Metal prices made in China? A network analysis of industrial metal futures

Siklos, Pierre L.; Stefan, Martin; Wellenreuther, Claudia Article — Published Version Metal prices made in China? A network analysis of industrial metal futures Journal of Futures Markets Provided in Cooperation with: John Wiley & Sons Suggested Citation: Siklos, Pierre L.; Stefan, Martin; Wellenreuther, Claudia (2020) : Metal prices made in China? A network analysis of industrial metal futures, Journal of Futures Markets, ISSN 1096-9934, Wiley, Hoboken, NJ, Vol. 40, Iss. 9, pp. 1354-1374, https://doi.org/10.1002/fut.22125 This Version is available at: https://hdl.handle.net/10419/230161 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by/4.0/ J Futures Markets. 2020;40:1354–1374.1354 | wileyonlinelibrary.com/journal/fut Received: 16 November 2019 | Accepted: 9 April 2020 DOI: 10.1002/fut.22125 RESEARCH ARTICLE Metal prices made in China? A network analysis of industrial metal futures Pierre L. Siklos 1 |Martin Stefan 2 |Claudia Wellenreuther 2,3 1 Department of Economics, Wilfrid Laurier University, Waterloo, Ontario, Canada 2 Department of Economics, Westfälische Wilhelms‐Universität Münster, Münster, Germany 3 Department of Energy and Environmental Economics, Hamburg Institute of International Economics (HWWI), Hamburg, Germany Correspondence Martin Stefan, Department of Economics, Westfälische Wilhelms‐Universität Münster, Am Stadtgraben 9, 48143 Münster, Germany. Email: [email protected]e Abstract In addition to being the world's greatest consumer and producer of industrial metals, China now also features the most actively traded industrial metal futures contracts worldwide. To examine China's role in the global price formation process of industrial metal futures markets, we use a sample of 29 futures contracts traded on exchanges in the United States, the United Kingdom, India, and China. We estimate vector autoregressive models and conduct variance decompositions, which are then visualized in the form of networks. The results indicate that China, despite its role as key actor in both real and financial industrial metal markets, is a price taker. KEYWORDS China, commodity markets, industrial metals, networks, price leadership 1|INTRODUCTION China's rapid industrialization and rise as an economic power have been accompanied by a voracious appetite for natural resources. This is particularly visible in the country's demand for industrial metals. As the country continues its process of urbanization and investment in infrastructure, China has evolved into the world's top consumer of refined aluminum, copper, nickel, steel, and zinc. In 1980, when China's policy of reform was just beginning, the country's share in worldwide consumption of these metals ranged between 3% and 4%. Today, Chinese consumption makes up 40% of the world's demand for lead and nickel and 50% of the world's demand for aluminum, copper, and zinc (World Bank, 2018). Similarly, the country has also developed into the top producer of these metals. In 2017, roughly half of all steel, refined aluminum and zinc, and 40% of refined copper and lead were produced in China (World Bank, 2018). Moreover, China's importance in the market for metals is not limited to the real side of the economy. Chinese commodity futures exchanges have, over the years and following a series of regulatory changes, evolved into the world's largest futures markets for numerous industrial metals. As documented by the Futures Industry Association's (FIA) 2018 volume survey, seven out of the ten most traded industrial metal futures contracts are traded on Chinese exchanges (Acworth, 2019). Moreover, the Shanghai Futures Exchange's (SHFE) steel rebar futures contract has grown into the most traded commodity futures contract worldwide. In light of these developments, this paper investigates the role of Chinese price leadership in industrial metal futures markets. We gather futures price data on 29 industrial metal contracts traded on six exchanges in the United States, the United Kingdom, India, and China. Our study is the most comprehensive analysis of industrial metal futures conducted to date. It includes futures contracts for copper, lead, nickel, iron, and several kinds of steel, and thus covers a much ------------------------------------------------------------------------------------------------------------------------------------------- This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2020 The Authors. The Journal of Futures Markets published by Wiley Periodicals LLC larger variety of commodities than earlier studies. To answer the question of whether China has become a price leader in these markets, the network approach of Diebold and Yilmaz (2012,2014) is used, which rests on variance decompositions of vector autoregressive (VAR) models' forecast errors. Based on these decompositions, so‐called connectedness tables are compiled which summarize how shocks to a specific futures price travel through the system of all prices of this commodity. These connectedness tables are then visualized in the form of graphical networks. Finally, we explore, via regressions, whether we can identify some economic and financial determinants of connectedness. We obtain two main findings. Chinese metal contracts are strongly interconnected with contracts traded on other exchanges. Thus, there are significant information flows between Chinese and selected Western exchanges. However, in these relationships, China is typically a net receiver of price shocks and not a large sender. Instead, price discovery is roughly equally shared between the U.S.