but less than 1, the dynamic conditional correla- parameter for the fitness of models. As λ is equal to the degree of
tions exhibit weak mean reversion. freedom, the larger λ is the closer the shape of the t-distribution is to a
The asymmetric effect is captured in the carbon futures returns by normal. We find that for the carbon futures returns and the energy
the DCC-T-GARCH model. As the asymmetric term γ is significantly index returns, the DCC-APGARCH model fits the data best, while for the
positive, the negative shocks in the carbon futures price tend to increase other price return series, the DCC-GJR-GARCH model fits the data best.
volatility more than the positive shocks do at the same magnitude. The We use the Ljung-Box test statistics on the standard residuals to eval-
asymmetric effect is also captured in the energy index returns by all uate the goodness of fit among these models. The results of the Ljung-
three GARCH models and in the commodity index returns by the DCC- Box test reported in Table 6 show that the residuals from all three
GJR-GARCH model. The asymmetric term γ is significantly negative in GARCH models can be considered as independently distributed.

8
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

Table 5
Estimation results from dynamic hedge ratio models.
DCC-APGARCH DCC-T-GARCH DCC-GJR-GARCH

Estimate Std. error Pr(>|t|) Estimate Std. error Pr(>|t|) Estimate Std. error Pr(>|t|)

Carbon
μ 0.0588 0.0449 0.1900 0.0599 0.0426 0.1590 0.0595 0.0444 0.1801
a −0.0046 0.0217 0.8339 −0.0042 0.0223 0.8506 −0.0038 0.0209 0.8572
ω 0.0623⁎ 0.0289 0.0310 0.0439⁎ 0.0157 0.0052 0.1126⁎ 0.0452 0.0128
α 0.0998⁎ 0.0157 0.0000 0.0961⁎ 0.0156 0.0000 0.0769⁎ 0.0153 0.0000
β 0.9071⁎ 0.0161 0.0000 0.9127⁎ 0.0151 0.0000 0.8987⁎ 0.0170 0.0000
γ 0.1665 0.0898 0.0637 0.2490⁎ 0.0754 0.0010 0.0352 0.0195 0.0712
δ 1.3782⁎ 0.2858 0.0000
λ 5.8631⁎ 0.6501 0.0000 5.8325⁎ 0.6435 0.0000 5.8399⁎ 0.6497 0.0000

VIX
μ 0.0060 0.0124 0.6301 0.0107 0.0109 0.3223 0.0027 0.0118 0.8204
a 0.0530⁎ 0.0217 0.0147 0.0516⁎ 0.0122 0.0000 0.0515⁎ 0.0219 0.0184
ω 0.0724⁎ 0.0127 0.0000 0.0762⁎ 0.0133 0.0000 0.0684⁎ 0.0119 0.0000
α 0.3830⁎ 0.0544 0.0000 0.3238⁎ 0.0341 0.0000 0.5331⁎ 0.0740 0.0000
β 0.6467⁎ 0.0420 0.0000 0.6993⁎ 0.0305 0.0000 0.6127⁎ 0.0384 0.0000
γ −0.2040⁎ 0.0552 0.0002 −0.2123⁎ 0.0612 0.0005 −0.2936⁎ 0.0778 0.0002
δ 1.6005⁎ 0.2182 0.0000
λ 3.5810⁎ 0.2679 0.0000 3.5293⁎ 0.2594 0.0000 3.7262⁎ 0.2594 0.0000

Commodity
μ −0.0247⁎ 0.0082 0.0027 −0.0251⁎ 0.0113 0.0267 −0.0255⁎ 0.0112 0.0223
a 0.0120 0.0161 0.4551 0.0105 0.0235 0.6540 0.0066 0.0210 0.7545
ω 0.0085 0.0185 0.6443 0.0099 0.0079 0.2131 0.0216⁎ 0.0074 0.0037
α 0.0753 0.0744 0.3113 0.0809⁎ 0.0310 0.0091 0.0694⁎ 0.0353 0.0493
β 0.9334⁎ 0.0801 0.0000 0.9266⁎ 0.0329 0.0000 0.8433⁎ 0.0379 0.0000
γ 0.1116 0.2250 0.6198 0.1482 0.0964 0.1241 0.1065⁎ 0.0345 0.0020
δ 0.8435 0.5438 0.1209
λ 4.8594⁎ 0.5861 0.0000 4.8870⁎ 0.5821 0.0000 4.9405⁎ 0.6097 0.0000

Energy
μ 0.0385 0.0240 0.1085 0.0439⁎ 0.0212 0.0389 0.0380 0.0239 0.1119
a 0.0343 0.0203 0.0906 0.0396 0.0210 0.0596 0.0339 0.0202 0.0930
ω 0.0351⁎ 0.0149 0.0188 0.0271⁎ 0.0124 0.0288 0.0387⁎ 0.0144 0.0071
α 0.0635⁎ 0.0144 0.0000 0.0737⁎ 0.0155 0.0000 0.0374⁎ 0.0116 0.0013
β 0.9204⁎ 0.0196 0.0000 0.9241⁎ 0.0193 0.0000 0.9190⁎ 0.0194 0.0000
γ 0.2031⁎ 0.0795 0.0107 0.2660⁎ 0.0972 0.0062 0.0449⁎ 0.0197 0.0226
δ 1.7558⁎ 0.2519 0.0000
λ 6.4775⁎ 0.9309 0.0000 6.3343⁎ 0.8804 0.0000 6.4766⁎ 0.9326 0.0000

Green bond
μ 0.0078 0.0066 0.2395 0.0087 0.0072 0.2302 0.0078 0.0066 0.2413
a −0.0295 0.0195 0.1304 −0.0317 0.0201 0.1144 −0.0295 0.0195 0.1308
ω 0.0006 0.0004 0.1418 0.0021⁎ 0.0009 0.0184 0.0006⁎ 0.0002 0.0154
α 0.0445⁎ 0.0138 0.0013 0.0659⁎ 0.0087 0.0000 0.0371⁎ 0.0067 0.0000
β 0.9532⁎ 0.0038 0.0000 0.9452⁎ 0.0069 0.0000 0.9535⁎ 0.0021 0.0000
γ 0.0722 0.0724 0.3186 0.0341 0.0911 0.7078 0.0129 0.0116 0.2654
δ 1.9508⁎ 0.5246 0.0002
λ 6.9656⁎ 0.9112 0.0000 6.5308⁎ 0.8216 0.0000 6.9760⁎ 0.9207 0.0000

DCC coefficients
φ1 0.0133⁎ 0.0024 0.0000 0.0117⁎ 0.0024 0.0000 0.0138⁎ 0.0024 0.0000
φ2 0.9676⁎ 0.0071 0.0000 0.9710⁎ 0.0068 0.0000 0.9660⁎ 0.0072 0.0000
λ 7.2504⁎ 0.4344 0.0000 7.1015⁎ 0.4170 0.0000 7.4472⁎ 0.4627 0.0000

Information criteria
AIC 13.284 13.303 13.292
BIC 13.283 13.303 13.291
HQ 13.330 13.345 13.334
Log-L −16,485 −16,514 −16,500

Notes: the table provides the estimation results of three dynamic hedge ratio models, i.e., the DCC-APGARCH, DCC-T-GARCH, DCC-GJR-GARCH models for the
carbon futures returns and the four market index returns. The four market indices include the VIX futures index, the commodity futures index, the energy index, and
the green bond index. AIC is the Akaike information criterion, BIC is the Bayesian information criterion, HQ is the Hannan-Quinn information criterion, and Log-L is
the log-likelihood value.
⁎
Denotes the significance at the 5% level.

