SSRN Id4613695

Paper prepared for International Review of Financial Analysis
ed
Title: Co-jump dynamicity in the commodity futures markets
iew
Authors: Lei Zhang1, Elie Bouri2 and Yan Chen1, *
1
Business School, Hunan University, Changsha 410082, China
v
2
School of Business, Lebanese American University, Lebanon
re
[email protected] (Lei Zhang)
[email protected] (Elie Bouri)
[email protected] (Yan Chen)
*Corresponding author: Yan Chen

er
pe
Abstract
This paper examines the co-jump transmission in 36 commodity futures returns using co-jump
network models. Specifically, it analyses co-jumping behavior in both static and time-varying
settings, considering the overall commodity markets and various commodity groups separately,
ot
which helps us understand the dynamic changes in co-jump dependencies at the overall and
sector levels across time. The main results reveal that co-jump heterogeneity exists among
commodities but is generally more apparent within each commodity group, and co-jumps vary
tn
over time. Gold exerts the strongest influence, with many commodity futures being influenced
by the jumps behavior in gold returns. During the COVID-19 outbreak and the Russia-Ukraine
war, the Energy group ranks highest in terms of co-jump network centrality. Some Agricultural
and Metals commodity futures such as Lumber, Nickel and Palladium show a higher centrality
rin
ranking after the Russia-Ukraine conflict. Our empirical analysis highlights the portfolio
implications and an additional examination shows that centrality information from the co-jump
network contains a highly and statistically forecasting power for US stock market volatility.
Keywords: Co-jump network; commodity futures; co-jump transmission; COVID-19 and
ep
Russia-Ukraine war; volatility forecasting implications.

Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4613695
Co-jump dynamicity in the commodity futures markets
ed
iew
Abstract
This paper examines the co-jump transmission in 36 commodity futures returns using
v
co-jump network models. Specifically, it analyses co-jumping behavior in both static
and time-varying settings, considering the overall commodity markets and various
re
commodity groups separately, which helps us understand the dynamic changes in co-
jump dependencies at the overall and sector levels across time. The main results reveal
that co-jump heterogeneity exists among commodities but is generally more apparent
within each commodity group, and co-jumps vary over time. Gold exerts the strongest
er
influence, with many commodity futures being influenced by the jumps behavior in
gold returns. During the COVID-19 outbreak and the Russia-Ukraine war, the Energy
group ranks highest in terms of co-jump network centrality. Some Agricultural and
pe
Metals commodity futures such as Lumber, Nickel and Palladium show a higher
centrality ranking after the Russia-Ukraine conflict. Our empirical analysis highlights
the portfolio implications and an additional examination shows that centrality
information from the co-jump network contains a highly and statistically forecasting
power for US stock market volatility.
ot
Keywords: Co-jump network; commodity futures; co-jump transmission; COVID-19

and Russia-Ukraine war; volatility forecasting implications.
tn
rin
ep
Pr
1. Introduction
ed
Discontinuous variations of the asset price process, called jumps, are rare events in
financial markets and encompass a multitude of crucial insights for asset pricing and
risk management. Bates (1991) argues that jumps incorporate the risk of financial
market crashes, and later researches emphasize the significance of jumps in various
domains, such as volatility forecasting (Andersen et al., 2007; Clements & Liao, 2017),
iew
risk premium explanation (Bollerslev and Todorov, 2011), return modelling (Cremers
et al., 2015; Bollerslev et al., 2015), diversification (Bollerslev et al., 2008), and options
pricing (Bates, 2019). Notably, the simultaneous occurrence of jumps in two or more
stocks or assets, called co-jumps, is broadly accepted. Previous studies (see, Gilder et
al., 2014; Caporin et al., 2017) examine co-jumps in the stock market, showing that the
v
simultaneous occurrence of large jumps in individual stocks is associated with market-
level news, including Fed Funds target rate announcements1, and suggesting that co-
re
jumps imply short-term predictability of stock returns and have an influence on the risk
premia (Caporin et al., 2017).
Much less attention has been given to the presence of co-jumps in commodities,
despite the emergence of commodities as an asset class highly appreciated by market
er
participants. The importance of commodity markets has been widely recognized over
the past two decades as an instrument useful for hedging and the portfolio choice, and
the dynamic behaviour of commodities has been prominently examined (Christoffersen
pe
et al., 2019)), especially during stress periods such as the 2008 global financial crisis,
the COVID-19 outbreak, and the Russian-Ukrainian war (Wang et al., 2022). Various
groups of commodities exist, such as Energy, Metals, Agricultural, and Grains, offering
diversification and hedging possibilities (Rehman et al., 2019; Abid et al., 2020). The
relationship across various commodities is shown to time-varying and tends to increase
ot
under crisis periods (Zhang and Broadstock, 2020; Bouri et al., 2021), reducing and
hedging diversification possibilities.
The prices of commodity futures tend to experience violent fluctuations, especially
tn
during crisis periods, which lead to enormous volatility. Jumps in returns differ from
jumps in volatility, although they can serve complementary purposes (Bandi et al.,
2016), and existing studies typically focus on jumps in returns. Several studies indicate
the presence of jumps in the commodity futures returns (Chatrath and Song, 1999;
rin
Diewald et al., 2015; Sévi, 2015; Chevallier and Ielpo, 2014; Prokopczuk et al., 2016;
Nguyen and Prokopczuk, 2019; Bouri and Gupta, 2020). However, the literature on co-
jumps in the commodity markets is limited. Laurini et al. (2020) examine jump and co-
jumps in oil prices and stock prices of integrated oil companies using a new multivariate
ep
stochastic volatility model with jumps. Jawadi et al. (2019) study the role of co-jump
to forecast the volatility in commodity futures markets. Other studies examine the co-
jump behaviour of commodity futures markets with different markets such as energy
spot and futures markets (Maslyuk et al., 2017), the U.S. interest rates and precious
Pr
1
Aït-Sahalia and Xiu (2016) use high-frequency data and argue that co-jumps arises from surprising
news announcements that occur mainly in the pre-market.
metals markets (Semeyutin et al., 2022). Bouri and Gupta (2020) study the presence of
jumps in the volatility of energy and non-energy commodity markets, showing that
ed
jumps tend to occur more with the Energy group than across the commodity groups,
and that jumps are associated with macroeconomic news surprises, especially in crude
oil. Semeyutin et al. (2021) find that three commodity futures (crude oil, gold, and
copper) exhibit co-jumping behaviour, which can trigger jumps in the next periods.
iew
They provide evidences that, despite the significant influence of jump risks on
commodities correlations, the potential for diversification benefits remains intact,
indicating that investors can exploit market-specific price jump in the copper and oil
markets to enhance diversification strategies.
The related literature on jumps and co-jumps lacks a comprehensive analysis of the
v
co-jump transmission in commodity futures returns within a co-jump network model in
both static and time-varying settings, and their feasible implications. Specifically, there
re
is no empirical evidence and model on how commodity futures returns co-jump and
how their co-jump behaviours evolve over time and during crisis periods, which if
addressed could help to understand the mechanisms of jump transmission in the
commodity futures markets and the portfolio implications, both of which should
er
provide useful information to market participants. In fact, the transmission of co-jump
in commodity futures returns have received very limited attention at both the aggregate
and industry-group levels, and the following questions remain unanswered: Which
pe
commodity futures act as bridges for cross-industry group jumps? Which commodity
futures hold the highest ranking in the co-jump network during the full sample period
and crisis periods? How does the dynamics of interrelated jumps in commodity futures
returns change during crisis events? What are the feasible implications of significant of
co-jumps in commodity markets?
ot
Our study provides answers to these important research questions in commodity

markets. Our study addresses the above-mentioned research gap by explicitly
identifying the co-jump transmission in 36 commodity futures returns using co-jump
tn
network models. We examine the co-jump dependency in commodity futures returns

within the overall commodity market and across different industry groups, from the
perspective of network analysis. We are particularly interested in understanding the
dynamic changes in these co-jump dependencies over time. The co-jump network
rin
model can reflect the dependent feature of edges in the real-world network, and
discover pairwise jump dependency among a large number of commodity futures
returns. This approach allows us to examine not only the occurrence of co-jumps in
commodity future markets but also the intensity of co-jumps between pairs of
ep
commodity futures returns from a network visualization perspective. Traditional co-

jump methods often focus on the co-jumps between two commodity futures returns,
whereas the co-jump network model can simultaneously consider the jump dependency
among multiple pair of commodity futures returns. This comprehensive consideration
Pr
helps to reveal the co-jump relationships between different commodity futures returns
and provides a more holistic understanding of commodity market dynamics.
Furthermore, the co-jump network model captures more complex commodity futures
market jump relationships by constructing a network of co-jump dependency among
commodity futures returns, which can reveal the co-jump transmission paths across and
ed
within commodity groups. This should provide more refined understandings of jump
risk and practical insights into the feasible implications of co-jumps in commodities.
The main contributions of this paper are summarized as follows: Firstly, following
the approach proposed by Ding et al. (2023) and adopted by Zhang et al. (2023) in the
iew
cryptocurrency markets, we apply the co-jump network framework to study co-jump
dependency among 36 commodity futures returns, offering a comprehensive
understanding of how co-jumps in commodity futures returns are transmitted across
time. The co-jump network model can identify important nodes in the commodity future
markets, which help reveal the commodity futures that have the largest impact on
v
market jumps. Specifically, unlike the work of Laurini et al. (2020) and Semeyutin et
al.(2021), we assess the intensity of co-jumps between commodity futures returns based
re
on the co-jump network, which helps us understand which commodity futures have
stronger impacts on others through jumps, and which have a weaker influence from
other commodity futures returns’ jumps. Secondly, we compare and analyse the
similarities and differences between whole commodity future markets’ co-jump
er
network and different industry groups’ co-jump network to explore time-varying nature
of co-jumps. Thirdly, we compare the structure of co-jump networks during the
COVID-19 crisis and Russia-Ukraine conflict to assess the diverse co-jump impact of
pe
these crises on commodity futures markets. Finally, we indicate the role of centrality
information from the co-jump network in forecasting US stock market volatility, which
extends the study of Semeyutin et al. (2021) and Zhang et al. (2023).
The empirical analysis shows that co-jumps exist in commodity futures returns and
are heterogeneous among commodities but are generally more apparent within each
ot
commodity group. The impact of co-jump centrality is time-varying, reflecting its