‐American, British, and Indian futures markets. This implies that China, despite its role as leading consumer, producer, and trader of industrial metals, is a price taker in the corresponding futures markets. Consequently, other factors hinder Chinese exchanges from playing a more important role in the global price formation process. A key factor in this regard might be the investor structure of Chinese commodity futures exchanges, which differs from its Western counterparts. First, government regulation largely excludes foreign investors from participating in Chinese markets. Second, various news accounts document that Chinese metal futures are traded by a large number of uninformed retail investors (see e.g., Financial Times, 2016a,2016b). The remainder of this paper is structured as follows. Section 2summarizes important steps in the development of Chinese commodity futures markets and earlier research on their role as price leader. Thereafter, Section 3explains the data used in this paper, while Section 4introduces the methodology. Section 5then presents the key results, and Section 6analyzes the determinants of connectedness. Section 7concludes. 2|INSTITUTIONAL BACKGROUND AND RELATED LITERATURE Today, Chinese futures trading occurs on five futures exchanges. The Dalian Commodity Exchange (DCE) and the Zhengzhou Commodity Exchange (ZCE) mainly focus on agricultural and chemical products, while the SHFE covers various metal contracts. Financial and crude oil futures contracts are traded at the China Financial Futures Exchange (CFX) and the Shanghai International Energy Exchange (INE). Despite the fact that Chinese futures markets continue to be relatively closed to foreign investors, who must generally rely on domestic intermediaries to conduct trades on their behalf, 1 many Chinese futures contracts now outstrip their Western counterparts in terms of trading volume. In international comparisons, Chinese metal futures contracts trade in remarkably high trading volumes, which is not surprising given the stylized facts outlined in the introduction. According to the FIA 2018 volume survey, seven out of the ten most traded industrial metal futures contracts are all traded on Chinese exchanges (Acworth, 2019). Moreover, the SHFE steel rebar futures contract has grown into the most traded commodity futures contract in the world. Against this backdrop, and given the fact that China has, over the years, evolved into the largest consumer and producer of numerous industrial metals (World Bank, 2018), a sizable body of literature has investigated the role of Chinese futures exchanges as potential price leaders in industrial metals. Being among the oldest industrial metal futures contracts traded in China, the SHFE's copper contract has been analyzed in numerous studies. Fung, Leung, and Xu (2003) use data from 1995 to 2001 and conduct a bivariate GARCH analysis of two copper futures contracts traded at the SHFE and the New York Commodity Exchange (COMEX). They find that the American contract dominates the information flow between the two markets. Liu and An (2011) study the same contracts but make use of a VECM‐GARCH framework and a price discovery metric developed by Lien and Shrestha (2009). The authors use data from 2004 to 2009 and conclude that the U.S. market generally leads its Chinese counterpart and is also dominant in the price discovery process. Li and Zhang (2013) study the case of copper on a broader basis by considering, in addition to the contracts traded at the SHFE and the COMEX, the contracts traded at the London Metal Exchange (LME) and the Multi Commodity Exchange (MCX) in Mumbai, India. Employing an SVAR model and data ranging from 2005 to 2011, the authors' results suggest that the LME contract is the key price maker. Conversely, the results of Rutledge, Karim, 1 Exceptions pertain to the INE's crude oil contract, the DCE's iron ore contract and the ZCE's Purified Terephthalic Acid (PTA) contract, which, following a new directive by the China Securities Regulatory Commission (2015), can be traded directly by qualified foreign brokerage firms without the need for a domestic intermediary. SIKLOS ET AL. | 1355 and Wang (2013), who use a VECM estimation and Granger‐causality tests based on data from 2006 to 2011, reveal no distinct leadership pattern between the copper contracts traded at the SHFE, LME, and COMEX. Studies featuring multiple metal contracts include that of Hua and Chen (2007), who study the markets for copper and aluminum. Using data from 1998 to 2002, the authors consider contracts traded at the SHFE and the LME. They employ Granger‐causality tests and find that the LME contracts Granger‐cause those of the SHFE, which implies that the Chinese contracts are price followers. Fung, Liu, and Tse (2010), who consider aluminum and copper contracts traded at the SHFE and the COMEX, use data from 1999 to 2009 and employ a VECM which accounts for structural breaks. They find that neither market dominates the information flow between them. Fung, Tse, Yau, and Zhao (2013) investigate the case of Chinese price leadership on an even broader basis by considering 16 different commodities including aluminum, copper, and zinc contracts traded at the SHFE and the LME. The authors' data range from different starting dates for each contract until 2011 and are analyzed using various regression techniques including error correction and GARCH models. Again, no clear pattern is found, as mixed and bidirectional results are obtained for the different industrial metal contracts. Lastly, the study by Kang and Yoon (2016), which is closely related to our