4.3. Dynamic conditional correlations between carbon futures returns and correlations are produced. We find that the dynamic conditional correla-
market index returns tions of the carbon and VIX pair fluctuate between a wide range of −0.44
to 0.09, with an average value that is not significantly different from zero.
In Table 7, we report the one-step-ahead dynamic conditional corre- The same characteristics are found in the conditional correlations of the
lations obtained from the dynamic hedge models. The dynamic hedge carbon and commodity pair. The low conditional correlations suggest that
models are refit every 20 observations. The estimation window is fixed at the stock and commodity markets are decoupled from the carbon futures
1490 observations and 1000 one-step-ahead dynamic conditional market, and their potential for hedging carbon risk is limited.

9
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

Table 6
Diagnostic tests of three dynamic hedge ratio models.
Carbon VIX Commodity Energy Green bond

DCC-APGARCH Q(30) 38.2600 21.7800 20.9200 32.9600 30.7500

p-Value 0.1400 0.8600 0.8900 0.3200 0.4300
DCC-T-GARCH Q(30) 39.7200 22.4900 20.0100 33.9500 31.6900
p-Value 0.1100 0.8400 0.9200 0.2800 0.3800
DCC-GJR-GARCH Q(30) 39.7200 22.4900 20.0100 33.9500 31.6900
p-Value 0.1100 0.8400 0.9200 0.2800 0.3800

Notes: the table reports the Ljung-Box test statistics on the standard residuals to evaluate the goodness of fit among three dynamic hedge ratio models, i.e., the DCC-
APGARCH, DCC-T-GARCH, and DCC-GJR-GARCH models. Q(30) is the Ljung-Box test statistic on the standard residuals at 30 lags.

Table 7 effectiveness of various hedging strategies.

Summary statistics of dynamic conditional correlations.
VIX Commodity Energy Green bond 4.4. Optimal hedge ratios and hedging effectiveness

DCC-APGARCH Mean −0.0720 −0.0370 0.0600 0.2390 In this subsection, we examine the implications of the results in
Max 0.0880 0.0570 0.2550 0.3930
previous sections for hedging strategies and portfolio risk management.
Min −0.4430 −0.2080 −0.0840 0.0620
SD 0.0850 0.0520 0.0620 0.0580 Building on modern portfolio theory, Johnson (1960) proposed the
DCC-T-GARCH Mean −0.0690 −0.0390 0.0600 0.2390 theory of minimum variance hedge ratio to minimize the risk associated
Max 0.0850 0.0590 0.2680 0.3850 with a portfolio. Suppose that we use asset j as a hedge for asset i. The
Min −0.4110 −0.2130 −0.0750 0.0720 return on a hedged portfolio including one dollar of asset i and γj, t
SD 0.0810 0.0530 0.0620 0.0590
DCC-GJR-GARCH Mean −0.0740 −0.0440 0.0580 0.2400
dollars of asset j can be written as
Max 0.0850 0.0440 0.2500 0.3830 RH , t = Ri, t j, t Rj, t , (9)
Min −0.4250 −0.2310 −0.0670 0.0650
SD 0.0820 0.0570 0.0590 0.0580 where Ri, t and Rj, t are the returns on assets i and j, and γj, t is the hedge
ratio.
Notes: the table reports the one-step ahead dynamic conditional correlations
According to Johnson (1960) and Baillie and Myers (1991), the
between the carbon futures returns and the returns of four market indices based
on three dynamic hedge ratio models, i.e., the DCC-APGARCH, DCC-T-GARCH, optimal hedge ratio can be obtained by partially diverting the variance
and DCC-GJR-GARCH models. The four market indices include the VIX futures of the hedged portfolio with respect to γj, t and setting the expression to
index, the commodity futures index, the energy index, and the green bond zero. We then have
index. The estimation window is fixed at 1490 observations and 1000 one-step-
cov (Ri, t , Rj, t )
ahead dynamic conditional correlations are produced. The dynamic hedge ratio j, t = .
models are refit every 20 observations. var (Rj, t ) (10)
In this study, we are interested in the rate at which a long position of
In Table 7, we find positive average conditional correlations for the one dollar in carbon futures can be hedged by a short position in a
carbon and energy pair and for the carbon and green bond pair, sug- market index. Therefore, the optimal hedge ratio is given by
gesting the potential of the energy and/or green bond index to hedge
carbon market risk. However, the conditional correlations of the carbon cov (R carbon, t , Rindex , t )
= ,
(11)
index , t
and energy pair fluctuates around a positive value of 0.06 with the var (Rindex, t )
occurrence of several negative values during the sample period, which
where var(Rindex, t) and cov(Rcarbon, t, Rindex, t) are conditional on in-
complicates the design of risk management strategies. Among the four
formation available up to time t-1. According to Kroner and Sultan
market indices, the green bond index is the only index that has positive
(1993) and Basher and Sadorsky (2016), the variance and covariance
conditional correlations over the whole sample period. With an average
forecasts can be obtained from the GARCH models, and the dynamic
conditional correlation as high as 0.24, the green bond index serves as a
optimal hedge ratios can be calculated and used to design dynamic
reliable instrument for the implementation of hedging strategies.
hedging strategies. We conduct a rolling window analysis to estimate
We now turn to the evolution of the dynamic conditional correla-
one-step ahead conditional variance and covariance forecasts, and then
tions between the carbon futures returns and the four market index
use these forecasts to calculate one-step ahead dynamic optimal hedge
returns. Fig. 3 plots the dynamic conditional correlations calculated by
ratios.
the DCC-APGARCH model. Over the whole sample period, there are no
For comparison purposes, we also construct constant hedge ratios
significant increases or decreases in the level of conditional correlations
from the “naive” OLS model that is still widely used in the hedging
for the carbon and VIX pair. The conditional correlations for the carbon
literature. The model is actually a regression specified as
and commodity pair show very similar patterns. By contrast, we find
obvious increases in the level of conditional correlations for the carbon
OLS
R carbon, t = index Rindex , t + carbon, t , (12)
and green bond pair and for the carbon and energy pair during the 2
where εcarbon, t~N(0, σ ). Therefore, the optimal hedge ratio can be
carbon market's volatile period. The average conditional correlation
calculated as
between the carbon futures returns and the green bond index returns is
0.32 before 2013, while it becomes 0.20 after 2013; the average con- OLS
=
carbon, index
,
(13)
index 2
ditional correlation between the carbon futures returns and the energy index
index returns is 0.14 before 2013, while it becomes 0.03 after 2013. 2
where σindex is the variance of index returns and σcarbon, index is the
These results suggest that the assumption of constant correlations is not
covariance between the carbon futures returns and index returns. In
appropriate.
practice, we refit the model every 20 observations. The advantage of
The analysis of dynamic conditional correlations sets the stage for
this method from a hedging perspective is that by updating the in-
the next subsection, in which we compare dynamic optimal hedge ra-
formation set we still obtain a more efficient estimate of the hedge
tios with constant optimal hedge ratios, and we evaluate the hedging
ratio, which takes time variation in the return distribution into account.