sensitivity to crisis periods. Notably, the dynamics ranking of co-jump influence are
revealed during some crises such as COVID-19 and Russia-Ukraine conflict, which
tn
show that co-jump dependencies have diverse change. During the COVID-19 outbreak
and the Russia-Ukraine war, the Energy group is ranked highest in terms of co-jump
network centrality. Further analysis shows that the mean-variance portfolio strategies
constructed based on the dynamics ranking of co-jump influence can achieve a higher
rin
Sharpe ratio compared with the baseline portfolio strategy. The overall findings suggest
that, when modelling the dynamic volatility of commodity futures, it is relevant to
consider not only the presence of co-jumps but also how commodity futures return co-
jump across time and which commodity futures have a strong co-jump impact on others.
ep
This also should apply to the analysis co-jumps in each commodity group.
2. Related studies
Besides the supply and demand forces, commodity prices are affected by various
Pr
potential factors such as hedging pressure, inflation uncertainty, and macroeconomic

conditions (Nguyen & Walther, 2020)). In the context of the overall commodity
markets, there exists a deeper market complexity, wherein vents such as extreme
weather conditions, natural disasters, heightened geopolitical risk, and political
interventions can disrupt the equilibrium of supply and demand relationships, leading
ed
to large price fluctuations and heightened volatility. For example, during the peak of
the COVID-19 outbreak, crude oil prices, notably WTI futures, collapsed and displayed
negative prices. Under the Russia-Ukraine war, the prices of crude oil, natural gas, and
wheat spiked and experienced large price fluctuations (Wang et al., 2022). Sharp market
iew
volatility can trigger a contagion effect among different commodity futures, further
exacerbating the risk interdependence within the commodity markets. This is not only
relevant within the same commodity group such as Energy, Agricultural, or Grains, but
across various commodity groups. In this regard, the biofuel expansion has triggered
strong linkages between Energy and Grains commodity groups (see, among others,
Bahel et al., 2013; Ji et al., 2018; Su et al., 2019). Furthermore, the financialization
v
phenomenon of commodities (Tang and Xiong, 2012) has eased the transmission of
re
return and volatility information among commodities. Commodities are diverse and
comprise various sectors, covering Energy, Metals, Agricultural, and Grains. In the
Energy sector, crude oil is a strategic commodity, crucial for the economic cycle. In the
Metals, gold is often used as an investment acting as a safe-haven during periods of
market stress. However, Iron, Aluminium, and Copper are essential industrial
er
commodities, with Copper in particular being regarded as a predictor of economic
trends, given its versatile usage in various industries covering electrical wiring,
plumbing, construction, and transportation. Wheat is a food staple in many countries,
pe
often used in the production of bread, pasta, and pastries.
The existing mainstream economic and financial literature focuses on commodities
in the several aspects, including price characteristics (Pindyck, 2001), information
interdependence (Ji et al., 2019), volatility spillovers (Zhang et al., 2023),
diversification and hedging (Rehman et al., 2019; Abid et al., 2020) 2 , and co-
ot
movements across timescales (Nekhili et al., 2021; Bouri et al., 2023). Evidence on the
long-term cyclical behaviour of commodity prices is well documented (Erten and
Ocampo, 2013; Rossen et al., 2015), and the recent evidence from Bouri et al. (2023)
tn
show evidence supporting the strong positive correlation at the long-term across various
commodities. However, they indicate that integration in the commodity markets
evolves over time and is influenced by crisis periods.
rin
Regarding jumps, several studies examine the presence of jumps in the commodity
futures. Chatrath and Song (1999) observe a negative correlation between the
probability of agricultural commodity price increases and the presence of speculators.
Diewald et al. (2015) study price jumps in energy futures and identified seasonal
ep
patterns in tail events. Sévi (2015) examine the existence of jumps in the crude oil
market. Chevallier and Ielpo (2014) investigate the impact of jumps in 20 major
commodities, revealing that the occurrence of jumps in commodity markets is more
frequent compared to other financial assets, and that there are significant variations in
Pr
2
These two studies point to the diversification and hedging possibilities across various groups of
commodities such as Energy, Metals, and Agricultural.
terms of quantity, scale, and direction of jumps across different commodity markets.
Prokopczuk et al. (2016) discover significant jump behavior in four energy markets,
ed
however, the jump models did not demonstrate significant predictive power for
volatility. Nguyen and Prokopczuk (2019) study the price jump behaviour in the
commodity market and find that jump correlations across commodities can be
substantial, varying depending on the different commodity sectors. For example,
iew
Energy, Metals and, Grains exhibit high jump correlations while Meats and Softs
commodities exhibit very weak jump correlations. They further show that commodities
weakly hedge jumps U.S. Dollar jumps, whereas they can hedge both the returns and
jumps of U.S. Treasury notes. Dutta et al. (2021) model the return volatility of crude
oil prices and show that accounting for the jump behaviour of crude oil implied
volatility is useful to improve the forecasting of the variance of crude oil returns.
v
The studies that are most closely related with our analysis on co-jumps in commodity
re
markets are two. The first is conducted by Bouri and Gupta (2020) who examine the
jumps in the volatility of various commodities using the GARCH-based approach of
Laurent et al. (2016) and show that the volatility of commodities jump together. They
further indicate that co-jumps tend to occur more within the commodity groups, and
er
that macroeconomic news surprises tend to drive jumps, especially in the crude market.
However, Bouri and Gupta (2020) do not consider the co-jump transmission in
commodity futures returns within a co-jump network model in both static and time-
pe
varying settings and the portfolio implications of co-jumps. The second study is
accomplished by Semeyutin et al. (2021) who use high frequency data and wavelet
decomposition approach to examine the influence of market-specific jumps and
systematic risks on the interdependence structure of commodities and the portfolio
inferences. They show that crude oil, gold, and copper futures exhibit co-jumping
behaviour, which can trigger jumps in the next periods. Semeyutin et al. (2021) focus
ot
on three commodities only, crude oil, gold, and copper, and overlook the comprehensive
analysis of co-jump network modelling and its portfolio implications for the entire
universe of commodity futures. Furthermore, they ignore the time-varying co-jumps in
tn
commodity futures and the central role of each individual commodity or commodity
groups across crisis periods, such as the COVID-19 and Russia-Ukraine war periods in
the co-jump network. This is where we seek to contribute to the research on co-jumps
in the commodity futures markets.
rin
For the rest of the paper, Section 3 offers the dataset, Section 4 describes the jump
and co-jump tests and the co-jump network model, Section 5 presents and discusses the
empirical results, Section 6 offers a portfolio allocation and volatility forecasting
ep
analysis, and Section 7 concludes.
3. Data
We use daily and 5-minute frequencies prices against the USD for 36 most common
Pr
commodity futures from Wind over the period from July 5, 2016 - July 1, 2023. Then,
we compute daily log-returns 𝑟𝑡 as log(𝑝𝑡 ) − log(𝑝𝑡−1 ) , where 𝑝𝑡 represents the
daily closing price of a commodity future at time t. In subsection 4.1, 5-minute prices
are employed to compute realized volatility (RV), realized bipower variation (BV), and
the daily threshold level. These metrics are subsequently juxtaposed against the daily
ed
returns to ascertain the presence of a jump or a co-jump on the given day. The
commodities belong to four categories, namely Energy (7 commodities), Metals (10
commodities), Agricultural (11 commodities), and Grain and Cereal (8 commodities).
Interestingly, we consider Carbon emission allowance because it represents an
iew
important instrument under the decarbonisation phenomenon. Following Bouri et al.
(2023), we use futures to the detriments of spot because they play an important and
leading role in the price formation of commodities (for energy, see Shrestha, 2014; for
metals, see Figuerola-Ferretti and Gonzalo (2010). Regarding the gasoline commodity,
we consider two prices: RBOB and London. This is because the RBOB gasoline futures
market is primarily influenced by factors specific to the United States, whereas London
v
gasoline futures, are influenced by factors specific to the European market (Mirantes et
re
al. 2012). Furthermore, commodity futures of the same variety do exist, and this is
reflected in the difference of the reported results on jumps in gasoline between RBOB
and London. Regarding Wheat, we consider two major trading hubs for wheat and
provide valuable information about global wheat market trends and price discovery.
Interestingly, we get different jumps behavior from the two types of Wheat futures.
er
Regarding Coffee, we use both London and US coffee commodities, because US Coffee
Futures are based on Arabica coffee, while London Coffee Futures are based on Robusta
coffee (Thompson, 1986).
pe
Appendix Figure A1 plots the prices of the 36 commodity futures under study, while
Appendix Figure A2 plots their logarithmic returns (multiplied by 100). We notice that
these assets exhibit significant variations in price and return dynamics, particularly
among commodity futures across different groups. During the peak of the COVID-19
outbreak around March 2020, many energy commodities display large price changes,
ot
whereas during the Russia-Ukraine war we notice pronounced changes in the prices of
wheat, oat, and Nickel. Table A1 presents the summary statistics of commodity futures
returns and the results of the stationarity test. RBOB Gasoline and Carbon Emission
tn
allowances3 have the highest mean return, whereas WTI Crude Oil has the lowest mean
return. Gold has the lowest standard deviation, which is not surprising given its safe
haven property. Excess kurtosis is dominant, especially in WTI and Nickel, whereas
skewness is negative in 22 commodity futures and positive in 14 commodity futures.
rin
All commodity futures return series are stationary, as indicated by the results of the
Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) tests.
4. Methods
ep
4.1. Detecting jump and pairwise co-jump scores

In this section, we present the methods used to identify jumps and co-jumps in
commodity futures.
Pr
3
Following Chevallier (2009), we consider carbon emission allowances as part of the commodity
markets.
Let us consider N logarithmic commodity future price processes {𝑌𝑖 (𝑡)}𝑁
𝑖=1 , with
commodity i obeys the following process:
ed
𝒅𝒀𝒊 (𝒕) = 𝝁𝒊 (𝒕)𝒅𝒕 + 𝝈𝒊 (𝒕)𝒅𝑾𝒊 (𝒕) + 𝜿𝒊 (𝒕)𝒅𝑱𝒊 (𝒕) (1)
where 𝜇𝑖 (𝑡) is the drift term of commodity future i, 𝜎𝑖 (𝑡) is the volatility process of
commodity i, 𝐽𝑖 (𝑡) and 𝜅𝑖 (𝑡) are the Poisson count process and the jump size process,
iew
respectively.
Suppose we collect commodity future prices at a frequency M, the s-th intraday
logarithmic return for commodity futures i on day t as is given by:
𝒔 𝒔−𝟏
𝒓𝒊,𝒕,𝒔 = 𝒀𝒊 (𝒕 − 𝟏 + 𝑴) − 𝒀𝒊 (𝒕 − 𝟏 + ) , 𝒔 = 𝟏, 𝟐, … , 𝑴. (2)
𝑴
v
Considering the physical measure, as M approaches infinity, the realized volatility
re
(RV) for commodity future i on day t has the following approximate measure:
𝒕
𝐥𝐢𝐦 𝑹𝑽𝒊,𝒕 → ∫𝒕−𝟏 𝝈𝟐𝒊 (𝒖)𝒅𝒖 + ∑𝒕−𝟏<𝒔≤𝒕 𝜿𝟐𝒊 (𝒔), (3)
𝑴→∞
𝑡
er
where ∫𝑡−1 𝜎𝑖2 (𝑢)𝑑𝑢 and ∑𝑡−1<𝑠≤𝑡 𝜅𝑖2 (𝑠) are the integrated volatility and the
quadratic variation due to the jump component, respectively. The integrated volatility
can be estimated by the realized bipower variation (BV) (Barndorff-Nielsen and
pe
Shephard (2004)):
𝒕 𝝅 𝑴
𝐥𝐢𝐦 𝑩𝑽𝒊,𝒕 → ∫𝒕−𝟏 𝝈𝟐𝒊 (𝒖)𝒅𝒖 , 𝑩𝑽𝒊,𝒕 = ∑𝑴−𝟏
𝒔=𝟏 |𝒓𝒊,𝒕,𝒔+𝟏 | |𝒓𝒊,𝒕,𝒔 | (4)
𝑴→∞ 𝟐 𝑴−𝟏
To identify commodity futures price jumps, the non-parametric threshold method