work, uses the approach proposed by Diebold and Yilmaz (2012) to study the SHFE's and LME's futures contracts for aluminum, copper, and zinc. Using data from 2007 to 2016, the authors find that price shocks typically originate in the LME's contracts and then travel to those of the SHFE. 2 The present paper extends this earlier research in three important ways: First, by analyzing 29 different contracts, some of which were launched as recently as November 2015, it is the most comprehensive study of industrial metal futures conducted to date. As our analysis covers futures contracts for copper, lead, nickel, iron, and several kinds of steel, we investigate a much larger variety of commodities than earlier studies. Second, by conducting variance decompositions of the forecast errors of various systems of different futures contracts, the network approach of Diebold and Yilmaz (2012,2014)enablesusto graphically visualize the inter‐dependencies between the different contracts. Third, we go beyond reporting market connectedness to consider the potential economic determinants of this phenomenon. 3|DATA To examine the role of Chinese price leadership in the market for industrial metal futures, we gather data on 29 industrial metal contracts at daily frequency beginning in November 2004 until August 2019. All price time series are retrieved from Thomson Reuters Datastream, whereby continuous series are constructed by switching to the nearest contract on the first day of each new trading month. The sample ranges for the individual commodity groups are dictated by the availability of data for the youngest futures contract in that group. Table 1lists the contracts used in the analysis and details the different contract specifications such as notation and size. Our sample covers five aluminum contracts, one cobalt contract, four copper contracts, one ferrosilicon contract, two iron ore contracts, three lead and nickel contracts, one silicon manganese contract, five steel contracts, one tin contract and three zinc contracts. The contracts are traded on six different exchanges, namely the COMEX, the LME, the MCX (in Mumbai, India), the SHFE, the DCE, and the ZCE. 3 Figure 1displays the futures price time series of the commodity contracts included in our analysis. All prices have been converted to USD per metric ton (USD/mt). In accordance with the law of one price, we observe relatively similar price movements among the different contracts for each type of commodity. Nonetheless, Chinese prices are most of the time noticeably higher for all commodities. This could be due to barriers to trade concerning the Chinese market. Regarding the steel market, we observe the greatest price differences within one commodity. This is because of the different types of steel included in our sample, which range from steel rebar to steel coils and steel scrap. The same holds for the aluminum market, where one can see large price differences between the LME's aluminum alloy contract and the other pure aluminum contracts. The unusual behaviour of the COMEX's aluminum price starting in late 2017 can be explained by the exceptionally low trading volumes of this contract. Summary statistics of the daily logarithmic futures returns are displayed in Table 2. The daily returns range from −0.32% to 0.33%. Standard deviations range from 0.01 to 0.02. Roughly two‐thirds of all return series exhibit a negative skewness, suggesting that severe price drops are more common than large price increases. For all return series we 2 Other applications of variance decompositions or the network approach of Diebold and Yilmaz (2012) to the case of Chinese commodity markets, include the studies of Yang and Leatham (1999) and Zhang and Wang (2014) who consider the markets for wheat and crude oil. 3 Note that the COMEX is since 2008 owned by the CME Group, while the LME is since 2012 owned by Hong Kong Exchanges and Clearing. 1356 | SIKLOS ET AL. observe kurtosis values well in excess of 3, which is the reference kurtosis value of the normal distribution. This implies that none of the series follow a normal distribution but feature fat tails instead. 4|METHODOLOGY To investigate the price leadership in the metal futures market, we follow the financial market connectedness approach of Diebold and Yilmaz (2012,2014). Within this framework, the informational spillovers between different metal TABLE 1 Industrial metal futures contracts Contract Exchange Notation Size Aluminum COMEX USD/mt 25 mt Aluminum LME USD/mt 25 mt Aluminum Alloy LME USD/mt 20 mt Aluminum MCX INR/kg 5 mt Aluminum SHFE RMB/mt 5 mt Cobalt LME USD/mt 1 mt Copper COMEX USD/lbs 25,000 lb Copper LME USD/mt 25 mt Copper MCX INR/kg 1 mt Copper SHFE RMB/mt 5 mt Ferrosilicon ZCE RMB/mt 5 mt Iron Ore DCE RMB/mt 100 mt Iron Ore COMEX USD/mt 500 mt Lead LME USD/mt 25 mt Lead MCX INR/kg 5 mt Lead SHFE RMB/mt 5 mt Nickel LME USD/mt 6 mt Nickel MCX INR/kg 250 kg Nickel SHFE RMB/mt 1 mt Silicon Manganese ZCE RMB/mt 5 mt Steel Scrap LME USD/mt 10 mt Steel Rebar LME USD/mt 10 mt Steel Coils COMEX USD/st 20 st Steel Rebar SHFE RMB/mt 10 mt Steel Coils SHFE RMB/mt 10 mt Tin LME USD/mt 5 mt Zinc LME USD/mt 25 mt Zinc MCX INR/kg 5 mt Zinc SHFE RMB/mt 5 mt Note: Contract sizes are reported in “mt,”“kg,”“st,”and “lb”referring to metric tons, kilograms, short tons (equivalent to roughly 0.907 metric tons), and pounds (equivalent to 0.453 kilograms), respectively. Abbreviations: COMEX, New York Commodity Exchange; DCE, Dalian Commodity Exchange; INR, Indian rupee; LME, London Metal Exchange; MCX, Multi Commodity Exchange (Mumbai, India); RMB, Chinese renminbi; SHFE, Shanghai