10
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

Fig. 3. Dynamic conditional correlations.

Notes: the figure plots the evolution of the dynamic conditional correlations between the carbon futures returns and the four market index returns calculated by the
DCC-APGARCH model. The four market indices include the VIX futures index, the commodity futures index, the energy index, and the green bond index.

We are interested in whether dynamic hedging strategies are econom- The summary statistics of the dynamic optimal hedge ratios pro-
ically distinguishable from the “naive” constant hedging strategies in duced by the DCC-APGARCH, DCC-T-GARCH, and DCC-GJR-GARCH
terms of providing higher hedging performance. models and the constant optimal hedge ratios produced by the OLS
We follow Ku, Chen, and Chen (2007) to compare the performance model are reported in Table 8. HE values are also reported. We find that
of different hedging strategies by comparing the variance of the hedged the green bond index is the best hedge against carbon market risk as it
portfolio to the variance of the unhedged portfolio and calculate the has the largest HE value among all the market indices no matter which
hedging effectiveness (HE) value as hedge model is used. On average, the HE value for a portfolio composed
of carbon futures and the green bond index based on dynamic hedge
varunhedged varhedged
HE = . models is 6.05 and the corresponding optimal hedge ratio is 0.03. From
varunhedged (14) an investor's perspective, about 6% of the return variance of a one-
dollar long position in carbon futures can be hedged by shorting 3 cents
Here, the unhedged portfolio includes one dollar of carbon futures,
of the green bond index.
while the hedged portfolio includes a long position of one dollar in
As shown in Table 8, all other market indices except for the green
carbon futures and a short position of γindex, t∗(γindexOLS) dollars in a
bond index are associated with some negative HE values from one or
market index. By construction, a larger HE value indicates better hed-
more hedge models. A negative HE value implies that the variance of
ging effectiveness.
the hedged portfolio is greater than that of the unhedged position, and
In essence, the calculated optimal hedge ratio can be positive or
it is thus unattractive to investors who attempt to minimize portfolio
negative. Referring to Eq. (9), a positive optimal hedge ratio indicates
risks. We also notice that, the average hedge ratios of the VIX and
that a long position of one dollar in carbon futures can be hedged by a
commodity indices are both negative across all the hedge models, im-
short position of γindex, t∗(γindexOLS) dollars in a market index, while a
plying that the risk in the carbon market could be diversified by pur-
negative optimal hedge ratio indicates that a long position of one dollar
chasing the VIX or commodity index, or to say by constructing a carbon
in carbon futures can be diversified by a long position of γindex, t∗
and VIX portfolio or a carbon and commodity portfolio. This finding is
(γindexOLS) dollars in a market index. When γindex, t∗(γindexOLS) is negative,
consistent with the result in Section 4.3 that the VIX and commodity
the evaluation of hedging effectiveness can be considered as the eva-
index returns have low correlations with the carbon futures returns.
luation of diversification performance.

11
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

Table 8 academics during volatile periods because hedging instruments have

Summary statistics of optimal hedge ratios and HE values. the most value during these periods. This motivates us to examine
Optimal hedge ratio HE (%) hedging performance in the market's volatile period.

Mean Max Min SD 4.5. Hedging effectiveness during volatile periods

DCC-APGARCH
VIX −0.0280 0.0780 −1.0870 0.0650 −1.7220 Our sample period covers the deep and lasting financial crises in the
Commodity −0.0110 0.0210 −0.0670 0.0150 0.7600 European Union and around the world. From the beginning of 2009,
Energy 0.0260 0.1400 −0.1380 0.0290 −0.0550 shortly after the collapse of Lehman brothers, the 2007 subprime crisis
Green bond 0.0340 0.1170 0.0050 0.0240 6.1170 in the U.S. has become a global financial crisis and it finally ended in
DCC-T-GARCH December 2010. Since April 1, 2010, the European debt crisis has be-
VIX −0.025 0.0500 −0.419 0.0450 0.2030 come a key focal point for the world markets and it finally ended in
Commodity −0.0110 0.0130 −0.064 0.0150 0.8090
December 31, 2012. Our sample period also covers different regulatory
Energy 0.0260 0.1400 −0.103 0.0280 0.0050
Green bond 0.0350 0.1180 0.0060 0.0250 6.1940 periods (Phase II and Phase III) which have been implemented within
the EU ETS. During 2012 to 2013, the carbon futures price experienced
DCC-GJR-GARCH
VIX −0.0270 0.0560 −0.4990 0.0470 0.2650
a sharp decline from 26 euro per ton of carbon dioxide to 5 euro. While
Commodity −0.0120 0.0200 −0.0550 0.0150 0.8360 some studies argue that the EUA price drop can be explained by the EU
Energy 0.0250 0.1430 −0.1240 0.0270 −0.091 ETS regulatory changes which introduced different trading schemes,
Green bond 0.0350 0.1130 0.0060 0.0240 5.8480 others (e.g., Aldy & Stavins, 2012) contend that the financial crises are
OLS responsible for the sharp drop in the EUA price. Thus, it is probable that
VIX −0.0240 −0.0100 −0.0350 0.0080 0.3710 the regulatory changes and financial crises may have led to structural
Commodity −0.0080 0.0000 −0.0120 0.0030 −0.0140
changes during our sample period.
Energy 0.0150 0.0350 0.0020 0.0090 0.0590
Green bond 0.0280 0.0510 0.0140 0.0090 4.0410 To identify structural breaks, we adopt the one-break minimum
Lagrange Multiplier (LM) unit root test proposed by Lee and Strazicich
Notes: the table reports the summary statistics of optimal hedge ratios and HE (2013). Unlike many earlier endogenous break tests that tend to iden-
values produced by three dynamic hedge models, i.e., the DCC-APGARCH, DCC- tify the break point one period prior to the true break point and thereby
T-GARCH, and DCC-GJR-GARCH models, and the OLS model. The optimal lead to greatest spurious rejections and maximized bias in the estima-
hedge ratio (γindex, t∗ or γindexOLS) is the rate at which a long position of one tion of the persistence parameter, Lee and Strazicich (2013) develop an
dollar in the carbon futures can be hedged by taking a short/long position in a
varunhedged varhedged endogenous one-break unit root test that is free of bias and spurious
market index. The HE value is calculated as HE = . The un-
var
unhedged rejections under the null and alternative hypotheses. As emphasized by
hedged portfolio includes one dollar of the carbon futures, while the hedged Lee and Strazicich (2013), there are many situations where a one-break
portfolio includes a long position of one dollar in the carbon futures and a long/ test is preferred since including unnecessary breaks can lead to less
short position of γindex, t∗(γindexOLS) dollars in a market index. The hedge models statistical power. We conduct the one-break minimum LM unit root test
are refit every 20 observations. For the dynamic hedge models, one-step ahead using the carbon futures return data and find a structural break point on
conditional variance and covariance forecasts are estimated and these forecasts
April 23, 2013. This time point is near the end of the sharp drop in the
are used to calculate one-step ahead dynamic optimal hedge ratios. For the OLS
carbon futures price, and is just at the end of the European debt crisis
model, one constant optimal hedge ratio is estimated for every 20 days.
and EUA Phase II. We also conduct the one-break minimum LM unit
root test using the other four market index returns and the results are
However, using the VIX and commodity indices to diversify carbon
reported in Table 9. The structural break points in the four market
market risk provides average HE values of −0.50 and 0.80, respec-
index returns are close to the one detected in the carbon futures returns.
tively, which are much lower than the HE value obtained by using the
Specifically, there is a structural break point on August 17, 2012 for the
green bond index as a hedge (6.05 on average). Therefore, the green
VIX index, on May 30, 2014 for the energy index, on December 10,
bond index is the most effective hedging instrument for carbon risk.
2012 for the commodity index, and on December 15, 2009 for the green
Further, we find that the HE values based on the dynamic hedge
bond index.
models are higher than those based on the OLS model for the carbon
Using the structural break point found in the carbon futures returns
and green bond portfolio and for the carbon and commodity portfolio.
as a cutoff, we separate the full sample period into two sub-periods. The
The evidence indicates that when the green bond index is used to hedge
first sub-period is from December 01, 2008 to April 23, 2013, which
carbon market risk, the dynamic hedge strategy is superior to the
covers the relatively volatile period of the 2008 global financial crisis,
constant hedge strategy. This is not surprising as dynamic hedge ratio
the 2010 European sovereign debt crisis, and the drastic decline in the
models can capture more important “stylized facts” that we have de-
EUA price from 2012 to 2013. The second sub-period is from April 24,
tected in the preliminary analysis, such as dynamic correlation and
2013 to August 31, 2018, which covers the relatively tranquil period.
volatility spillover for the carbon futures returns and the green bond
The summary statistics of the optimal hedge ratios and HE values for
index returns.
the two sub-periods are reported in Table 10. We find that the green
We further plot the evolution of the dynamic optimal hedge ratios
bond index as the hedging instrument has the largest HE value among
based on the DCC-APGARCH model in Fig. 4. As shown, the dynamic
all the market indices, in both the turmoil and tranquil periods and
optimal hedge ratios of the VIX index are relatively stable with the
across all the dynamic hedge models. These results provide strong
occurrence of several extremely negative values around June 2016,7
evidence that the green bond index is the best hedge for carbon futures
while the dynamic optimal hedge ratios of all the other three market
and performs well even in the crisis period.
indices exhibit a deceasing trend, likely due to the occurrence and re-
Based on the dynamic hedge models, we find that the green bond
covery of financial crises in the first half of the sample period. As em-
index is the only market index that consistently produces positive op-
phasized by Junttila, Pesonen, and Raatikainen (2018), risk manage-
timal hedge ratios over the crisis period. When the optimal hedge ratio
ment and asset allocation are very important to practitioners and
is positive, a short position in the market index should be added to the
portfolio, while a long position in the market index should be added to
7
It is the strong negative correlation between the carbon futures returns and the portfolio when the optimal ratio is negative. The feature that the
the VIX index returns during a long period before June 2016 that leads to the optimal hedge ratios of the green bond index stay positive over the
negative covariance forecast and thus the negative optimal hedge ratio. crisis period makes green bonds an ideal hedge in terms of simplicity in