ot
introduced by Mancini (2009) is used. Compared with other methods, such as the one
adopted by Bouri and Gupta (2020) 4 , the non-parametric threshold estimation can
reduce the sensitivity to model misspecification and outliers, which make it more robust
tn
in facing the noise and extreme observations, and also allowing for the adaptive
selection of threshold levels, which can vary over time and capture changes in the
intensity and occurrence of jumps. More specifically, for each commodity future i, the
daily threshold level is given as:
rin
𝟏 𝟎.𝟒𝟗
𝝂𝒊,𝒕 = 𝝉 × (𝑴) × √𝐦𝐢𝐧 (𝑹𝑽𝒊,𝒕 , 𝑩𝑽𝒊,𝒕 ), (5)
where we set 𝜏= 4 following Ding et al. (2023) and establish a daily threshold level.
Assuming certain regularity conditions on 𝜇𝑖 (t) and 𝜎𝑖 (𝑡), and as the frequency M
ep
approaches infinity, the consistent detection of jumps is possible (see Mancini, 2009).
To achieve a trade-off between mitigating the influence of microstructure noise and
adhering to the methodology proposed by Ding et al. (2023), we employ 5-minute price
data to compute realized volatility (RV), realized bipower variation (BV), and the daily
Pr
4
The authors detect jumps in the volatility process, using the GARCH-jump model of Laurent et al.
(2016).
threshold level. These metrics are then utilized to determine the presence of a jump or
co-jump on the specific day We denote the cumulative number of jumps detected for
ed
commodity future i from day 1 to day t as:
𝑱𝒊,𝒕 = ∑𝟏≤𝒅≤𝒕,𝟏≤𝒔≤𝑴 𝟏{|𝒓𝒊,𝒅,𝒔|>𝝂𝒊,𝒅} (6)
where {1} is the indicator function.
iew
Following Gilder et al. (2014), we define co-jumps between any pair of commodities
returns as the simultaneous occurrence of jumps in the pair of commodities returns, i.e.
the intersection of their individual jumps. We denote the cumulative number of co-
jumps between commodity future i and commodity future j from day 1 to day t as:
v
𝑪𝑱𝒊𝒋,𝒕 = ∑𝟏≤𝒅≤𝒕,𝟏≤𝒔≤𝑴 𝟏{|𝒓𝒊,𝒅,𝒔 |>𝝂𝒊,𝒅,|𝒓𝒋,𝒅,𝒔|>𝝂𝒋,𝒅} (7)
re
The co-jump indices between pairs of commodity futures returns have three useful
characteristics. Firstly, they are non-negative, where higher values indicate a stronger
co-jump relationship. Secondly, the co-jumps indices are scaled, allowing for
comparisons between relationships among pairs of commodity futures. Thirdly, the co-
jumps indices are symmetric, reflecting the symmetry of the co-jump relationship
between two commodity futures.
er
4.2. Co-jump matrices and networks
pe
Traditional co-jump methods only consider the co-jumps between two financial
assets, but cannot simultaneously consider the jump dependencies among multiple
commodity futures. The co-jump network model can address this problem, which helps
to reveal the co-jump dependency between commodity futures and provide more
holistic understanding for co-jump dynamicity of commodity futures.
ot
There are several aspects of the co-jump network, which can extend our
understanding of co-jumps in commodity future markets. Firstly, the co-jump network
tn
helps us to identify which commodity futures tend more to experience a co-jumping

behaviour. This allows us to understand the jump relationships and interdependencies
across commodity futures. Secondly, the co-jump network analysis can shed light on
the efficiency of the commodity future markets, given that the frequent and persistent
rin
occurrence of co-jumps may be exploited by market participants to generate abnormal

returns 5 . Finally, the co-jump network can be used to optimize commodity future
portfolios. By considering the centrality ranking of co-jump influence between
commodity futures, investors and portfolio managers can construct portfolios that are
ep
better diversified and resilient to co-jump events.

Using the pairwise co-jump measures, we construct the corresponding daily 𝑁 × 𝑁
Pr
5
For example, when co-jumps occur due to news or similar events, it suggests a strong relationship or
co-movement between two markets. If this relationship persists and happens frequently, it may indicate
that the market is not fully efficient in incorporating the information from these news events into the
prices of the affected commodities.
co-jump matrix A(T) for commodity futures based on the approach of Ding et al. (2023),
as follows:
ed
𝑱𝒊 (𝑻) 𝑪𝑱𝒊𝒋 (𝑻)
(𝑨(𝑻))𝒊𝒊 = 𝒇𝒐𝒓 𝟏 ≤ 𝒊 ≤ 𝑵, 𝒂𝒏𝒅 (𝑨(𝑻))𝒊𝒋 = 𝒇𝒐𝒓 𝟏 ≤ 𝒊 ≠ 𝒋 ≤ 𝑵 (8)
𝑻 𝑻
The co-jump network can represent complex structures in the commodity futures,
iew
and has a symmetrical matric. We consider dynamic co-jump networks of commodity
futures, 𝐺𝑡 = (𝑉𝑡 , 𝐸𝑡 ), 𝑡 ∈ 𝑇 , where each vertex 𝑁𝑖,𝑡 ∈ 𝑉 represents a certain
commodity futures at time t, and a weighted edge 𝑒𝑖,𝑗 (𝑡) ∈ 𝐸 denotes a type of co-
jump dependency between two commodity futures at time t; the weight being the
corresponding co-jump intensity.
v
5. Empirical results
re
5.1. Results of commodity future jumps
Table 2 presents the statistics for the jumps in commodity futures over the period
July 5, 2016 - July 1, 2023. We notice that Gold, Wheat, Live Cattle, Soybeans, and
Aluminum exhibit the lowest jump activity, with 70, 137, 154, 173 and 179 jumps
er
recorded, respectively, representing 4.1%, 8.03%, 9.02%, 10.13% and 10.49% of
observed days. Conversely, Natural Gas, Carbon Emissions, Lumber, WTI Crude Oil,
and RBOB Gasoline account for the highest number of jumps, with 788, 718, 673, 544
pe
and 536 jumps, respectively, representing 46.16%, 42.06%, 39.43%, 31.87% and 31.4%
of observed days. These findings generally demonstrate the recurring nature of jumps
in Energy commodities in particular. Furthermore, our results indicate that strategic
Energy commodity futures such as crude oil tend to exhibit comparatively more jumps,
which bears significance for the modelling of volatility (Chen et al., 2019; Dutta et al.,
ot
2021).
Table 2. Statistics of jumps detected

tn
Industry group Commodity Number % of days Industry group Commodity Number % of days
future of jumps with jumps future of jumps with jumps
Metals Silver 298 17.46% Agricultural US Cocoa 360 21.09%
rin
Metals Palladium 424 24.84% Agricultural Live Cattle 154 9.02%

Metals Platinum 256 15% Agricultural Lean Hogs 447 26.19%
Metals Lead 235 13.77% Energy RBOB Gasoline 536 31.4%
Metals Copper 226 13.24% Energy WTI Crude Oil 544 31.87%
ep
Metals Aluminum 179 10.49% Energy Brent Crude Oil 488 28.59%
Metals Zinc 309 18.1% Energy London Gasoline 477 27.94%
Metals Tin 208 12.19% Energy Natural Gas 788 46.16%
Metals Nickel 434 25.42% Energy Heating Oil 466 27.3%
Pr
Metals Gold 70 4.1% Energy Carbon Emission 718 42.06%

Agricultural London Cocoa 283 16.58% Grain and Cereal London Wheat 137 8.03%
Agricultural London Coffee 227 13.3% Grain and Cereal Oats 482 28.24%
Agricultural London Sugar 227 13.3% Grain and Cereal Corn 253 14.82%
Agricultural Lumber 673 39.43% Grain and Cereal Rice 208 12.19%
ed
Agricultural Orange Juice 510 29.88% Grain and Cereal Soybeans 173 10.13%
Agricultural US Sugar 368 21.56% Grain and Cereal Soybean Oil 269 15.76%
Agricultural US Coffee 471 27.59% Grain and Cereal Soybean Meal 239 14%
Agricultural US Cotton 355 20.8% Grain and Cereal US Wheat 423 24.78%
iew
Note: This table presents the number of jumps detected in the return series of the 36 commodity futures under study,
and the percentage (%) of days with jumps.
5.2. Co-jump network analysis I: total 36 commodity futures