Futures Exchange; USD, U.S. dollar; ZCE, Zhengzhou Commodity Exchange. SIKLOS ET AL. | 1357 (a) (b) (c) (d) (e) (f) (g) (h) (i) (j) FIGURE 1 Futures price time series. COMEX contracts are highlighted in blue, LME contracts in red, MCX contracts in green, SHFE contracts in orange, DCE contracts in cyan, ZCE contracts in purple. Steel rebar, aluminum alloy, and the silicon manganese contracts are depicted using dashed lines. All prices have been converted to USD/mt. Subfigures (a) through (j) show the price time series for aluminum, copper, lead, nickel, steel, iron, zinc, silicons, cobalt and tin, respectively [Color figure can be viewed at wileyonlinelibrary.com] 1358 | SIKLOS ET AL. futures contracts are studied using a network interpretation of a VAR model's variance decomposition. The starting point of this approach is estimating the following covariance stationary VAR( p ) model: ∑ rrεΦ=+ , t i p iti t =1 −(1) where the vector rrr r=( , ,…, ) ′ ttt nt1, 2, , contains nlogarithmic futures return time series, and εt is an n× 1 vector of white noise disturbances with covariance matrix Ω . 4 TABLE 2 Summary statistics of returns Contract Exchange Obs. Min Mean Max SD Skew. Kurt. Aluminum COMEX 1,385 −0.20 −0.00 0.04 0.01 −6.22 130.07 Aluminum LME 1,386 −0.08 −0.00 0.05 0.01 0.18 5.91 Aluminum Alloy LME 1,386 −0.07 −0.00 0.06 0.01 −0.24 9.39 Aluminum MCX 1,386 −0.10 0.00 0.08 0.01 0.39 11.13 Aluminum SHFE 1,386 −0.04 −0.00 0.04 0.01 0.19 5.86 Cobalt LME 982 −0.16 0.00 0.12 0.02 −0.58 18.76 Copper COMEX 3,854 −0.12 0.00 0.12 0.02 −0.13 7.19 Copper LME 3,854 −0.10 0.00 0.12 0.02 −0.01 7.51 Copper MCX 3,853 −0.12 0.00 0.10 0.02 −0.10 7.68 Copper SHFE 3,854 −0.07 0.00 0.06 0.01 −0.28 6.27 Ferrosilicon ZCE 982 −0.28 0.00 0.20 0.02 −2.70 55.71 Iron Ore DCE 1,527 −0.32 −0.00 0.10 0.02 −2.91 32.73 Iron Ore COMEX 1,528 −0.09 −0.00 0.16 0.02 0.15 6.93 Lead LME 2,199 −0.08 −0.00 0.08 0.02 −0.01 5.17 Lead MCX 2,199 −0.09 −0.00 0.09 0.01 0.11 6.38 Lead SHFE 2,198 −0.05 −0.00 0.05 0.01 −0.17 7.59 Nickel LME 1,153 −0.09 0.00 0.07 0.02 −0.18 4.64 Nickel MCX 1,153 −0.08 0.00 0.07 0.02 −0.01 4.52 Nickel SHFE 1,152 −0.06 0.00 0.06 0.01 −0.08 5.29 Silicon Manganese ZCE 982 −0.28 0.00 0.33 0.02 0.42 68.77 Steel Scrap LME 981 −0.08 0.00 0.10 0.02 −0.22 7.06 Steel Rebar LME 981 −0.05 0.00 0.05 0.01 0.05 6.01 Steel Coils COMEX 982 −0.06 0.00 0.11 0.01 1.65 18.87 Steel Rebar SHFE 982 −0.09 0.00 0.10 0.02 −0.18 7.80 Steel Coils SHFE 982 −0.08 0.00 0.08 0.02 0.10 7.59 Tin LME 982 −0.07 0.00 0.04 0.01 −0.35 6.09 Zinc LME 3,242 −0.11 −0.00 0.10 0.02 −0.05 5.43 Zinc MCX 3,242 −0.09 −0.00 0.10 0.02 −0.07 5.68 Zinc SHFE 3,242 −0.06 −0.00 0.05 0.01 −0.39 5.71 Abbreviations: COMEX, New York Commodity Exchange; DCE, Dalian Commodity Exchange; LME, London Metal Exchange; MCX, Multi Commodity Exchange (Mumbai, India); SHFE, Shanghai Futures Exchange; ZCE, Zhengzhou Commodity Exchange. 4 As our empirical application includes 29 futures return time series, we use an elastic net shrinkage approach based on Zou and Hastie (2005)to eliminate statistically superfluous variables and improve the performance of the VAR. SIKLOS ET AL. | 1359 4.1 |Forecast error variance decomposition The VAR model above can be represented as a vector moving average (VMA) model of the form ∑ ∞ rεΨ= , titi i − =0 (2) where Ψ idenotes the nn×moving average coefficient matrices. These are determined by Ψ ΦΨ= + ii1−1 ⋯ Φ ΨΦΨ++ ipi p 2−2−for i >0 , while I Ψ =n0and Ψ 0= iif i <0 . Based on the VMA model in Equation (2), one can compute the generalized H ‐step ahead forecast error variance decompositions d ij Hof Koop, Pesaran, and Potter (1996) and Pesaran and Shin (1998)as ∑ ∑ ee ee d σΨΩ ΨΩΦ =() () , ′ ′′ ihj ihhi ij Hjj h H h H −1 =0 −12 =0 −1 (3) where σ jj is the standard deviation of ε j t , , while e i is the n× 1 selection vector consisting of zeros only except for its i ‐th element, which is equal to one. This decomposition captures the contribution that shocks to variable j make to the H ‐ step‐ahead error variance when forecasting variable i . In the case of i j=, Diebold and Yilmaz (2012) refer to d ij Has the own variance share. Correspondingly, if ≠ i j, d ij His called the cross variance share. Note that this type of variance decomposition, unlike conventional variance decompositions, does not make use of a Cholesky factorization of Ω and is thus independent of the ordering of the time series in the system. However, as the shocks to the model's variables are not orthogonalized, a variable i 's different variance shares due to shocks in variable j generally do not add up to one, that is ∑ ≠d 1 j n ij H =1 . Therefore, to allow straightforward comparisons between the different shocks sent by a variable j to another variable i , the variance decompositions are normalized and converted into percentages by computing ∑⋅ d d d ˜= 100 . ij Hij H j n ij H =1 (4) Thus, by construction, ∑ d ˜=10 0 j nij =1 and ∑ ⋅dn ˜= 100 ij nij ,=1 . 5 4.2 |Measuring connectedness Following Diebold and Yilmaz (2014), the variance decomposition computed above can be interpreted as a measure of the H ‐step ahead gross pairwise directional connectedness from variable j to variable i , that is ← Cd=˜ . ij Hij H (5) As ← C ij H will generally not be equal to ← Cji H, the net flow of shocks between the two variables, or net pairwise directional connectedness from variable j to variable i , is calculated as Cdd=˜−˜ . ij Hji H ij H (6) To gauge a variable's relative importance as sender or receiver of shocks in the system, Diebold and Yilmaz (2014) compute two measures of total directional connectedness. The first of these measures, ← CiH•, summarizes all those parts of a variable i 's forecast error variance decomposition that are due to shocks from another variable j . Hence, this measure is calculated as ∑ ← ≠ Cd=˜. iH j ji n ij H • =1, (7) 5 An alternative forecast error variance decomposition is developed by Lanne and Nyberg (2016). 