12
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

Fig. 4. Dynamic optimal hedge ratios.

Notes: the figure plots the evolution of the dynamic optimal hedge ratios based on the DCC-APGARCH model. The optimal hedge ratio is the rate at which a long
position of one dollar in carbon futures can be hedged by a short position in a market index. The four market indices include the VIX futures index, the commodity
futures index, the energy index, and the green bond index.

Table 9 carbon futures produces a HE value of only 0.34. As pointed out by

Results of the one-break minimum LM unit root test. Chen and Sutcliffe (2012), the average position in the hedging instru-
TB K τ
ment, or the average hedge ratio, can be considered as a proxy for the
transaction cost. The transaction cost per unit of hedging effectiveness
Carbon 2013/4/23 3 −24.8753⁎ (TC/HE) can therefore be considered as the tradeoff between transac-
VIX 2012/8/17 8 −14.0770⁎ tion costs and hedging effectiveness. As the TC/HE ratio of the green
Commodity 2012/12/10 7 −15.3286⁎ 0.023
Energy 2014/5/30 7 −15.2459⁎
bond index ( 5.84 = 0.004 ) is much lower than that of the energy index
0.024
Green bond 2009/12/15 7 −4.6588⁎ ( 0.34 = 0.071), the green bond index is not only an effective but also an
inexpensive hedge for carbon risk.
Notes: the table reports the results of the one-break minimum LM unit root test
Comparing the transaction costs in the crisis and tranquil periods,
for the time series of daily returns of the carbon futures, the VIX futures index,
we find that a one-dollar investment in carbon futures can be hedged by
the commodity futures index, the energy index, and the green bond index. TB is
a short position of 5.5 cents in the green bond index in the crisis period,
the time period of the structural break, K is the lag order, and τ is the one-break
minimum LM unit root test statistic. The lag order is chosen following a general- in contrast to a short position of 2.3 cents in the tranquil period.
to-specific approach as suggested by Lee and Strazicich (2013). The 5% critical Considering the tradeoff between transaction costs and hedging effec-
value is −3.2300. tiveness, the TC/HE ratio of green bond index in the crisis period
0.055 0.023
⁎
Indicates significance at the 5% level. ( 1.16 = 0.047 ) is larger than that in the tranquil period ( 5.84 = 0.004 ),
indicating that a more costly position in the green bond index is re-
the design of hedging strategies. quired in order to hedge carbon market risk during the crisis period.
In the tranquil period, taking a short position of about 2.3 cents in When evaluated in the crisis period, the OLS model produces low
the green bond index to hedge a one-dollar investment in carbon fu- negative HE values in all the hedged portfolios and thus its hedging
tures produces a HE value of about 5.84, while taking almost the same performance is inferior to that of the dynamic hedge models. As the
position in the energy index to hedge the one-dollar investment in more parsimonious OLS model fails to capture the complex price

13
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

Table 10
Summary statistics of optimal hedge ratios and HE values for two sub-periods.
2008/12/1–2013/4/23 2013/4/24–2018/8/31

Optimal hedge ratio HE (%) Optimal hedge ratio HE (%)

Mean Max Min SD Mean Max Min SD

DCC-APGARCH
VIX −0.0300 0.0020 −0.1390 0.0200 −2.3440 −0.0110 0.0170 −0.0640 0.0100 −0.4230
Commodity −0.0010 0.0350 −0.0290 0.0120 −0.1760 −0.0060 0.0130 −0.0390 0.0080 −0.4360
Energy 0.0470 0.1200 −0.0190 0.0330 −3.4060 0.0240 0.0640 0.0000 0.0130 0.4050
Green bond 0.0540 0.1190 0.0040 0.0280 1.2520 0.0230 0.0390 0.0060 0.0070 5.9180