We provide a comprehensive visualization of commodity futures co-jump
v
dependency by constructing a co-jump network-based matrix, denoted by 𝐴2016−2023 ,
which cover all 36 commodity futures from July 15, 2016 to July 1, 2023. Figure 1
re
shows this co-jump network based on the total 36 commodity futures returns. Each
vertex in the graph represents an individual commodity future. The colour distribution,
ranging from deep blue to deep red, reflects the intensity of co-jump dependencies
between pairs of commodity futures returns. er
From Figure 1, we can find some meaningful patterns in the cross-sectional structure
of commodity futures returns. Firstly, Gold has the deepest red distribution, indicating
its lowest sensitivity to jumps from the other 35 commodity futures returns. This result
pe
is not surprising given evidence from the academic literature on the hedging and safe-
haven properties of Gold during crisis periods (Baur and Lucey, 2010). While analysing
jumps and co-jumps in three commodities, crude oil, gold, and copper, Semeyutin et al.
(2021) point to the safe-haven property of gold. Some under-the-radar commodity
futures, e.g. Live Cattle, London Sugar and London Coffee, also exhibit lower jump
ot
sensitivity when other commodity futures returns occur jumps. This can be explained
by the low level of market integration of these various commodities, which do not
necessarily share many common factors affecting their demand/supply and thus their
tn
price formation. Secondly, commodity futures belonging to the Energy group have a
higher likelihood of experiencing co-jumps, which is in line with Bouri and Gupta
(2020). The major colours of the commodity futures within the Energy Industry group
are blue, which reflect the higher sensitivity to jumps. This result is not surprising given
rin
the huge price fluctuations experienced by crude oil and natural gas during crisis
periods, especially the COVID-19 and the Russia-Ukraine war (Wang et al., 2022;
Bouri et al., 2023). Thirdly, commodity futures belonging to the Energy and
Agricultural groups have higher co-jump dependency (the major colours of the
ep
commodity futures connecting the Energy and Agricultural industry group is blue,
compared to other industry groups). This result can be explained by the biofuel
expansion that has intensified the linkages between Energy and Agricultural markets
(Bahel et al., 2013; Ji et al., 2018; Su et al., 2019). This result is generally in line with
Pr
Nguyen and Prokopczuk (2019) who show that Energy and Grains experience high
jump correlations.
We have the following potential explanations for the above results. The jump
sensitivity of commodity futures returns may be related to their specific characteristics.
ed
Firstly, Gold is typically considered a safe-haven asset in commodity futures, so its
price fluctuations are typically influenced by financial markets and the global economy,
and are less susceptible to other commodity futures. In contrast, price fluctuations in
other commodities (e.g., Live Cattle, London Sugar and London Coffee) are likely to
iew
be influenced primarily by supply and demand factors, making them have lower
sensitive to jumps when other commodity futures return occur jumps. Second, energy
markets are usually influenced by geopolitical risks, supply disruptions, and global
economic conditions, which can lead to common price fluctuations in commodity
futures within the Energy industry group, making their co-jumps more likely. Finally,
Energy and Agricultural are fundamental to the functioning of the economy and they
v
are inextricably linked. For example, higher energy prices may lead to higher
re
agricultural production and transportation costs, which in turn affect the prices of
agricultural commodities. Furthermore, higher energy prices trigger more demand for
biofuel and thus for some grains such as Corn, Sugar, Wheat, and Soybeans (Bahel et
al., 2013; Ji et al., 2018; Su et al., 2019). As a result, Energy and Agricultural industry
groups tend to experience have a higher co-jump dependence.
er
pe
ot
tn
rin
Figure 1. Co-jump network based total 36 commodity futures

ep
Notes: The connection between pairs of commodity futures indicates the strength of co-jumps in different groups.
The colour gradient in the legend at the bottom right represents the strength of co-jump relationships, ranging from
deep blue to deep red. The colour gradient in the legend at the top right represents the width of the connection
between pairs of commodity futures, ranging from light grey to black.
Pr
In order to analyse the more jump influence nodes based total 36 commodity futures,
we consider the approach of Mantegna (1999) to filter some unimportant edges using a
minimum spanning tree (MST) that maximizes the sum of the co-jump scores.
Additionally, we use eigenvector centrality, based on the work of Bonacich (1987), to
measure the influence of each commodity future on the co-jump network. Figure 2
ed
shows the co-jump network based on the total 36 commodity futures after filtering the
edges. We notice that Gold has the strongest jump influence, indicating that the jumps
in most other commodity futures returns are influenced by the jumps in Gold future
returns, especially Heating Oil, Sugar, and Wheat. Specifically, once a jump in gold
iew
future returns occur, this increases the probability of a jump in other commodity futures
returns. This result is in line with and adds to Nekhili et al. (2021) who report evidence
of a unidirectional co-movement running from gold to the remaining precious metals.
The colours of the edges connecting the other commodity futures to Gold, ranging from
deep blue to red, reflect the varying degrees of influence that Gold jumps have on the
jumps in commodity futures returns. Commodity futures with edges closer to deep blue
v
are more likely to be influenced by jumps in Gold. Interestingly, Cocoa jumps have a
re
week influence on the jumps in some metals, such as Platinum, Lead, Aluminum, and
Tin.
er
pe
ot
tn
rin
Figure 2. Co-jump network based on MST for the full sample period
ep
Notes: See notes to Figure 1.
5.3. Co-jump network analysis II: Inter-industry group

Pr
In the previous subsection 4.2, we analyse the co-jump network based on the total 36
commodity futures. Now we focus on four co-jump networks based Inter-industry
group, which can capture commodity futures returns co-jump dependency within
different industry group from July 5, 2016 to July 1, 2023. Figure 3 shows the four co-
jump networks based Inter-industry group. Each vertex in the subgraphs represent an
ed
individual commodity future. The colour distribution, ranging from deep blue to deep
red, reflects the intensity of co-jump relationships within the commodity group.
The general perception among investors is that commodity futures within the same
commodity group may be more prone to co-jumping behaviour, which is due to the
iew
higher similarity in the demand and supply of commodities belonging to the same group
and the factors that affect the price formation of these commodities. From the four co-
jump networks, we reveal some meaningful phenomena in various Inter-industry group.
For the Metals Futures group, Nickel is more susceptible to other commodity futures,
especially Zinc and Palladium. Not surprisingly, Gold has the lowest jump sensitivity,
v
probably because of its hedging and safe-haven properties. For the Agricultural Futures
group, Lumber is more susceptible to other commodity futures, especially US C Coffee
re
and Orange Juice. London Coffee has the strongest jump dependence on London Sugar,
whereas Cotton has the most stable jump sensitivity. For the Energy Futures group,
RBOB Gasoline and London Gasoline have relatively lower jump sensitivity, whereas
WTI Crude Oil and Brent Crude Oil have the strongest jump dependence because they
er
are more sensitive to global risk factors and macroeconomic conditions. In this regard,
Bouri and Gupta (2020) underline the high sensitivity of crude oil to macroeconomic
news while associating between jumps in the volatility of crude oil and macroeconomic
pe
news surprises. For the Grain and Cereal Futures group, London Wheat and Rice have
relatively lower jump sensitivity, whereas US Wheat and Oats have the strongest jump
dependence.
ot
tn
rin
ep
Figure 3. Four co-jump networks based Inter-industry group

Pr
Notes: The connection between pairs of commodity futures indicates the strength of co-jumps in different groups.
The colour gradient in the legend at the top right represents the intensity of co-jump relationships, ranging from deep
blue to deep red. The colour gradient in the legend at the bottom right represents the width of the connection between
pairs of commodity futures, ranging from light grey to black.
Figure 4 presents the four co-jump networks after filtering some unimportant edges.
It shows more detailed information compared to Figure 2, especially the jump
ed
connectivity within each commodity group. The topology can reveal the jump
dependencies between the nodes within different commodity groups in the co-jump
network. Gold, Live Cattle, Carbon Emissions and London Wheat are the core nodes
within their respective groups and play an important role in the connectivity of the
iew
entire co-jump network. We also find that there are relatively independent groups of
nodes in Four co-jump networks, such as Sliver, US C Coffee, Natural Gas and Wheat.
These results are important for studying the jump dependency based co-jump networks
among different commodity group.
v
re
er
pe
ot
Figure 4. Four co-jump networks based on MST

tn
5.4. Dynamics ranking of co-jump influence: total 36 commodity futures

The rolling-sample ranking of the co-jump centrality analysis provides insight into
the dynamics of the co-jump influence across 36 commodity future markets over time
rin
(from 2017 to 2023). Figure 5 shows co-jump network centrality rankings represented
by colours ranging from deep blue to yellow, indicating rankings from 1 to 36.
During the full period, the Energy group relatively ranked highest in terms of co-
ep
jump network centrality. This suggests that Energy futures returns jump during the full
period had more significant impact on other commodity futures returns, making them
more susceptible to jumps in Energy futures returns. One potential explanation for this
phenomenon is that the jumping behaviour of Energy futures returns has amplification
Pr
and incentive effects. During the COVID-19 and the Russia-Ukraine conflict, Natural
Gas, RBOB Gasoline, and WTI Crude Oil consistently maintained their dominant
position in the centrality ranking, which can be attributed to several factors. Firstly,
Natural Gas, RBOB Gasoline and WTI Crude Oil, as energy resources, have supply
chain relationships with other commodities. Therefore, when the prices of Natural Gas,
ed
RBOB Gasoline and WTI Crude Oil jump, the supply chains of other commodities are
affected6, causing the prices of other commodities to jump as well. Secondly, extreme
price fluctuations in Natural Gas, RBOB Gasoline and WTI crude oil during crisis
periods reflect changes in global financial markets and fluctuations in market sentiment,
iew
so this may further alter investor expectations for other commodities and have an impact
on commodity futures returns jumps. Thirdly, in a time of high geopolitical risk such as
during the Russia-Ukraine war (Wang et al., 2022), Natural Gas, RBOB Gasoline and
WTI Crude Oil emerge as hedging assets (Mensi et al., 2021). Furthermore, some
Agricultural and Metals commodity futures such as Lumber, Nickel and Palladium have
significantly higher centrality rankings after the Russia-Ukraine conflict. The above-
v
mentioned results offer insights regarding the time-varying co-jumps behaviour in the
re
commodity markets, which nicely extend the previous findings of Nguyen and
Prokopczuk (2019), Gupta and Bouri (2020), and Semeyutin et al. (2021).
er
pe
ot
tn
rin
ep
Figure 5. Dynamic ranking of co-jump influence for the 36 commodity futures

Notes: In the colour bar (see the legend on the right), each colour stands for a ranking of co-jump influence, varying
from 1 to 36.
Pr
5.5. Dynamics ranking of co-jump influence: Inter-industry group
6
Gozgor et al. (2023) relate the global supply chain pressure with the commodity markets.
In the previous subsection 5.4, we analyse the dynamics ranking of co-jump
influence based on the total 36 commodity futures. Here, we focus on the dynamics
ed
ranking of co-jump influence based Inter-industry group, which can capture the
dynamic co-jump influence within different commodity groups. Compared to a
comprehensive analysis of all commodity futures, an inter-group study has more
significant theoretical and practical advantages. First, we can better understand and
iew
capture the co-jump dependency and interaction effects between various commodity
groups and can more accurately analyse and explain the dynamic jump impacts across
commodity groups. Second, we can better explain the dynamics of the co-jump impact
in the commodity markets. Finally, based on the co-jump impact between different
commodity groups, investors can better develop trading strategies and optimize their
portfolios.
v
Figure 6 presents the centrality rankings of co-jump networks for various commodity
re
groups, represented by a colour scheme ranging from deep blue to yellow. Although
Gold did not achieve an ideal centrality ranking in the comprehensive analysis of all
commodity futures, it maintained its dominant position in the centrality ranking within
the Metals Futures group during the COVID-19 pandemic and the Russia-Ukraine
er
conflict. This indicates that the returns jump of Gold during crisis periods had a
significant impact on other commodities in the Metals Futures group, making them
more susceptible to jumps when Gold returns jump. In contrast to the results obtained
pe
from the comprehensive analysis of the entire commodity futures market, WTI crude
oil and natural gas exhibited lower dynamic rankings within the Energy group,
indicating their relatively lower propensity for jump spillovers within the Energy group.
However, during the COVID-19 pandemic, Carbon Emissions continued to maintain
its dominant position within the Energy group, while Heating Oil maintained its
dominant position during the Russia-Ukraine conflict. It is worth noting that Rice
ot
maintained a higher dynamic ranking within the Grain and Cereal Futures group,
despite its inconspicuous position in the overall commodity futures markets.
tn
rin
ep
Pr
Figure 6. Dynamic ranking of co-jump influence for different industry groups
ed
Notes: In the colour bar (see the legend), each colour stands for a ranking of co-jump influence.
6. Application to portfolio allocation and volatility forecasting