1360 | SIKLOS ET AL. Conversely, the second measure ← C j H • summarizes all the contributions that variable i makes to the forecast error variance of another variable j . This measure is therefore given by ∑ ← ≠ Cd=˜ . j H i ij n ij H • =1, (8) The difference between these two metrics is the net total directional spillover ↔←← CCC=− . iHiHi H ••• (9) If a market's net spillover is above zero, the market sends more shocks than it receives. Conversely, if the market's net spillover is below zero, the market is a net receiver of price signals. Lastly, to capture the system's total connectedness, Diebold and Yilmaz (2014) sum up all of the normalized cross variance shares. To allow for comparing the total connectedness values of different variable systems, this measure C H is also normalized by n: ∑ ≠ Cnd=1˜ . H ij ij n ij H ,=1, (10) It holds by construction that the system's total connectedness is equal to the (normalized) sum of all shocks sent or equivalently all shocks received. Given these measures of connectedness, a connectedness table for the VAR system of Equation (1) is constructed as follows: Table 3. The main diagonal elements, apart from C H , display how the variance of a specific return series is driven by the series's own shocks. The off‐diagonal elements, except those at the margin of the connectedness table, represent the fraction of a return series's variance that is due to shocks in the other return series. The bottom row elements summarize the total impact that the return series have on the variance of the other return series, while the elements of the right‐most column summarize the total of shocks that the return series receive from the other series in the system. Thus, the greater a futures return series' total directional connectedness, the greater its role in price leadership. Conversely, if a return series features a large row sum, it features a high total directional connectedness from others and is therefore a strong recipient of price signals originating from other futures contracts. Our analysis covers 29 different futures price time series, rendering an analysis of all impulse response functions unfeasible. However, as shown by Diebold and Yilmaz (2014), the variance decompositions matrix described above can be interpreted as a network. The nodes of this network are the different variables of the VAR system, that is, in our case the different metal futures contracts, while the connections between the contracts are determined by the magnitudes of the different variance decompositions. The advantage of interpreting the variance decompositions in this way is that it allows for straightforward visualizations of market interdependencies and information flows using previously developed graph‐drawing algorithms, which are discussed below in greater detail. TABLE 3 Concept of connectedness tables r 1 r 2  r n From others r 1 ← C H 1 1 ← CH 1 2 ⋯ ← C n H 1 ← C H 1• r 2 ← CH 2 1 ← CH 2 2 ⋯ ← C n H 2← C H 2 • ⋮⋮⋮ ⋱ ⋮⋮ r n ← Cn H 1 ← Cn H2 ⋯ ← Cnn H← C n H• To others ← CH • 1 ← C H •2 ⋯ ← CH • 1 CH Note: Connectedness table as proposed by Diebold and Yilmaz (2014). SIKLOS ET AL. | 1361 7|CONCLUSION Over the past two decades, China has become the greatest consumer and producer of numerous industrial metals. Moreover, China has recently launched a number of futures contracts for these metals, and these have become some of the most highly traded futures contracts worldwide. This paper investigates the question of whether these new markets are important in the formation of international metal prices. We follow the network approach by Diebold and Yilmaz (2012,2014)andconsider 29 metal contracts, traded on six exchanges in the United States, the United Kingdom, India, and China. Despite their large trading volumes, our results indicate that the Chinese futures contracts are not price leaders. Our analysis comprised three steps. First, we analyzed the overall network structure across all industrial metal futures contracts included in our sample. Unsurprisingly, futures contracts of the same underlying commodities were grouped closely together. Of these clusters, the copper and zinc clusters were found to be the most important ones regarding the transmission of price signals. Furthermore, the Chinese contracts appeared to play a minor role within the different commodity clusters. In a second step, we repeated the earlier analysis, but for each of the different commodity clusters separately. The results of this step confirm those of the first one: Chinese contracts were again found to be net recipients of price shocks. Next, we conducted time‐varying network analyses to study how China's role of price leadership varies over time. The results suggest that China's passive role in the price discovery process is relatively stable over time. Lastly, we used a dynamic fixed effects panel regression to study the determinants of connectedness. Apart from lagged spillovers, relative volatility and real sector flows are the strongest determinants of connectedness. TABLE 4 Regression results Aluminum Copper Lead Iron & steel Zinc ← Cijt,− 1 10 1.007*** 0.952*** 0.919*** 0.927*** 0.931*** (0.013) (0.013) (0.016) (0.011) (0.016) V OLAjt,0.007 −0.032*** −0.023*** 0.013*−0.027 (0.009) (0.007) (0.005) (0.007) (0.015) ILLIQjt,0.006 −1.033 −1.034*** 0.000 −9.420 (0.006) (0.881) (0.160) (0.000) (8.013) S PMATjt,−0.000 −0.000 −0.000 0.001** −0.000 (0.000) (0.000) (0.000) (0.000) (0.000) IMij t,−0.008 0.000 0.012 0.016** 0.013*** (0.007) (0.001) (0.022) (0.008) (0.002) E Xij t ,−0.009 0.000 0.104 0.004*** 0.014*** (0.011) (0.001) (0.188) (0.001) (0.002) TEDj t ,0.152 0.038 −0.097 0.270 −0.084 (0.177) (0.024) (0.076) (0.171) (0.046) V IXj t ,−0.008 −0.006 −0.041** −0.000 0.018** (0.014) (0.005) (0.011) (0.009) (0.006) E PUjt,−0.000 0.001** 0.000 0.001** 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) Const. 