DCC-T-GARCH
VIX −0.0290 0.0030 −0.1500 0.0210 −2.4440 −0.0110 0.0170 −0.0750 0.0100 −0.4330
Commodity −0.0010 0.0320 −0.0250 0.0120 −0.1960 −0.0070 0.0100 −0.0350 0.0080 −0.4650
Energy 0.0510 0.1350 −0.0190 0.0360 −3.9250 0.0230 0.0630 −0.0010 0.0130 0.2940
Green bond 0.0560 0.1280 0.0070 0.0290 1.0590 0.0230 0.0400 0.0060 0.0070 5.6550

DCC-GJR-GARCH
VIX −0.0280 0.0030 −0.1340 0.0190 −2.1860 −0.0100 0.0180 −0.0720 0.0110 −0.4580
Commodity −0.0010 0.0340 −0.0360 0.0120 −0.1490 −0.0060 0.0140 −0.0420 0.0080 −0.4790
Energy 0.0490 0.1220 −0.0160 0.0330 −3.6820 0.0240 0.0640 −0.0010 0.0130 0.3470
Green bond 0.0540 0.1190 0.0040 0.0270 1.1940 0.0230 0.0390 0.0060 0.0070 5.9270

OLS
VIX −0.0730 −0.0450 −0.1050 0.0150 −16.8260 −0.0240 −0.0200 −0.0260 0.0020 −1.3810
Commodity 0.0800 0.1340 0.0390 0.0230 −8.2860 0.0130 0.0220 0.0050 0.0050 0.5350
Energy 0.0800 0.1340 0.0390 0.0230 −8.2860 0.0130 0.0220 0.0050 0.0050 0.5350
Green bond 0.0860 0.1450 0.0500 0.0210 −45.2470 0.0190 0.0220 0.0160 0.0020 5.5650

Notes: the table reports the summary statistics of optimal hedge ratios and HE values produced by three dynamic hedge models, i.e., the DCC-APGARCH, DCC-T-
GARCH, and DCC-GJR-GARCH models, and the OLS model for two sub-periods. The first sub-period is from December 01, 2008 to April 23, 2013, which covers the
relatively volatile period of the 2008 global financial crisis, the 2010 European sovereign debt crisis, and the drastic decline in the EUA price from 2012 to 2013. The
second sub-period is from April 24, 2013 to August 31, 2018, which covers the relatively tranquil period. The optimal hedge ratio (γindex, t∗ or γindexOLS) is the rate at
which a long position of one dollar in the carbon futures can be hedged by taking a short/long position in a market index. The HE value is calculated as
varunhedged varhedged
HE = var
. The unhedged portfolio includes a long position of one dollar in the carbon futures, while the hedged portfolio includes a long position of one
unhedged
dollar in the carbon futures and a long/short position of γindex, t∗(γindexOLS) dollars in a market index.

dynamics in the volatile period, it is likely to cause poor performance management strategy is larger than that produced by using the green
due to high estimation error. bond index as the hedging instrument. These results confirm our finding
that the green bond index is not only an effective but also an in-
expensive hedge for carbon futures.
5. Robustness tests

The HE values in the previous section are obtained under specific 6. Conclusion
assumptions about the forecast horizon and the number of model refits.
This section provides additional information on how robust these values Carbon performance is greatly unstable due to various factors such
are to changes in the forecast horizon and the number of model refits. as climate change, uncertainty in carbon control regulations and green
Table 11 presents the estimation results with the forecast lengths of 800 policies, and geopolitical situations. Serving as a barometer for changes
and 1200 days, and Table 12 reports the estimation results with the in these factors, the green bond market index is closely related to the
hedge models refit every 10 and 30 days. As shown, the HE values are carbon market. Meanwhile, green bonds have been raising social
fairly similar across different forecast horizons and model refits. The awareness among investors who are considering not only benefiting
green bond index as the hedging instrument has the largest HE value from their investment objectives but also adding values to the climate
among all the market indices based on all the hedge models. When the change or environmental themes. The search for such investments may
green bond index is used to hedge carbon market risk, the dynamic have led to increased popularity of green bonds and contributed to
hedge strategy has better performance than the constant hedge enhanced correlation between the green bond market and the carbon
strategy. market.
We further report the summary statistics of the optimal hedge ratios Our study explores the hedging effect of green bonds on carbon
and HE values based on multi-step ahead forecasts in both the volatile market risk. For comparison purposes, we examine the relationship
and tranquil periods in Table 13. The multi-step ahead forecasts can be between the carbon futures returns and the returns of four major
useful because they would provide investors with flexibility in de- market indices, i.e., the VIX index, the commodity index, the energy
signing appropriate hedging strategies. As shown, the green bond index index, and the green bond index. We find that the connectedness be-
still has the largest HE value among all the market indices, based on tween the carbon futures returns and the green bond index returns is
both the 20- and 30-step ahead forecasts, in both the turmoil and the highest and this connectedness is extremely pronounced during the
tranquil periods, and across all the dynamic hedge models. These market's volatile period. Further, we employ three dynamic hedge
findings confirm that the green bond index is the best hedge for carbon models (DCC-APGARCH, DCC-T-GARCH, and DCC-GJR-GARCH
risk and still performs well in the volatile period. As implied by the models) to outline dynamic hedging strategies and compare their
positive HE value associated with the energy index in the tranquil hedging performance with that of the constant hedge model (OLS
period, another robust result is that investors can seek hedge oppor- model). We find that the green bond index is an effective and in-
tunities by investing in a portfolio composed of carbon futures and the expensive hedging instrument for carbon risk. Among the four market
energy index. However, the TC/HE ratio produced by this risk indices, the green bond index as the hedging instrument has the largest

14
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

Table 11
Summary statistics of optimal hedge ratios and HE values from different forecast lengths.
Forecast length = 800 Forecast length = 1200

Optimal hedge ratio HE (%) Optimal hedge ratio HE (%)

Mean Max Min SD Mean Max Min SD

DCC-APGARCH
VIX −0.0240 0.0450 −0.5770 0.0460 −0.3100 −0.0260 0.0720 −1.0800 0.0590 −1.8470
Commodity −0.0090 0.0230 −0.0710 0.0150 0.7790 −0.0090 0.0180 −0.0550 0.0140 0.4970
Energy 0.0250 0.0900 −0.1070 0.0230 0.2620 0.0250 0.1530 −0.1420 0.0300 0.0670
Green bond 0.0320 0.1240 0.0080 0.0250 4.6210 0.0310 0.1110 0.0060 0.0230 5.4060

DCC-T-GARCH
VIX −0.0230 0.0490 −0.4520 0.0420 −0.1150 −0.0230 0.0460 −0.4270 0.0410 0.0660
Commodity −0.0090 0.0220 −0.0570 0.0150 0.8160 −0.0100 0.0160 −0.0620 0.0140 0.6750
Energy 0.0260 0.0950 −0.0860 0.0230 0.2820 0.0250 0.1410 −0.1080 0.0280 0.0650
Green bond 0.0320 0.1280 0.0080 0.0250 4.6890 0.0310 0.1110 0.0060 0.0230 5.5440

DCC-GJR-GARCH
VIX −0.0260 0.0510 −0.5390 0.0460 −0.0660 −0.0260 0.0720 −0.4980 0.0430 0.1030
Commodity −0.0100 0.0340 −0.0560 0.0160 0.9100 −0.0110 0.0180 −0.0610 0.0150 0.7040
Energy 0.0250 0.0870 −0.1070 0.0230 0.2360 0.0240 0.1480 −0.1410 0.0290 0.0070
Green bond 0.0320 0.1190 0.0070 0.0240 4.5320 0.0310 0.1070 0.0050 0.0220 5.1630