6.1 Application to portfolio allocation
iew
In practical applications, investors may be more concerned about constructing
optimal investment portfolios to maximize their returns. Especially in periods of crisis
when frequent price jumps occur (Arouri et al., 2019), investors can choose commodity
futures with lower jump dependencies to diversify risks and achieve relatively higher
portfolio returns. To demonstrate the economic value of the co-jump network in
v
commodity future markets, we develop different daily rebalanced mean-variance
strategies catering to the needs of different investors based on the dynamic ranking of
re
co-jump influence for each trading day throughout the study period. In the portfolio
analysis, we differentiate between two investment scenarios: one for investors in the
commodity industry and another for overall commodity investors.
6.1.1 Mean-variance portfolio construction

er
Similar with Zhang et al. (2023) and based upon the assumption that 𝐸(Σ̂𝑡 |Σ̂𝑡−1 ) =
Σ̂𝑡−1 , we estimate the mean-variance portfolio (MV) weights 𝑤𝑡 by solving the
pe
following constrained optimization question for day t:
̂𝒕−𝟏 𝒘𝒕
𝐦𝐢𝐧 𝒘𝑻𝒕 𝚺
𝒘𝒕
ot
⃗ =𝟏
𝒔𝒖𝒃𝒋𝒆𝒄𝒕 𝒕𝒐 𝒘𝑻𝒕 𝟏 (9)
‖𝒘𝒕 ‖𝟏 ≤ 𝒍,
tn
where Σ̂𝑡−1 denotes the covariance matrix. 𝑙 ≥ 0 denotes the level of leverage, and
if 𝑙 = 1, leverage is not allowed and all weights are non-negative.
The daily portfolio return can be calculated as:
rin
𝒓𝑴𝑽
𝒕 = 𝒘𝑻𝒕 𝒓𝒕 (10)
where 𝒓𝑡 is a vector of commodity futures logarithmic returns, and 𝒘𝑇𝑡 is a vector of
weight estimation values.
ep
The evaluation metric of portfolio performance are the annualized volatility and Sharpe
ratio. The annualized Sharpe ratio (Sharpe, 1994) is calculated as:
̅̅̅̅̅̅
𝒓𝑴𝑽 ×𝟐𝟓𝟐
𝒔𝒓𝑴𝑽 = (11)
𝝈𝑴𝑽
Pr
where 𝑟̅̅̅̅̅
𝑀𝑉 denotes the average daily return of the mean-variance portfolio and
𝜎 𝑀𝑉 = 𝑠𝑡𝑑(𝑟𝑡𝑀𝑉 × √252) is the annualized volatility.
6.1.2. Analysis of portfolio performance: total 36 commodity futures
ed
Here, we examine various mean-variance portfolio strategies and leverage
constraints for optimization. This optimization is based on dynamically ranking the
influence of co-jumps across 36 commodity futures markets over a specific time period
(from 2017 to 2023). To ensure an out-of-sample analysis, we calculate the mean-
variance portfolio weights for day t+1 using covariance estimates derived from a rolling
iew
window of 300 observations, covering the time period from t-299 to t. Additionally, we
construct a co-jump network based on the rolling historical window of 300 samples (t-
299:t) and select different numbers of commodity futures to build the investment
portfolio strategy for day t+1, based on the ranking of co-jump influence. This implies
that we periodically update the strategy using a rolling window of historical data.
v
The annualized Sharpe ratio is presented in Table 3 for different mean-variance
re
portfolios, which can further compare after adjusting for portfolio returns. The co-jump
dependency-driven portfolio outperform the baseline mean-variance portfolio.
According to the ranking of co-jump influence across 36 commodity future markets,
portfolios built on the top10 commodity futures in terms of centrality exhibit highest
Sharpe ratios compared to the other mean-variance portfolios. However, increasing the
er
number to 20 deteriorates the Sharpe ratio. Portfolios built on the Top20 commodity
futures demonstrate lower Sharpe ratios than the baseline mean-variance portfolio. As
leverage constraints are relaxed, all mean-variance portfolios show an increasing
pe
annualized Sharpe ratio. We offer explanations for these findings: The selection of the
top five, ten, and fifteen commodity futures, which possess the significant influence on
the jumping behaviour of other commodity futures returns while being less susceptible
to such behaviour themselves, implies their relative jump independence from other
commodity futures returns. Consequently, when combined into a portfolio, these
ot
commodity futures achieve improved risk diversification, thereby reducing portfolio

volatility. In contrast, commodity futures with lower centrality are more impacted by
other commodity futures. It is noteworthy that, with Top20 commodity futures, the co-
tn
jump dependency among commodity futures could be overestimated, leading to

increased portfolio volatility and uncertainty, and results in low level of Sharpe ratios.
Additionally, we explore the trajectory of portfolio gains by plotting the cumulative
returns of multiple mean-variance portfolio strategies in Figure 7. The portfolios
rin
derived from the co-jump network model outperform the baseline strategy by attaining
higher profits while experiencing less volatility and shorter downside periods. The Top5
Rank-MV portfolio strategy, which considers the top five commodity futures based on
the ranking of co-jump influence, exhibits significant profit differentials compared to
ep
other investment portfolios. Importantly, under periods of crisis such as the COVID-19
pandemic or the Russia-Ukraine conflict, the Top5 Rank-MV portfolio rapidly expands
this profit differential. However, as the portfolio includes more commodity futures, the
profit differential gradually diminishes and may even fall below the benchmark strategy.
Pr
We provide explanations for these findings. Firstly, the top five commodity futures that
are ranked based on co-jump influence may be less susceptible to the jumping
behaviour of other commodity futures returns. This can result in higher profits in the
Top5 Rank-MV portfolio compared to other portfolios, particularly during crisis
periods. Secondly, following crisis periods, certain commodities may experience
ed
significant price jumps due to supply disruptions, changes in demand, or geopolitical
factors. The Top5 Rank-MV portfolio may effectively capture these returns jumps,
leading to increased profits. Finally, as more commodity futures are added to the
portfolio, such as the Top20 Rank-MV portfolio, the portfolio may exhibit higher
iew
correlation and a greater tendency to jump together, which can result in lower profits.
Table 3. Annualized Sharpe ratio of mean-variance portfolios

Leverage Baseline-MV Top5 Rank-MV Top10 Rank- Top15 Rank- Top20 Rank-
MV MV MV
v
1 0.9571 1.2932 1.4022 1.1896 0.5053
2 1.1485 1.3415 1.4654 1.2852 0.6204
re
3 1.2123 1.3576 1.4865 1.3171 0.6588
4 1.2442 1.3656 1.4970 1.3331 0.6780
5 1.2633 1.3705 1.5033 1.3426 0.6895
6 1.2761 er 1.3737 1.5075 1.3490 0.6972
7 1.2852 1.3760 1.5105 1.3536 0.7026
8 1.2921 1.3777 1.5128 1.3570 0.7068
9 1.2974 1.3791 1.5146 1.3596 0.7100
pe
10 1.3016 1.3801 1.5160 1.3618 0.7125
Notes: This table shows the annualized Sharpe ratio of mean-variance portfolios under different constrained leverage.
Baseline-MV indicates a portfolio strategy that considers the all commodity futures to construct optimal investment
portfolios. Top5 Rank-MV indicates a portfolio strategy that considers the top five commodity futures in terms of
ranking of co-jump influence. Top10 Rank-MV indicates a portfolio strategy that considers the top ten commodity
ot
futures in terms of ranking of co-jump influence. Top15 Rank-MV indicates a portfolio strategy that considers the
top fifteen commodity futures in terms of ranking of co-jump influence. Top20 Rank-MV indicates a portfolio
strategy that considers the top twenty commodity futures in terms of ranking of co-jump influence.
tn
rin
ep
Pr
Figure 7. Cumulative returns of portfolios
6.2 Application to volatility forecasting
ed
The present study examines the co-jump behavior of commodity futures in 36
commodities or different inter-industry groups using co-jump network models. To
highlight the significance of each individual commodity and commodity group, we
construct static MSTs as depicted in Figure 2 and Figure 4. Moreover, we provide
insights into the stability and evolution of co-jump dependencies based on Eigenvector
iew
centrality, which measures the influence of nodes in the co-jump network (Figures 5-
6). In practical applications, investors are often more concerned with asset pricing,
investment portfolios, and risk management. However, accurately predicting financial
asset volatility is a key input for addressing these real-world applications. Therefore,
we develop a novel perspective for volatility forecasting which reflects the economic
v
value of the co-jump network in commodity futures markets. Specifically, we utilize
information on the centrality measures within the co-jump network structure, namely
re
Eigenvector centrality, Closeness centrality, and Normalized tree length, to forecast the
volatility of the S&P 500 index.
6.2.1 Model specification
er
Realized volatility is a widely employed measure to capture the true volatility of the
underlying asset, as it filters out market noise (Andersen et al., 2003). Specifically, for
a given day t, the formula for realized volatility is as follows:
pe
𝑹𝑽𝒕 = ∑𝑴 𝟐
𝒋=𝟏 𝒓𝒕,𝒋 , 𝒕 = 𝟏, . . , 𝑻 （12）
Where, M denotes the intraday trading frequency of the underlying asset in day t,
2
𝑟𝑡,𝑗denotes the square of the intraday logarithmic return on the underlying asset at the
j-th in day t.
ot
In line with prior research (Wang et al., 2018), the autoregressive model has been
extensively employed in the context of volatility forecasting. This model is expressed
as follows:
tn
𝒑−𝟏
𝑽𝒕+𝟏 = 𝝎 + ∑𝒊=𝟎 𝜶𝒊 𝑽𝒕−𝒊 + 𝜷𝑽𝒕,𝒌 + 𝜺𝒕+𝟏 （13）
Where, 𝑉𝑡 = log (𝑅𝑉𝑡 ) and 𝑉𝑡,𝑘 are the log realized volatilities of the predicted
and forecast variables, respectively. 𝜀𝑡+1 is the error term and follows an independent
rin
and identically normal distribution. The long lag length p can reflect the strong
autocorrelation of the predictor variables, and is set equal six by Wang et al., (2018).
Based the researches of Ji et al. (2018), we build three measures of centrality as:
ep
𝟏
Eigenvector centrality: 𝑪𝒆 (𝝂𝒊 ) = 𝝀 ∑𝒏𝒋=𝟏 𝚨𝒋,𝒊 𝑪𝒆 (𝝂𝒋 ) (14)
Closeness centrality: 𝑪𝑪(𝒊) = ∑(𝒊,𝒋) 𝑹𝒊𝒋 , 𝒊 ≠ 𝒋 （15）