0.804 1.956*** 3.778*** −1.214** 2.713** (0.763) (0.544) (0.517) (0.504) (1.049) R ¯20.943 0.920 0.876 0.873 0.905 Note: The table displays the results of the two‐stage panel fixed effects regressions for the different commodity subnetworks. The aluminum sample starts on May 6, 2014, the copper sample on November 18, 2004, the lead sample on March 24, 2011, the nickel sample on March 27, 2015, the iron and steel sample on November 23, 2015, and the zinc sample on March 26, 2007. All samples end on August 27, 2019. Standard errors are displayed in parentheses. *p< .1. **p< .05. ***p< .01. 1368 | SIKLOS ET AL. In conclusion, our results provide strong evidence that metal prices are currently not made in China. Price leadership, however, does not appear to be limited to Western markets, as India also appears to be an important transmitter of price signals. Over the past years, Chinese regulators have been able to develop active futures markets for many different commodities including various industrial metals. However, further steps have to be taken to strengthen the role of Chinese markets in terms of price leadership. Most importantly, Chinese markets must become more accessible to foreign investors. But unless such measures are extended to additional markets, price differentials between Chinese and Western markets will continue to exist, since no arbitrage trading will be possible between these two trading venues. A first step in this direction might be the opening of the DCE iron ore futures contract to overseas investors in May 2018. This will allow foreign investors to provide additional liquidity and exploit price differentials between Western and Chinese futures markets. Moreover, Chinese regulators should aim for greater participation of institutional investors to dampen the effects of retail investors and noise traders. This will improve the ability of Chinese markets to more accurately pick up new fundamental information and become an important price maker in industrial metals. Finally, it is also likely that, as markets in China mature while opening up to foreign investors, the performance of Chinese markets will change. Together with the impending slowdown of China's economy, there is greater scope for a shift away from behavior associated with price taking. Future research, using event studies or structural break tests, may reveal whether this policy change alters the importance of this contract in the global price formation process of iron ore. 11 Moreover, if Chinese market regulators become more transparent and provide greater information about the investor structure in Chinese commodity futures markets, future work will also be able to examine the role of Chinese retail investors and their impact on Chinese price leadership. Such an extension will also provide useful insights to policy makers and regulators. ACKNOWLEDGMENTS We thank the editor Robert Webb and an anonymous referee for helpful comments. We would also like to thank our colleague Christoph Sulewski for his valuable help concerning the network analysis. DATA AVAILABILITY STATEMENT The futures price data used in this study are available from Thomson Reuters Datastream (2019). Restrictions apply to the availability of these data, which were used under license for this study. Import and export data used in this study are available from the International Trade Center (2019;https://marketanalysis.intracen.org/en/home). Economic policy uncertainty (EPU) data were taken from Baker et al. (2016;https://www.policyuncertainty.com/). ORCID Martin Stefan http://orcid.org/0000-0002-0106-436X Claudia Wellenreuther http://orcid.org/0000-0001-7329-8221 REFERENCES Acworth, W. (2019). 2018 annual volume survey. Market Voice,5(1). Retrieved from https://www.fia.org/articles/fia‐releases‐half‐year‐data‐ futures‐and‐options‐volume‐trends Amihud, Y. (2002). Illiquidity and stock returns: Cross‐section and time‐series effects. Journal of Financial Markets,5(1), 31–56. Baker, S. R., Bloom, N., & Davis, S. J. (2016). Measuring economic policy uncertainty. The Quarterly Journal of Economics,131(4), 1593–1636. Retrieved from https://www.policyuncertainty.com/ China Securities Regulatory Commission (2015). Interim measures for overseas traders and overseas brokers to manage futures of certain types of futures. Retrieved from http://www.dce.com.cn/dalianshangpin/fg/fz/jysgzhgz/2048658/index.html Diebold, F. X., & Yilmaz, K. (2012). Better to give than to receive: Predictive directional measurement of volatility spillovers. International Journal of Forecasting,28(1), 57–66. Diebold, F. X., & Yilmaz, K. (2014). On the network topology of variance decompositions: Measuring the connectedness of financial firms. Journal of Econometrics,182(1), 119–134. 