OLS
VIX −0.0270 −0.0100 −0.0340 0.0060 0.3620 −0.0220 −0.0110 −0.0350 0.0070 0.4760
Commodity −0.0080 −0.0020 −0.0110 0.0020 −0.2570 −0.0080 0.0010 −0.0140 0.0040 0.0490
Energy 0.0180 0.0340 0.0050 0.0070 0.2730 0.0130 0.0400 −0.0010 0.0100 −0.0580
Green bond 0.0300 0.0510 0.0170 0.0080 2.6760 0.0260 0.0570 0.0130 0.0100 1.2320

Notes: the table reports the summary statistics of optimal hedge ratios and HE values produced by three dynamic hedge models, i.e., the DCC-APGARCH, DCC-T-
GARCH, and DCC-GJR-GARCH models, and the OLS model from different forecast lengths. The optimal hedge ratio (γindex, t∗ or γindexOLS) is the rate at which a long
varunhedged varhedged
position of one dollar in the carbon futures can be hedged by taking a short/long position in a market index. The HE value is calculated as HE = var
.
unhedged
The unhedged portfolio includes a long position of one dollar in the carbon futures, while the hedged portfolio includes a long position of one dollar in the carbon
futures and a long/short position of γindex, t∗(γindexOLS) dollars in a market index.

Table 12
Summary statistics of optimal hedge ratios and HE values from different model refits.
Refit every 10 observations Refit every 30 observations

Optimal hedge ratio HE (%) Optimal hedge ratio HE (%)

Mean Max Min SD Mean Max Min SD

DCC-APGARCH
VIX −0.0280 0.0780 −1.0870 0.0650 −1.7350 −0.0280 0.0800 −0.9800 0.0620 −1.3300
Commodity −0.0110 0.0210 −0.0670 0.0150 0.7240 −0.0110 0.0220 −0.0700 0.0150 0.7690
Energy 0.0260 0.1400 −0.1380 0.0290 −0.0770 0.0260 0.1400 −0.1290 0.0280 −0.0410
Green bond 0.0340 0.1170 0.0050 0.0250 6.1480 0.0340 0.1140 0.0050 0.0240 6.1070

DCC-T-GARCH
VIX −0.0250 0.0500 −0.4190 0.0450 0.2320 −0.0250 0.0490 −0.4220 0.0450 0.2000
Commodity −0.0110 0.0130 −0.0640 0.0150 0.7980 −0.0110 0.0130 −0.0640 0.0150 0.8070
Energy 0.0260 0.1410 −0.1030 0.0280 −0.0150 0.0260 0.1400 −0.0970 0.0280 0.0070
Green bond 0.0350 0.1180 0.0060 0.0250 6.1850 0.0350 0.1150 0.0060 0.0250 6.1990

DCC-GJR-GARCH
VIX −0.0270 0.0600 −0.4990 0.0470 0.2390 −0.0270 0.0510 −0.4990 0.0470 0.2640
Commodity −0.0120 0.0200 −0.0560 0.0150 0.8340 −0.0120 0.0210 −0.0550 0.0150 0.8420
Energy 0.0250 0.1430 −0.1240 0.0280 −0.1170 0.0250 0.1430 −0.1200 0.0270 −0.0920
Green bond 0.0350 0.1130 0.0060 0.0240 5.8430 0.0350 0.1120 0.0060 0.0240 5.8500

OLS
VIX −0.0230 −0.0100 −0.0350 0.0080 0.3820 −0.0240 −0.0100 −0.0340 0.0080 0.3630
Commodity −0.0080 0.0000 −0.0120 0.0030 −0.0220 −0.0080 −0.0010 −0.0300 0.0030 −0.1430
Energy 0.0150 0.0350 0.0020 0.0090 0.0460 0.0140 0.0350 −0.0690 0.0120 −0.0530
Green bond 0.0270 0.0510 0.0140 0.0090 4.0370 0.0280 0.0710 0.0140 0.0100 3.9060

Notes: the table reports the summary statistics of optimal hedge ratios and HE values produced by three dynamic hedge models, i.e., the DCC-APGARCH, DCC-T-
GARCH, and DCC-GJR-GARCH models, and the OLS model for different model refits. The optimal hedge ratio (γindex, t∗ or γindexOLS) is the rate at which a long position
varunhedged varhedged
of one dollar in the carbon futures can be hedged by taking a short/long position in a market index. The HE value is calculated as HE = var
. The
unhedged
unhedged portfolio includes one dollar of the carbon futures, while the hedged portfolio includes a long position of one dollar in the carbon futures and a long/short
position of γindex, t∗(γindexOLS) dollars in a market index.

15
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

Table 13
Summary statistics of optimal hedge ratios and HE values from multi-step ahead forecasts.
Panel A: 20-step ahead forecasts

2008/12/1–2013/4/23 2013/4/24–2018/8/31

Optimal hedge ratio HE (%) Optimal hedge ratio HE (%)

Mean Max Min SD Mean Max Min SD

DCC-APGARCH
VIX −0.0300 0.0010 −0.1450 0.0210 −2.4390 −0.0100 0.0180 −0.0720 0.0110 −0.4320
Commodity 0.0000 0.0330 −0.0250 0.0120 −0.2790 −0.0060 0.0140 −0.0410 0.0080 −0.5300
Energy 0.0500 0.1370 −0.0190 0.0370 −3.9400 0.0240 0.0640 −0.0010 0.0120 0.3500
Green bond 0.0560 0.1250 0.0070 0.0290 0.8740 0.0230 0.0390 0.0060 0.0070 5.9150

DCC-T-GARCH
VIX −0.0300 0.0000 −0.1320 0.0200 −2.2530 −0.0100 0.0170 −0.0640 0.0110 −0.3950
Commodity 0.0000 0.0350 −0.0250 0.0120 −0.2330 −0.0060 0.0130 −0.0350 0.0080 −0.4850
Energy 0.0460 0.1170 −0.0190 0.0330 −3.5750 0.0240 0.0640 0.0000 0.0120 0.4040
Green bond 0.0540 0.1120 0.0040 0.0270 1.1110 0.0230 0.0390 0.0060 0.0070 5.8970

DCC-GJR-GARCH
VIX −0.0280 0.0010 −0.1260 0.0190 −2.1380 −0.0110 0.0170 −0.0750 0.0100 −0.4060
Commodity −0.0010 0.0340 −0.0330 0.0120 −0.2030 −0.0070 0.0100 −0.0330 0.0080 −0.4910
Energy 0.0490 0.1190 −0.0160 0.0330 −3.7880 0.0230 0.0630 −0.0010 0.0120 0.3030
Green bond 0.0540 0.1120 0.0040 0.0270 1.0850 0.0230 0.0400 0.0060 0.0070 5.6520

OLS
VIX −0.0740 −0.0480 −0.1050 0.0150 −18.7150 −0.0230 0.0160 −0.0260 0.0070 −1.1490
Commodity −0.0020 0.0160 −0.1120 0.0210 −9.5540 −0.0100 0.0330 −0.0130 0.0070 −0.5190
Energy 0.0900 0.3780 0.0410 0.0540 −68.6390 0.0180 0.1900 0.0050 0.0300 0.1850
Green bond 0.0900 0.1450 0.0550 0.0220 −95.1130 0.0210 0.0900 0.0160 0.0120 4.0460