Pr
𝟏
Normalized tree length: 𝑳(𝒕) = 𝑵−𝟏 ∑𝒆(𝒊𝒋)∈𝑴𝑺𝑻 𝒆𝒊𝒋 , （16）
Where, 𝝀 is a constant, 𝑪𝒆 = (𝑪𝒆 (𝝂𝟏 ), 𝑪𝒆 (𝝂𝟐 ), … , 𝑪𝒆 (𝝂𝒏 ))𝑻 is the center vector of
ed
all vertices, 𝑹𝒊𝒋 𝑖𝑠the shortest path from i to j in the MST, 𝒆𝒊𝒋 is the edge in the MST.
In this study, we employ two different techniques, namely rolling and recursive
window estimation, to forecast out-of-sample volatility. To accomplish this, we divide
the full sample T of each predictor variable and stock volatility series into an in-sample
iew
portion, consisting of M observations, and an out-of-sample portion, consisting of T -
M observations. The equation for a one-step backward prediction is as follows:
̂ 𝑴+𝟏 = 𝝎 𝒑−𝟏 ̂ 𝑴 𝑽𝑴,𝒌
𝑽 ̂ 𝑴 + ∑𝒊=𝟎 𝜶
̂ 𝒊,𝑴 𝑽𝑴−𝒊 + 𝜷 (17)
To assess the forecasting performance of the autoregressive (AR) model, researchers
v
often rely on the out-of-sample 𝑹𝟐𝑶𝑶𝑺 metric. This metric quantifies the out-of-sample
predictive ability of two nested mean models (Rapach et al., 2010). Specifically, the
re
out-of-sample 𝑹𝟐𝑶𝑶𝑺 is defined as follows:
𝑴𝑺𝑷𝑬
∆𝑹𝟐𝑶𝑶𝑺 = 𝟏 − 𝑴𝑺𝑷𝑬𝒎𝒐𝒅𝒆𝒍 , (18)
𝒃𝒆𝒏𝒄𝒉
1
er
Where, 𝑀𝑆𝑃𝐸𝑖 = 𝑇−𝑀 ∑𝑇𝑡=𝑀+1(𝑉𝑡 − 𝑉̂𝑡,𝑖 )2 (𝑖 = 𝑚𝑜𝑑𝑒𝑙, 𝑏𝑒𝑛𝑐ℎ) . 𝑉𝑡 𝑎𝑛𝑑 𝑉̂𝑡,𝑖 are the
2
true and predicted values of log volatility, respectively. A positive ∆𝑅𝑂𝑂𝑆 value
suggest that the model of interest have higher prediction accuracy than the bench model.
pe
Furthermore, the Clark and West (2007) statistic is employed to test the null
2 2
hypothesis and alternative hypotheses, 𝐻0 : ∆𝑅𝑂𝑂𝑆 ≥ 0 𝑣. 𝑠. 𝐻1 : ∆𝑅𝑂𝑂𝑆 < 0 . The
Clark and West statistic has the form as follow:
̂ 𝒕,𝒃𝒆𝒏𝒄𝒉 )𝟐 − (𝑽𝒕 − 𝑽
𝒇𝒕 = (𝑽𝒕 − 𝑽 ̂ 𝒕,𝒎𝒐𝒅𝒆𝒍 )𝟐 + (𝑽
̂ 𝒕,𝒃𝒆𝒏𝒄𝒉 −𝑽
̂ 𝒕,𝒎𝒐𝒅𝒆𝒍 )𝟐 (19)
ot
The t-statistic is obtained by regressing {𝑓𝑡 : 𝑡 = 𝑀 + 1, … , 𝑇} on a constant.

tn
6.2.2 Dynamic analysis and forecasting results

Similar to Subsection 5.4, we employ a rolling window approach based on 300 days
to estimate the centrality measures. For each centrality measure, we focus solely on the
rin
evolution of the entire market and the four respective groups. Figure A3 illustrates the
Eigenvector centrality index for each commodity type and overall commodities, based
on the time-varying MST of the co-jump network. The Metals group exhibits relative
stability prior to the COVID-19 period, followed by an upward trend. The Agricultural
ep
and Grain and Cereal groups display similar influence trends, but behave in contrast to
the Energy group during the period from 2018 to 2019. Additionally, the Energy group
experiences significant fluctuations, whereas the Metals, Agricultural, and Grain and
Cereal groups exhibit narrower variability. Similarly, Figures A4 and A5 depict the
Pr
Closeness centrality index and Normalized tree length index, respectively, for each
commodity type and overall commodities, based on the time-varying MST of the co-
jump network. The Closeness centrality index measures the degree of co-jumping
between variables, with higher values indicating a higher degree of co-jumping with
other variables. As shown in Figure A4, four groups generally exhibit similar trends.
ed
All four groups experience oscillations prior to the COVID-19 pandemic, with the
Energy group reaching a major peak during the pandemic and hitting a low point in
mid-2021; it also reaches a peak during the Russia-Ukraine conflict. In Figure 5, the
Normalized tree length index measures the evolution of the importance of each group.
iew
We can observe that prior to the COVID-19 pandemic, all four groups fluctuate within
a normal range. The occurrence of the pandemic leads to abnormal upward fluctuations
in the Energy and Agricultural groups, and their importance continues to increase in the
later stages of the pandemic, reaching a peak in mid-2021.
Previous studies have shown that commodity futures information can effectively
v
forecast US stock market volatility (Liang et al., 2020; Liu and Guo, 2022). However,
there is a lack of understanding regarding the underlying network structure that drives
re
this forecasting ability. Therefore, we examine the forecastability of S&P 500 volatility
based on the centrality variable of the common jump network in the commodity markets.
By doing so, we provide valuable insights into the relationship between commodity
network structure and stock market volatility forecasting. Table 4 presents the
er
forecasting results based on three centrality variables for each commodity type and
overall commodities. The results are summarized are as follows: firstly, compared to
the benchmark model, the Eigenvector centrality index and Closeness centrality index
pe
2
significantly improve the out-of-sample 𝑅𝑂𝑂𝑆 based on the rolling window approach.
The improvement in predictive ability is statistically significant based on the Clark and
West (2007) statistic. These pieces of evidence suggest that centrality information from
the co-jump network is highly predictive of US stock market volatility. Secondly, the
predictive ability varies across different groups, with the Energy and Agricultural
groups exhibiting stronger predictive power. However, the forecasting performance
ot
based on the Recursive window approach is poor, as it is consistently outperformed by

the benchmark model. This may indicate that the predictive power of centrality
information in the co-jump network is relatively short-lived.
tn
Table 4. Out-of-sample forecasting results

Eigenvector centrality Closeness centrality Normalized tree length
rin
Rolling window ∆𝑹𝟐𝑶𝑶𝑺 ∆𝑹𝟐𝑶𝑶𝑺 ∆𝑹𝟐𝑶𝑶𝑺

Metals Group 1.0465*** 0.4513*** 0.2694*
Agricultural Group 1.9617*** 1.2962*** -0.4071
ep
Energy Group 1.9181*** 2.0624*** -0.5459

Grain and Cereal Group 1.3098*** 1.1637*** -0.2202
Overall 1.7515*** 1.3527*** -0.6289
Recursive window ∆𝑹𝟐𝑶𝑶𝑺 ∆𝑹𝟐𝑶𝑶𝑺 ∆𝑹𝟐𝑶𝑶𝑺
Pr
Metals Group -0.4859 -0.12808 -0.1159

Agricultural Group -0.7427 -0.3083 -0.1859
Energy Group -0.2972 -0.6619 -0.1648
Grain and Cereal Group -1.0400 -0.2153 -0.1583
Overall -0.7158 -0.3608 -0.0266
ed
Notes: This table presents the out-of-sample forecasting results for the predictive regressions using three centrality
variables. The forecasts are generated using both the rolling window and recursive window approaches, with an
2
initial sample period of 300 days. The table reports the out-of-sample 𝑅𝑂𝑂𝑆 , which is defined as the percentage
reduction in mean squared predictive error (MSPE) of the centrality model compared to the benchmark AR(6) model.
iew
The asterisks *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels based the p-values of
Clark and West (2007) (C-W) tests, respectively.
7．Conclusion
This paper addresses the question of how co-jump in 36 commodity futures are
v
transmitted by constructing co-jump network models, which offers a new perspective
for investigating co-jump dependencies among commodity returns. We also capture the
re
volatility forecasting implications, which helps in better understanding the implications
of co-jump dynamics on the performance of the US stock market volatility forecasting
based on centrality information from the co-jump network.
Our results show that gold has the strongest influence, indicating that the jumps in
er
most other commodity futures returns are influenced by the jumps in gold returns. This
may seem contradictory if the jumps occur simultaneously. However, it is important to
note that the influence of gold on other commodities in this context refers to the overall
pe
relationship between their price movements. When gold experiences a jump, it suggests
that there is a higher likelihood of jumps occurring in other commodity futures as well.
This does not necessarily mean that gold directly causes jumps in other commodities or
that the jumps occur at the exact same time. Instead, it suggests a broader relationship
between gold and other commodities during periods of market stress or significant
ot
events. Therefore, gold serves as an indicator or leading indicator for jumps in other
commodity futures, rather than a direct causal influence. The commodity futures within
the Energy group have a higher likelihood of experiencing co-jumps. From the co-jump
tn
networks analysis based Inter-industry group, it appears that co-jump heterogeneity

exists among pairs of commodity futures returns with different commodity groups. The
impact of co-jump centrality varies over time, reflecting its sensitivity to crisis periods;
we also discuss the dynamics ranking of co-jump influence during some crises
rin
including COVID-19 and Russia-Ukraine conflict and find that co-jump dependencies
have diverse change, which provide useful insights for portfolio management such as
crisis-specific adjustments, risk diversification, and asset allocation. We further study
the role of centrality information from the co-jump network in forecasting US stock
ep
market volatility. Compared to the benchmark model, the Eigenvector centrality index
2
and Closeness centrality index significantly improve the out-of-sample 𝑅𝑂𝑂𝑆 based on
the rolling window approach. Our study also indicates that constructing a portfolio with
a larger number of commodity futures may not always be the most effective approach,
Pr
but rather it may be sufficient to consider just five or ten commodities to outperform
the baseline strategy. This aligns with the concept of “less is more” in value investing,
where quality of investments outweighs quantity, which is applicable in the commodity
market, and we explain it from the perspective of the ranking of co-jump centrality. The
findings also suggest that, when modelling the dynamic behaviour of commodity
ed
futures returns, it is relevant to consider not only the presence of co-jumps but also how
commodity futures co-jump across time and which commodity futures have a strong
co-jump impact on others. Furthermore, it is relevant to reflect the results of co-jumps
in the various commodity groups, which suggest that not all commodity groups are
iew
alike when it comes to the co-jumping behaviour of commodities returns.
Future research can extend our current findings by decomposing the co-jump
network into positive and negative components or by considering the co-jump impact
between stocks and commodity futures within a network model and the portfolio
implications. In another possible extension, researchers can build on the highly
v
interrelated co-jump commodity pairs to assess whether traders and investors can
exploit co-jump events to generate abnormal profits.
re
References
Abid, I., Dhaoui, A., Goutte, S., & Guesmi, K. (2020). Hedging and diversification
across commodity assets. Applied Economics, 52(23), 2472-2492.
er
Aït-Sahalia, Y., & Xiu, D. (2016). Increased correlation among asset classes: Are
volatility or jumps to blame, or both?. Journal of Econometrics, 194(2), 205-219
Andersen, T. G., Bollerslev, T., Diebold, F. X., & Labys, P. (2003). Modeling and
pe
forecasting realized volatility. Econometrica, 71(2), 579-625.
Andersen, T. G., Bollerslev, T., & Diebold, F. X. (2007). Roughing it up: Including
jump components in the measurement, modeling, and forecasting of return
volatility. The review of economics and statistics, 89(4), 701-720.
Arouri, M., M’saddek, O., Nguyen, D. K., & Pukthuanthong, K. (2019). Cojumps and
ot
asset allocation in international equity markets. Journal of Economic Dynamics and