11 As previously discussed, our results reveal that both Western and Indian futures exchanges are of greater importance than their Chinese counterparts. Historically, India's commodity futures markets have also been relatively closed to foreign investors. Nevertheless, Chinese price leadership might be impaired by additional factors that do not impede Indian markets. These obstacles include, for example, less weight on the rule of law in China than in India (as measured, e.g., by the World Bank's 2020 governance indicators) and China's relatively heavier state‐control of its economy. However, the approach adopted in this paper is unable to identify the relative importance of these institutional factors. SIKLOS ET AL. | 1369 Financial Times (2016a). China warns off commodities market punters. Retrieved from https://www.ft.com/content/5094d2e8‐1387‐11e6‐ 91da‐096d89bd2173 Financial Times (2016b). Chinese retail investors throw global commodities into a tailspin. Retrieved from https://www.ft.com/content/ be92835e‐0dcd‐11e6‐b41f‐0beb7e589515 Fruchterman, T. M. J., & Reingold, E. M. (1991). Graph drawing by force‐directed placement. Software: Practice and Experience,21(11), 1129–1164. Fung, H.‐G., Leung, W. K., & Xu, X. E. (2003). Information flows between the U.S. and China commodity futures trading. Review of Quantitative Finance and Accounting,21(3), 267–285. Fung, H.‐G., Liu, Q. W., & Tse, Y. (2010). The information flow and market efficiency between the U.S. and Chinese aluminum and copper futures markets. Journal of Futures Markets,30(12), 1192–1209. Fung, H.‐G., Tse, Y., Yau, J., & Zhao, L. (2013). A leader of the world commodity futures markets in the making? The case of China's commodity futures. International Review of Financial Analysis,27, 103–114. Guhathakurta, K., Dash, S. R., & Maitra, D. (2020). Period specific volatility spillover based connectedness between oil and other commodity prices and their portfolio implications. Energy Economics,85, 104566. Hua, R., & Chen, B. (2007). International linkages of the Chinese futures markets. Applied Financial Economics,17(16), 1275–1287. Indriawan, I., Liu, Q., & Tse, Y. (2019). Market quality and the connectedness of steel rebar and other industrial metal futures in china. Journal of Futures Markets,39(11), 1383–1393. International Trade Center (ITC) (2019). Market analysis tools portal. Retrieved from https://marketanalysis.intracen.org/en/home Kang, S. H., & Yoon, S.‐M. (2016). Dynamic spillovers between Shanghai and London nonferrous metal futures markets. Finance Research Letters,19, 181–188. Koop, G., Pesaran, M., & Potter, S. M. (1996). Impulse response analysis in nonlinear multivariate models. Journal of Econometrics,74(1), 119–147. Lanne, M., & Nyberg, H. (2016). Generalized forecast error variance decomposition for linear and nonlinear multivariate models. Oxford Bulletin of Economics and Statistics,78(4), 595–603. Li, Z., & Zhang, L. H. (2013). An empirical study of international linkages of the Shanghai copper futures market. The Chinese Economy, 46(3), 61–74. Lien, D., & Shrestha, K. (2009). A new information share measure. Journal of Futures Markets,29(4), 377–395. Liu, Q., & An, Y. (2011). Information transmission in informationally linked markets: Evidence from US and Chinese commodity futures markets. Journal of International Money and Finance,30(5), 778–795. Liu, Q., Tse, Y., & Zhang, L. (2018). Including commodity futures in asset allocation in China. Quantitative Finance,18(9), 1487–1499. Pesaran, H. H., & Shin, Y. (1998). Generalized impulse response analysis in linear multivariate models. Economics Letters,58(1), 17–29. Rutledge, R. W., Karim, K., & Wang, R. (2013). International copper futures market price linkage and information transmission: Empirical evidence from the primary world copper markets. Journal of International Business Research,12(1), 113–131. Thomson Reuters Datastream (2019). Refinitiv. Retrieved from https://www.refinitiv.com/en Wang, Y., Zhang, Z., Li, X., Chen, X., & Wei, Y. (2019). Dynamic return connectedness across global commodity futures markets: Evidence from time and frequency domains. Physica A: Statistical Mechanics and its Applications, 123464. Retrieved from https://www. sciencedirect.com/science/article/pii/S0378437119319326?via%3Dihub World Bank. (2018). Commodity markets outlook, April 2018, World Bank, Washington, DC. World Bank (2020). Worldwide governance indicators. Retrieved from https://info.worldbank.org/governance/wgi/ Xiao, B., Yu, H., Fang, L., & Ding, S. (2019). Estimating the connectedness of commodity futures using a network approach. Journal of Futures Markets,40(4), 598–616. Yang, J., & Leatham, D. J. (1999). Price discovery in wheat futures markets. Journal of Agricultural and Applied Economics,31(2), 359–370. Zhang, B., & Wang, P. (2014). Return and volatility spillovers between China and world oil markets. Economic Modelling,42, 413–420. Zou, H., & Hastie, T. (2005). Regularization and variable selection via the elastic net. Journal of the Royal Statistical Society: Series B (Statistical Methodology),67(2), 301–320. How to cite this article: Siklos PL, Stefan M, Wellenreuther C. Metal prices made in China? A network analysis of industrial metal futures. J Futures Markets. 