Panel B: 30-step ahead forecasts

2008/12/1–2013/4/23 2013/4/24–2018/8/31

Optimal hedge ratio HE (%) Optimal hedge ratio HE (%)

Mean Max Min SD Mean Max Min SD

DCC-APGARCH
VIX −0.0300 0.0000 −0.1320 0.0200 −2.3460 −0.0100 0.0180 −0.0760 0.0110 −0.4150
Commodity −0.0010 0.0320 −0.0290 0.0120 −0.1910 −0.0060 0.0140 −0.0420 0.0080 −0.4480
Energy 0.0480 0.1170 −0.0100 0.0320 −3.4550 0.0240 0.0630 −0.0010 0.0120 0.3020
Green bond 0.0550 0.1120 0.0050 0.0270 0.8030 0.0230 0.0390 0.0060 0.0070 5.9180

DCC-T-GARCH
VIX −0.0300 0.0010 −0.1450 0.0210 −2.5850 −0.0100 0.0170 −0.0670 0.0110 −0.3930
Commodity −0.0010 0.0390 −0.0260 0.0120 −0.2760 −0.0050 0.0120 −0.0390 0.0080 −0.3910
Energy 0.0520 0.1370 −0.0100 0.0350 −3.8350 0.0240 0.0630 −0.0010 0.0120 0.3530
Green bond 0.0570 0.1250 0.0060 0.0290 0.6210 0.0230 0.0390 0.0060 0.0070 5.9150

DCC-GJR-GARCH
VIX −0.0280 0.0010 −0.1260 0.0190 −2.1790 −0.0100 0.0170 −0.0770 0.0110 −0.3910
Commodity −0.0010 0.0310 −0.0360 0.0120 −0.2080 −0.0070 0.0100 −0.0350 0.0080 −0.4340
Energy 0.0500 0.1190 −0.0050 0.0320 −3.7920 0.0230 0.0630 −0.0010 0.0120 0.2640
Green bond 0.0550 0.1120 0.0040 0.0270 0.8840 0.0230 0.0400 0.0060 0.0070 5.6720

OLS
VIX −0.0730 −0.0200 −0.1050 0.0190 −15.7820 −0.0300 −0.0210 −0.1310 0.0250 −2.3700
Commodity −0.0030 0.0160 −0.0680 0.0180 −3.4320 −0.0130 −0.0090 −0.0450 0.0080 −1.8380
Energy 0.0920 0.2180 0.0440 0.0380 −29.3210 0.0140 0.0420 0.0050 0.0080 0.5440
Green bond 0.0900 0.1450 0.0600 0.0220 −56.4320 0.0220 0.0760 0.0160 0.0130 4.3830

Notes: the table reports the summary statistics of optimal hedge ratios and HE values produced by three dynamic hedge models, i.e., the DCC-APGARCH, DCC-T-
GARCH, and DCC-GJR-GARCH models, and the OLS model based on multi-step ahead forecasts in two sub-periods. The first sub-period is from December 01, 2008 to
April 23, 2013, which covers the relatively volatile period of the 2008 global financial crisis, the 2010 European sovereign debt crisis, and the drastic decline in the
EUA price from 2012 to 2013. The second sub-period is from April 24, 2013 to August 31, 2018, which covers the relatively tranquil period. The optimal hedge ratio
(γindex, t∗ or γindexOLS) is the rate at which a long position of one dollar in the carbon futures can be hedged by taking a short/long position in a market index. The HE
varunhedged varhedged
value is calculated as HE = var
. The unhedged portfolio includes a long position of one dollar in the carbon futures, while the hedged portfolio includes
unhedged
a long position of one dollar in the carbon futures and a long/short position of γindex, t∗(γindexOLS) dollars in a market index.

HE values based on all the dynamic hedge models. From an investor's models can capture the dynamic correlation and volatility spillover
perspective, about 6% of the return variance of a one-dollar long po- between the carbon futures and market index returns. Based on the
sition in carbon futures can be hedged by shorting 3 cents of the green dynamic hedge model, the green bond index as the hedge instrument
bond index. We further find that the dynamic hedge models are su- can still produce positive HE values in the volatile period while other
perior to the OLS model in the volatile period as more sophisticated market indices fail. A promising direction to explore would be how the