Control, 98, 1-22.
Bahel, E., Marrouch, W., & Gaudet, G. (2013). The economics of oil, biofuel and food
tn
commodities. Resource and energy economics, 35(4), 599-617.

Bandi, F. M., & Reno, R. (2016). Price and volatility co-jumps. Journal of Financial
Economics, 119(1), 107-146
Barndorff-Nielsen, O. E., & Shephard, N. (2004). Power and bipower variation with
rin
stochastic volatility and jumps. Journal of financial econometrics, 2(1), 1-37.

Bates, D, S., (2019). How crashes develop: Intradaily volatility and crash evolution.
Journal of Finance. 74(1), 193-238.
Bates, D. S. (1991). The crash of ʼ87: was it expected? The evidence from options
ep
markets. The journal of finance, 46(3), 1009-1044.

Baur, D. G., & Lucey, B. M. (2010). Is gold a hedge or a safe haven? An analysis of
stocks, bonds and gold. Financial review, 45(2), 217-229.
Bollerslev, T., & Todorov, V. (2011). Tails, fears, and risk premia. The Journal of
Pr
Finance, 66(6), 2165-2211.

Bollerslev, T., Law, T. H., & Tauchen, G. (2008). Risk, jumps, and
diversification. Journal of Econometrics, 144(1), 234-256.
Bollerslev, T., Todorov, V., & Xu, L. (2015). Tail risk premia and return
predictability. Journal of Financial Economics, 118(1), 113-134.
ed
Bonacich, P. (1987). Power and centrality: A family of measures. American journal of
sociology, 92(5), 1170-1182.
Bouri, E., & Gupta, R. (2020). Jumps in energy and non-energy commodities. OPEC
Energy Review, 44(1), 91-111.
iew
Bouri, E., Lucey, B., Saeed, T., & Vo, X.V. (2021). The realized volatility of commodity
futures: Interconnectedness and determinants. International Review of Economics
and Finance, 73, 139-151.
Bouri, E., Nekhili, R., and Todorova, N. (2023). Dynamic co-movement in major
commodity markets during crisis periods: A wavelet local multiple correlation
analysis. Finance Research Letters, 55, 103996.
v
Caporin M, Kolokolov A, & Renò R. (2017). Systemic co-jumps. Journal of Financial
re
Economics, 126(3), 563-591.
Chatrath, A., & Song, F. (1999). Futures commitments and commodity price
jumps. Financial Review, 34(3), 95-111.
Chen, Y., Ma, F., & Zhang, Y. (2019). Good, bad cojumps and volatility forecasting:
New evidence from crude oil and the US stock markets. Energy Economics, 81, 52-
62.
er
Chevallier, J., & Ielpo, F. (2014). Twenty years of jumps in commodity
markets. International Review of Applied Economics, 28(1), 64-82.
pe
Chevallier, J. (2009). Carbon futures and macroeconomic risk factors: A view from the
EU ETS. Energy Economics, 31(4), 614-625.
Christoffersen, P., Lunde, A., & Olesen, K. V. (2019). Factor structure in commodity
futures return and volatility. Journal of Financial and Quantitative Analysis, 54(3),
1083–1115.
ot
Clements, A., & Liao, Y. (2017). Forecasting the variance of stock index returns using
jumps and cojumps. International Journal of Forecasting, 33(3), 729-742.
Cremers, M., Halling, M., & Weinbaum, D. (2015). Aggregate jump and volatility risk
tn
in the cross‐section of stock returns. The Journal of Finance, 70(2), 577-614.

Diewald, L., Prokopczuk, M., & Simen, C. W. (2015). Time-variations in commodity
price jumps. Journal of Empirical Finance, 31, 72-84.
Ding, Y., Li, Y., Liu, G., & Zheng, X. (2023). Stock co-jump networks. Journal of
rin
Econometrics.
Dutta, A., Bouri, E., & Roubaud, D. (2021). Modelling the volatility of crude oil returns:
jumps and volatility forecasts. International Journal of Finance and Economics,
26(1), 889-897.
ep
Erten, B., & Ocampo, J. A. (2013). Super cycles of commodity prices since the mid-
nineteenth century. World development, 44, 14-30.
Figuerola-Ferretti, I., & Gonzalo, J. (2010). Modelling and measuring price discovery
in commodity markets. Journal of Econometrics, 158(1), 95-107.
Pr
Gilder, D., Shackleton, M. B., & Taylor, S. J. (2014). Cojumps in stock prices:
Empirical evidence. Journal of Banking & Finance, 40, 443-459.
Gozgor, G., Khalfaoui, R., & Yarovaya, L. (2023). Global supply chain pressure and
commodity markets: Evidence from multiple wavelet and quantile connectedness
ed
analyses. Finance Research Letters, 54, 103791.
Jawadi, F., Louhichi, W., Ameur, H. B., & Ftiti, Z. (2019). Do jumps and co-jumps
improve volatility forecasting of oil and currency markets?. The Energy
Journal, 40(Special Issue).
iew
Ji, Q., Bouri, E., Roubaud, D., & Shahzad, S. J. H. (2018). Risk spillover between
energy and agricultural commodity markets: A dependence-switching CoVaR-
copula model. Energy Economics, 75, 14-27.
Ji, Q., Bouri, E., Roubaud, D., & Kristoufek, L. (2019). Information interdependence
among energy, cryptocurrency and major commodity markets. Energy
Economics, 81, 1042-1055.
v
Laurent, S., Lecourt, C., & Palm, F. C. (2016). Testing for jumps in conditionally
re
Gaussian ARMA–GARCH models, a robust approach. Computational Statistics &
Data Analysis, 100, 383-400.
Laurini, M. P., Mauad, R. B., & Aiube, F. A. L. (2020). The impact of co-jumps in the
oil sector. Research in International Business and Finance, 52, 101197.
Liang, C., Ma, F., Li, Z., & Li, Y. (2020). Which types of commodity price information
er
are more useful for predicting US stock market volatility?. Economic Modelling, 93,
642-650.
Liu, G., & Guo, X. (2022). Forecasting stock market volatility using commodity futures
pe
volatility information. Resources Policy, 75, 102481.
Mancini, C. (2009). Non-parametric threshold estimation for models with stochastic
diffusion coefficient and jumps. Scandinavian Journal of Statistics, 36(2), 270-296.
Mantegna, R. N. (1999). Hierarchical structure in financial markets. The European
Physical Journal B-Condensed Matter and Complex Systems, 11, 193-197.
ot
Maslyuk-Escobedo, S., Rotaru, K., & Dokumentov, A. (2017). News sentiment and
jumps in energy spot and futures markets. Pacific-Basin Finance Journal, 45, 186-
210.
tn
Mirantes, A. G., Población, J., & Serna, G. (2012). The stochastic seasonal behaviour
of natural gas prices. European financial management, 18(3), 410-443.
Mensi, W., Rehman, M. U., & Vo, X. V. (2021). Dynamic frequency relationships and
volatility spillovers in natural gas, crude oil, gas oil, gasoline, and heating oil
rin
markets: Implications for portfolio management. Resources Policy, 73, 102172.

Nekhili, R., Sultan, J., & Mensi, W. (2021). Co-movements among precious metals and
implications for portfolio management: A multivariate wavelet-based dynamic
analysis. Resources Policy, 74, 102419.
ep
Nguyen, D. B. B., & Prokopczuk, M. (2019). Jumps in commodity markets. Journal of

Commodity Markets, 13, 55-70.
Nguyen, D. K., & Walther, T. (2020). Modeling and forecasting commodity market
volatility with long-term economic and financial variables. Journal of Forecasting,
Pr
39(2), 126–142.
Pindyck, R. S. (2001). The dynamics of commodity spot and futures markets: a
primer. The energy journal, 22(3).
Prokopczuk, M., Symeonidis, L., & Wese Simen, C. (2016). Do jumps matter for
volatility forecasting? Evidence from energy markets. Journal of Futures
ed
Markets, 36(8), 758-792.
Rehman, M. U., Bouri, E., Eraslan, V., & Kumar, S. (2019). Energy and non-energy
commodities: An asymmetric approach towards portfolio diversification in the
commodity market. Resources Policy, 63, 101456.
iew
Rapach, D. E., Strauss, J. K., & Zhou, G. (2010). Out-of-sample equity premium
prediction: Combination forecasts and links to the real economy. The Review of
Financial Studies, 23(2), 821-862.
Rossen, A. (2015). What are metal prices like? Co-movement, price cycles and long-
run trends. Resources Policy, 45, 255-276.
Semeyutin, A., & Downing, G. (2022). Co-jumps in the US interest rates and precious
v
metals markets and their implications for investors. International Review of
re
Financial Analysis, 81, 102078.
Semeyutin, A., Gozgor, G., Lau, C. K. M., & Xu, B. (2021). Effects of idiosyncratic
jumps and co-jumps on oil, gold, and copper markets. Energy Economics, 104,
105660.
Sévi, B. (2015). Explaining the convenience yield in the WTI crude oil market using
er
realized volatility and jumps. Economic Modelling, 44, 243-251.
Sharpe, W. F. (1998). The sharpe ratio. Streetwise–the Best of the Journal of Portfolio
Management, 3, 169-185.
pe
Shrestha, K. (2014). Price discovery in energy markets. Energy Economics, 45, 229-
233.
Su, C. W., Wang, X. Q., Tao, R., & Oana-Ramona, L. (2019). Do oil prices drive
agricultural commodity prices? Further evidence in a global bio-energy context.
Energy, 172, 691-701.
ot
Tang, K., & Xiong, W. (2012). Index investment and the financialization of
commodities. Financial Analysts Journal, 68(6), 54-74.
Thompson, S. (1986). Returns to storage in coffee and cocoa futures markets. The
tn
Journal of Futures Markets (1986-1998), 6(4), 541.