2020;40:1354–1374. https://doi.org/10.1002/fut.22125 APPENDIX A This appendix shows the connectedness table for the entire network of industrial metal markets considered in this article. The following numbers are used to refer to the different futures markets: 1. Aluminum–COMEX 2. Aluminum–LME 1370 | SIKLOS ET AL. 3. Aluminum Alloy LME 4. Aluminum–MCX 5. Aluminum–SHFE 6. Cobalt–LME 7. Copper–COMEX 8. Copper–LME 9. Copper–MCX 10. Copper–SHFE 11. Ferrosilicon–ZCE 12. Iron Ore–DCE 13. Iron Ore–COMEX 14. Lead–LME 15. Lead–MCX 16. Lead–SHFE 17. Nickel–LME 18. Nickel–MCX 19. Nickel–SHFE 20. Silicon Manganese–ZCE 21. Steel Scrap–LME 22. Steel Rebar–LME 23. Steel Coils–COMEX 24. Steel Rebar–SHFE 25. Steel Coils–SHFE 26. Tin–LME 27. Zinc–LME 28. Zinc–MCX 29. Zinc–SHFE SIKLOS ET AL. | 1371 TABLE A1 Connectedness table (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (1) 53.62 9.75 0.46 7.26 1.25 0.08 2.70 3.67 2.50 0.29 0.13 0.03 0.77 2.01 2.09 (2) 5.44 29.91 1.39 22.12 1.14 0.00 4.22 5.17 4.46 0.42 0.00 0.01 0.75 2.68 2.77 (3) 0.63 3.49 75.07 3.70 0.68 0.17 1.33 1.56 2.23 0.72 0.02 0.14 0.77 1.05 1.40 (4) 4.15 22.64 1.51 30.62 0.97 0.01 3.48 3.85 5.65 0.41 0.01 0.07 0.65 1.79 3.72 (5) 2.32 8.53 0.73 7.03 34.08 0.07 3.65 3.89 3.29 7.01 0.03 0.12 1.47 1.72 1.56 (6) 0.14 0.00 0.21 0.05 0.19 94.45 0.03 0.43 0.06 0.27 0.12 0.07 0.20 0.55 0.17 (7) 1.03 2.88 0.36 2.32 0.84 0.00 20.42 16.72 15.35 1.88 0.00 0.01 1.94 4.57 3.63 (8) 1.33 3.35 0.41 2.44 0.77 0.08 15.87 19.39 13.63 2.05 0.00 0.06 1.78 5.12 3.70 (9) 0.97 3.10 0.62 3.82 0.64 0.01 15.58 14.58 20.73 1.71 0.03 0.01 1.56 3.35 4.92 (10) 0.82 2.21 0.39 1.83 3.11 0.06 12.16 12.53 10.51 14.32 0.05 0.21 2.05 3.46 2.81 (11) 0.23 0.07 0.02 0.07 0.09 0.09 0.05 0.05 0.15 0.28 72.37 0.34 0.68 0.05 0.05 (12) 0.26 0.53 0.23 0.57 0.42 0.06 1.82 1.94 1.52 1.50 0.34 59.34 12.14 0.76 0.59 (13) 0.65 1.22 0.43 1.01 0.88 0.08 4.67 4.71 3.80 2.80 0.38 3.48 40.58 1.59 1.22 (14) 1.06 2.53 0.40 1.66 0.44 0.14 6.33 7.46 4.57 1.03 0.01 0.07 0.91 28.20 18.01 (15) 1.09 2.61 0.52 3.43 0.29 0.04 4.97 5.34 6.65 0.75 0.01 0.04 0.68 17.85 27.95 (16) 0.67 1.47 0.21 1.25 1.72 0.17 3.76 4.73 3.65 4.51 0.04 0.13 1.22 10.22 9.05 (17) 0.98 4.04 0.25 3.20 0.93 0.00 7.22 7.98 6.47 0.95 0.00 0.00 1.65 3.46 2.86 (18) 0.86 3.71 0.33 4.74 0.80 0.00 6.61 6.65 7.95 0.82 0.01 0.01 1.50 2.24 3.73 (19) 0.73 2.70 0.32 2.48 1.99 0.00 5.19 5.58 5.02 5.73 0.00 0.14 1.51 2.12 2.11 (20) 0.05 0.23 0.10 0.41 0.21 0.11 0.36 0.41 0.57 0.81 17.52 0.55 1.15 0.51 0.40 (21) 0.14 0.37 0.17 0.39 0.40 0.15 0.93 1.06 0.94 0.85 0.76 0.26 4.41 0.37 0.22 (22) 0.04 0.12 0.08 0.32 0.12 0.10 0.72 0.55 0.79 0.25 0.32 0.39 4.18 0.22 0.27 (23) 0.03 0.01 0.10 0.02 0.01 0.19 0.08 0.02 0.01 0.20 0.09 0.00 0.21 0.06 0.26 (24) 0.49 0.89 0.19 0.85 0.89 0.30 2.33 2.66 2.11 3.85 1.76 2.57 9.05 1.24 0.95 (25) 0.45 0.93 0.08 0.61 1.72 0.09 2.40 2.39 1.92 3.33 1.34 1.05 7.17 0.96 0.78 (26) 1.21 1.95 0.28 1.46 1.28 0.07 4.58 5.37 3.88 2.13 0.06 0.25 1.86 2.87 1.82 1372 | SIKLOS ET AL. TABLE A1 (Continued) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (27) 1.61 2.92 0.37 2.21 0.54 0.05 7.05 8.49 5.51 1.06 0.00 0.06 1.80 7.60 5.36 (28) 1.32 2.83 0.60 3.80 0.36 0.02 6.38 6.67 7.33 0.79 0.00 0.03 1.79 5.20 8.08 (29) 1.02 1.97 0.38 1.86 2.13 0.08 5.19 6.11 4.58 5.52 0.04 0.38 1.67 4.69 4.19 to 29.70 87.07 11.15 80.91 24.77 2.23 129.67 140.58 125.10 51.91 23.06 10.49 65.51 88.31 86.72 (16) (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) (27) (28) (29) from (1) 0.10 2.21 1.88 0.34 0.00 0.12 0.03 0.02 0.13 0.24 1.24 3.69 3.08 0.31 46.38 (2) 0.13 5.08 4.54 0.35 0.05 0.17 0.04 0.00 0.10 0.19 1.11 3.75 3.67 0.32 70.09 (3) 0.00 0.79 1.02 0.56 0.08 0.21 0.08 0.08 0.13 0.00 0.41 1.19 1.94 0.55 24.93 (4) 0.10 4.11 5.94 0.41 0.12 0.19 0.13 0.01 0.18 0.08 0.85 2.90 5.05 0.41 69.38 (5) 1.96 3.58 3.19 2.64 0.13 0.33 0.11 0.003 0.60 1.10 1.33 2.72 2.43 4.37 65.92 (6) 0.45 0.02 0.07 0.05 0.14 0.23 0.14 0.18 0.67 0.15 0.13 0.33 0.16 0.36 5.55 (7) 0.33 6.20 5.52 0.50 0.06 0.28 0.20 0.02 0.20 0.27 1.78 6.18 5.65 0.86 79.58 (8) 0.64 6.50 5.28 0.57 0.07 0.31 0.14 0.00 0.29 0.28 1.99 7.08 5.62 1.26 80.61 (9) 0.52 5.63 6.74 0.56 0.12 0.29 0.22 0.003 0.23 0.21 1.53 4.91 6.59 0.80 79.27 (10) 2.28 4.68 4.16 3.81 0.19 0.36 0.15 0.03 1.21 0.89 1.71 4.79 4.23 5.00 85.68 (11) 0.12 0.04 0.04 0.01 18.51 0.89 0.38 0.08 3.09 1.81 0.09 0.08 0.06 0.21 27.63 (12) 0.44 1.70 1.50 0.70 0.55 0.99 1.10 0.00 4.09 1.70 0.77 1.43 1.35 1.67 40.66 (13) 0.95 3.31 2.92 0.90 0.51 2.87 2.55 0.11 4.58 3.91 1.64 3.55 3.50 1.20 59.42 (14) 1.36 4.05 2.54 0.18 0.18 0.16 0.10 0.02 0.25 0.24 1.54 9.23 6.37 0.97 71.80 (15) 1.27 3.32 4.19 0.23 0.13 0.09 0.10 0.08 0.17 0.19 0.97 6.46 9.80 0.78 72.05 (16) 26.61 2.66 2.27 1.89 0.12 0.17 0.06 0.04 1.20 0.44 0.80 6.75 5.90 8.30 73.39 (17) 0.38 23.75 19.49 1.42 0.01 0.54 0.30 0.01 0.11 0.32 2.32 5.77 4.93 0.66 76.25 (18) 0.39 20.00 24.37 1.47 0.00 0.44 0.30 0.02 0.10 0.18 1.68 4.37 6.15 0.58 75.63 (19) 1.55 14.98 13.73 20.30 0.01 0.51 0.28 0.01 0.42 0.31 1.87 3.62 3.57 3.25 79.70 (20) 0.07 0.16 0.13 0.03 68.51 0.17 0.13 0.03 3.12 1.48 0.11 0.89 0.64 1.12 31.49 (Continues) SIKLOS ET AL. | 1373 TABLE A1 (Continued) (16) (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) (27) (28) (29) from (21) 0.15 1.45 1.16 0.59 0.17 60.90 16.90 0.11 1.87 1.77 0.47 1.18 1.37 0.49 39.10 (22) 0.02 0.95 0.95 0.32 0.12 18.10 66.17 0.69 1.39 1.05 0.34 0.45 0.83 0.14 33.83 (23) 0.14 0.02 0.07 0.01 0.04 0.18 1.01 97.08 0.05 0.05 0.01 0.02 0.01 0.04 2.92 (24) 1.88 2.75 2.32 1.00 1.91 1.54 1.19 0.03 41.19 6.96 0.94 2.26 2.10 3.81 58.81 (25) 0.76 1.53 1.12 0.46 1.15 1.68 1.00 0.03 8.86 53.94 0.46 1.26 1.03 1.52 46.06 (26) 0.19 5.07 3.62 1.41 0.05 0.38 0.25 0.00 0.48 0.18 52.50 3.45 2.23 1.11 47.50 (27) 0.69 5.69 4.20 0.24 0.23 0.44 0.15 0.00 0.19 0.20 1.53 23.20 16.52 2.09 76.80 (28) 0.63 4.78 5.81 0.30 0.13 0.50 0.28 0.003 0.23 0.11 0.98 16.35 22.97 1.72 77.03 (29) 4.68 3.82 3.32 2.33 0.36 0.40 0.15 0.01 1.40 0.48 1.17 13.89 12.22 15.98 84.02 to 22.17 115.07 107.71 23.27 25.15 32.53 27.47 1.62 35.32 24.77 29.76 118.55 117.01 43.90 57.98 Note: The main body of this connectedness tables shows the net pairwise directional spillovers between the different futures markets. The right‐most column and the bottom row summarize the total directional spillovers from and to other markets. 1374 | SIKLOS ET AL.