16
J. Jin, et al. International Review of Financial Analysis 71 (2020) 101509

financialization has made the carbon futures market more vulnerable to Hatemi-J, A. (2012). Asymmetric causality tests with an application. Empirical Economics,
the contagion effect of financial markets. Understanding the underlying 43(1), 447–456.
Hintermann, B. (2017). Market power in emission permit markets: Theory and evidence
drivers of carbon risk can provide reliable guidance on how future from the EU ETS. Environmental & Resource Economics, 66, 89–112.
carbon prices respond to economic shocks. This can assist policymakers Hoffmann, V. H., & Busch, T. (2008). Corporate carbon performance indicators. Journal of
in achieving the desired quantity of emissions through setting the target Industrial Ecology, 12(4), 505–520.
Holt, C. A., & Shobe, W. M. (2016). Reprint of: Price and quantity collars for stabilizing
quantity of emissions and designing the trading mechanisms. emission allowance prices: Laboratory experiments on the EU ETS market stability
reserve. Journal of Environmental Economics and Management, 80, 69–86.
Acknowledgments International Capital Market Association (ICMA) (2017). The green bond principles 2017.
Available at https://www.icmagroup.org/assets/documents/Regulatory/Green-
Bonds/GreenBondsBrochure-JUNE2017.pdf.
The research is financially supported by National Natural Science Jaraite-Kažukausk, J., & Kažukauskas, A. (2015). Do transaction costs influence firm
Foundation of China (No.71850007; No.71671008). trading behaviour in the European emissions trading system? Environmental and
Resource Economics, 62, 583–613.
Ji, Q., Zhang, D., & Geng, J. (2018). Information linkage, dynamic spillovers in prices and
References volatility between the carbon and energy markets. Journal of Cleaner Production, 198,
972–978.
Agliardi, E., & Agliardi, R. (2019). Financing environmentally-sustainable projects with Johnson, L. L. (1960). The theory of hedging and speculation in commodity futures.
green bonds. Environment & Development Economics, 24(6), 608–623. Review of Economic Studies, 27, 139–151.
Alberola, E., Chevallier, J., & Cheze, B. (2008). Price drivers and structural breaks in Jung, J., Herbohn, K., & Clarkson, P. (2018). Carbon risk, carbon risk awareness and the
European carbon prices 205-2007. Energy Policy, 36, 787–797. cost of debt financing. Journal of Business Ethics, 150, 1151–1171.
Aldy, J. E., & Stavins, R. N. (2012). Using the market to address climate change: Insights Junttila, J., Pesonen, J., & Raatikainen, J. (2018). Commodity market based hedging
from theory & experience. American Academy of Arts and Sciences, 141(2), 45–61. against stock market risk in times of financial crisis: The case of crude oil and gold.
Baillie, R. T., & Myers, R. J. (1991). Bivariate GARCH estimation of the optimal com- Journal of International Financial Markets, Institutions & Money, 56, 255–280.
modity futures hedge. Journal of Applied Econometrics, 6, 109–124. Kara, M., Syri, S., Lehtilä, A., Helynen, S., Kekkonen, V., Ruskaand, M., & Forsström, J.
Balcılar, M., Demirer, R., Hammoudeh, S., & Nguyen, D. K. (2016). Risk spillovers across (2008). The impacts of EU CO2 emissions trading on electricity markets and elec-
the energy and carbon markets and hedging strategies for carbon risk. Energy tricity consumers in Finland. Energy Economics, 30, 193–211.
Economics, 54, 159–172. Kroner, K. F., & Sultan, J. (1993). Time-varying distributions and dynamic hedging with
Basher, A. S., & Sadorsky, P. (2016). Hedging emerging market stock prices with oil, gold, foreign currency futures. Journal of Financial & Quantitative Analysis, 28, 535–551.
VIX, and bonds: A comparison between DCC, ADCC and GO-GARCH. Energy Ku, Y.-H. H., Chen, H.-C., & Chen, K.-h. (2007). On the application of the dynamic con-
Economics, 54, 235–247. ditional correlation model in estimating optimal time-varying hedge ratios. Applied
Batten, J. A., Ciner, C., & Lucey, B. M. (2017). The dynamic linkages between crude oil Economics Letters, 14(7), 503–509.
and natural gas markets. Energy Economics, 62, 155–170. Lee, H. T. (2010). Regime switching correlation hedging. Journal of Banking and Finance,
Benz, E., & Trück, S. (2009). Modeling the price dynamics of CO2 emission allowances. 34, 2728–2741.
Energy Economics, 1(31), 4–15. Lee, H. T., & Yorder, J. K. (2007). A bivariate markov regime switching GARCH approach
Bredin, D., Hyde, S., & Muckley, C. (2014). A microstructure analysis of the carbon fi- to estimate time varying minimum variance hedge ratios. Applied Economics, 39,
nance market. International Review of Financial Analysis, 34, 222–234. 1253–1265.
Byun, S. J., & Cho, H. (2013). Forecasting carbon futures volatility using GARCH models Lee, J., & Strazicich, M. C. (2013). Minimum LM unit root test with one structural break.
with energy volatilities. Energy Economics, 40, 207–221. Economics Bulletin, 33(4), 2483–2492.
Carmona, R., Fehr, M., Hinz, J., & Porchet, A. (2010). Market design for emission trading Luo, C., & Wu, D. (2016). Environment and economic risk: An analysis of carbon emission
schemes. SIAM Review, 52(3), 403–452. market and portfolio management. Environmental Research, 149, 297–301.
Chen, F., & Sutcliffe, C. (2012). Better cross hedges with composite hedging? Hedging Lutz, B. J., Pigorsch, U., & Rotfub, W. (2013). Nonlinearity in cap-and-trade systems: The
equity portfolios using financial and commodity futures. European Journal of Finance, EUA price and its fundamentals. Energy Economics, 40, 222–232.
18(6), 575–595. Makridou, G., Doumpos, M., & Galariotis, E. (2019). The financial performance of firms
Chevallier, J. (2009). Carbon futures and macroeconomic risk factors: A view from the EU participating in the EU emissions trading scheme. Energy Policy, 129, 250–259.
ETS. Energy Economics, 31, 614–625. McKenzie, M., Mitchell, H., Brooks, R., & Faff, R. (2001). Power arch modelling of
Chevallier, J. (2011). Detecting instability in the volatility of carbon prices. Energy commodity futures data on the London metal exchange. European Journal of Finance,
Economics, 33, 99–110. 7, 22–38.
Chung, S. L., Tsai, W. C., Wang, Y. H., & Weng, P. S. (2011). The information content of Monasterolo, I., & Raberto, M. (2018). The EIRIN flow-of-funds behavioural model of
the s&p500 index and vix options on the dynamics of the s&p 500 index. Journal of green fiscal policies and green sovereign bonds. Ecological Economics, 144, 228–243.
Futures Markets, 31(12), 1170–1201. Morgan Stanley (2017). Behind the green bond boom. Retrieved from https://www.
Conrad, C., Rittler, D., & Rotfuß, W. (2012). Modeling and explaining the dynamics of morganstanley.com/ideas/green-bond-boom.
European Union allowance prices at high-frequency. Energy Economics, 34(1), Nguyen, D. K., Sousa, R. M., & Uddin, G. S. (2015). Testing for asymmetric causality
316–326. between U.S. equity returns and commodity futures returns. Finance Research Letters,
Diebold, F. X., & Yilmaz, K. (2014). On the network topology of variance decompositions: 12, 38–47.
Measuring the connectedness of financial firms. Journal of Econometrics, 182(1), Philip, D., & Shi, Y. (2016). Optimal hedging in carbon emission markets using Markov
119–134. regime switching models. Journal of International Financial Markets, Institutions &
Ding, Z., Granger, C. W. J., & Engle, R. F. (1993). Long memory property of stock market Money, 43, 1–15.
returns and a new model. Journal of Empirical Finance, 1, 83–106. Reboredo, J. C. (2018). Green bond and financial markets: Co-movement, diversification
Engle, R. F. (2002). Dynamic conditional correlation: A simple class of multivariate and price spillover effects. Energy Economics, 74, 38–50.
GARCH models. Journal of Business & Economic Statistics, 20, 339–350. Sanin, M. E., Violante, F., & Bataller, M. M. (2015). Understanding volatility dynamics in
Fan, J. H., Rocan, E., & Akimov, A. (2014). Estimation and performance evaluation of the EU-ETS market. Energy Policy, 82, 321–331.
optimal hedge ratios in the carbon market of the European Union Emissions Trading Stern, N. (2016). The criticality of the next 10 years: Delivering the global agenda and
Scheme. Australian Journal of Management, 39, 73–91. building infrastructure for the 21st century. Speech at the London School of Economics
Gianfrate, G., & Peri, M. (2019). The green advantage: Exploring the convenience of is- in occasion of the 10 years of the Stern review, 28 October, 2016, London (UK).
suing green bonds. Journal of Cleaner Production, 219, 127–135. Tian, Y., Akimov, A., Roca, E., & Wong, V. (2016). Does the carbon market help or hurt
Glosten, L. R., Jagannathan, R., & Runkle, D. E. (1993). On the relation between the the stock price of electricity companies? Further evidence from the European context.
expected value and the volatility of the nominal excess return on stocks. Journal of Journal of Cleaner Production, 112(2), 1619–1626.
Finance, 48(5), 1779–1801. Zakoian, J. M. (1994). Threshold heteroskedastic models. Journal of Economic Dynamics
Hamilton, J. D. (2003). What is an oil shock? Journal of Econometrics, 113(2), 363–398. and Control, 18(5), 931–955.
Hamilton, J. D. (2011). Nonlinearities and the macroeconomic effects of oil prices. Zhang, Y., Liu, Z., & Yu, X. (2017). The diversification benefits of including carbon assets
Macroeconomic Dynamics, 15(S3), 364–378. in financial portfolios. Sustainability, 9, 437.

17