Wang, Y., Bouri, E., Fareed, Z., & Dai, Y. (2022). Geopolitical risk and the systemic
risk in the commodity markets under the war in Ukraine. Finance Research Letters,
103066.
rin
Wang, Y., Wei, Y., Wu, C., & Yin, L. (2018). Oil and the short-term predictability of
stock return volatility. Journal of Empirical Finance, 47, 90-104.
Zhang, D., & Broadstock, D. C. (2020). Global financial crisis and rising connectedness
in the international commodity markets. International Review of Financial Analysis,
ep
68, 101239.
Zhang, L., Bouri, E., & Chen, Y. (2023). Co-jump Dynamicity in the Cryptocurrency
Market: A Network Modelling Perspective. Finance Research Letters, 104372.
Pr
Appendix
ed
v iew
re
er
Figure A1. Plots of price levels of commodity futures
pe
ot
tn
rin
ep
Pr
Figure A2. Plots of return levels of commodity futures
Table A1. Summary statistics of commodity futures daily returns
ed
Commodity group Individual Mean Max Min SD Skewness Kurtosis P-value P-value
Commodity ADF test PP test
futures
Metals Silver 0.0272 9.3037 -11.6138 1.8174 -0.3393 9.1057 0.000 0.000
iew
Metals Palladium 0.0625 20.4748 -20.4808 2.1672 -0.0588 14.8509 0.000 0.000
Metals Platinum -0.0123 10.4413 -11.6485 1.5172 -0.159 8.1565 0.000 0.000
Metals Lead 0.0212 6.5147 -6.2536 1.4185 0.0599 4.4143 0.000 0.000
Metals Copper 0.0521 7.5223 -7.024 1.363 -0.0349 4.7969 0.000 0.000
Metals Aluminum 0.0352 6.2296 -7.1703 1.3007 0.1342 5.6492 0.000 0.000
Metals Zinc 0.0323 7.5386 -6.8353 1.5741 -0.0116 4.2203 0.000 0.000
v
Metals Tin 0.0504 10.1155 -9.1367 1.5147 -0.4716 10.1031 0.000 0.000
Metals Nickel 0.1067 68.5833 -40.7192 3.2972 8.8378 219.4544 0.000 0.000
re
Metals Gold 0.0244 5.9454 -4.9854 0.9199 -0.1437 7.7018 0.000 0.000
Agricultural London Cocoa 0.0227 9.7705 -7.5332 1.5965 0.4082 6.0984 0.000 0.000
Agricultural London Coffee 0.0373 6.1832 -6.3089 1.3586 0.1495 4.2728 0.000 0.000
Agricultural London Sugar 0.0373 6.1832
er -6.3089 1.3586 0.1495 4.2728 0.000 0.000
Agricultural Lumber 0.0122 30.6998 -33.4783 3.4119 -0.4876 21.8825 0.000 0.000
Agricultural Orange Juice 0.0573 9.3404 -10.7306 2.0922 -0.0529 4.393 0.000 0.000
Agricultural US Sugar 0.0207 11.4203 -7.5299 1.769 0.2249 4.9723 0.000 0.000
pe
Agricultural US C Coffee 0.0195 9.8938 -8.4546 1.9925 0.3749 4.5277 0.000 0.000
Agricultural US Cotton 0.0037 31.3801 -6.5544 1.8599 2.9609 50.903 0.000 0.000
Agricultural US Cocoa 0.0202 6.3963 -16.3551 1.6829 -0.4882 8.3081 0.000 0.000
Agricultural Live Cattle 0.0298 7.2551 -14.4808 1.3126 -1.514 20.2706 0.000 0.000
Agricultural Lean Hogs 0.0423 28.361 -20.9125 2.8799 0.3416 20.9311 0.000 0.000
ot
Energy RBOB Gasoline 1.6334 43.1469 -39.5783 6.6982 2.1116 13.7255 0.000 0.000
Energy WTI Crude Oil -0.0923 37.6623 -305.966 8.5686 -28.48 980.7913 0.000 0.000
Energy Brent Crude Oil 0.1217 21.0186 -24.4036 2.5308 -0.5685 19.327 0.000 0.000
tn
Energy London Gasoline 0.1179 15.026 -33.5304 2.5647 -1.169 25.2671 0.000 0.000
Energy Natural Gas 0.1303 21.8943 -16.5282 3.7012 0.2075 6.2777 0.000 0.000
Energy Heating Oil 0.1037 15.0127 -21.9277 2.4772 -0.5617 13.4246 0.000 0.000
Energy Carbon Emissions 0.2149 17.5129 -16.2511 2.9023 -0.1162 6.3623 0.000 0.000
rin
Grain and Cereal London Wheat 0.0379 17.0431 -10.3356 1.3145 1.606 27.2686 0.000 0.000
Grain and Cereal Oats 0.0683 29.5319 -29.0689 2.5613 -0.5838 29.1852 0.000 0.000
Grain and Cereal Corn 0.0424 6.4051 -17.3865 1.6325 -1.141 15.0812 0.000 0.000
Grain and Cereal Rice 0.0433 10.2987 -25.8962 1.619 -2.2785 46.384 0.000 0.000
Grain and Cereal Soybeans 0.0275 6.6372 -8.1981 1.2736 -0.3005 6.1666 0.000 0.000
ep
Grain and Cereal Soybean Oil 0.0612 8.642 -9.4293 1.6186 -0.1003 6.519 0.000 0.000
Grain and Cereal Soybean Meal 0.026 12.7829 -13.332 1.6108 -0.5796 14.7422 0.000 0.000
Grain and Cereal US Wheat 0.0451 21.7761 -10.6823 2.0289 0.9115 12.3574 0.000 0.000
Notes: This table reports summary statistics of daily returns of the 36 most common commodity futures over the
Pr
sample period July 5, 2016 – July 1, 2023. It includes, the mean of returns, standard deviation (SD), largest and
smallest return observations (denoted by max and min, respectively), skewness, and kurtosis. The table also reports
the p-value of the Augmented Dickey-Fuller (ADF) test and Phillips-Perron(PP) test.
ed
v iew
re
er
Figure A3: Eigenvector centrality index for each type of commodity and
overall commodities based on the time-varying MST.
pe
ot
tn
rin
ep
Pr
Figure A4: Closeness centrality index for each type of commodity and overall
commodities based on the time-varying MST.
ed
v iew
re
er
pe
ot
Figure A5: Normalized tree length index for each type of commodity and overall
commodities based on the time-varying MST.
tn
rin
ep
Pr

SSRN Id4613695

Uploaded by

Copyright:

Available Formats

SSRN Id4613695

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

SSRN Id4613695

Uploaded by

Copyright:

Available Formats

Paper prepared for International Review of Financial Analysis

*Corresponding author: Yan Chen

Russia-Ukraine war; volatility forecasting implications.

Keywords: Co-jump network; commodity futures; co-jump transmission; COVID-19

Our study provides answers to these important research questions in commodity

network models. We examine the co-jump dependency in commodity futures returns

commodity futures returns from a network visualization perspective. Traditional co-

commodity group. The impact of co-jump centrality is time-varying, reflecting its

potential factors such as hedging pressure, inflation uncertainty, and macroeconomic

analysis, and Section 7 concludes.

4.1. Detecting jump and pairwise co-jump scores

To identify commodity futures price jumps, the non-parametric threshold method

where {1} is the indicator function.

helps us to identify which commodity futures tend more to experience a co-jumping

occurrence of co-jumps may be exploited by market participants to generate abnormal

better diversified and resilient to co-jump events.

Table 2. Statistics of jumps detected

Metals Palladium 424 24.84% Agricultural Live Cattle 154 9.02%

Metals Gold 70 4.1% Energy Carbon Emission 718 42.06%

5.2. Co-jump network analysis I: total 36 commodity futures

Figure 1. Co-jump network based total 36 commodity futures

Notes: See notes to Figure 1.

5.3. Co-jump network analysis II: Inter-industry group

Figure 3. Four co-jump networks based Inter-industry group

Figure 4. Four co-jump networks based on MST

5.4. Dynamics ranking of co-jump influence: total 36 commodity futures

Figure 5. Dynamic ranking of co-jump influence for the 36 commodity futures

5.5. Dynamics ranking of co-jump influence: Inter-industry group

6. Application to portfolio allocation and volatility forecasting

6.1.1 Mean-variance portfolio construction

𝜎 𝑀𝑉 = 𝑠𝑡𝑑(𝑟𝑡𝑀𝑉 × √252) is the annualized volatility.

commodity futures achieve improved risk diversification, thereby reducing portfolio

jump dependency among commodity futures could be overestimated, leading to

Table 3. Annualized Sharpe ratio of mean-variance portfolios

Figure 7. Cumulative returns of portfolios

Closeness centrality: 𝑪𝑪(𝒊) = ∑(𝒊,𝒋) 𝑹𝒊𝒋 , 𝒊 ≠ 𝒋 （15）

The t-statistic is obtained by regressing {𝑓𝑡 : 𝑡 = 𝑀 + 1, … , 𝑇} on a constant.

6.2.2 Dynamic analysis and forecasting results

based on the Recursive window approach is poor, as it is consistently outperformed by

Table 4. Out-of-sample forecasting results

Rolling window ∆𝑹𝟐𝑶𝑶𝑺 ∆𝑹𝟐𝑶𝑶𝑺 ∆𝑹𝟐𝑶𝑶𝑺

Energy Group 1.9181*** 2.0624*** -0.5459

Metals Group -0.4859 -0.12808 -0.1159

networks analysis based Inter-industry group, it appears that co-jump heterogeneity

asset allocation in international equity markets. Journal of Economic Dynamics and

commodities. Resource and energy economics, 35(4), 599-617.

stochastic volatility and jumps. Journal of financial econometrics, 2(1), 1-37.

markets. The journal of finance, 46(3), 1009-1044.

Finance, 66(6), 2165-2211.

in the cross‐section of stock returns. The Journal of Finance, 70(2), 577-614.

markets: Implications for portfolio management. Resources Policy, 73, 102172.

Nguyen, D. B. B., & Prokopczuk, M. (2019). Jumps in commodity markets. Journal of

Journal of Futures Markets (1986-1998), 6(4), 541.

Figure A2. Plots of return levels of commodity futures

You might also like

Energy Group 1.9181* 2.0624* -0.5459