Demystifying commodity futures in

John Hua Fan , Tingxi Zhang

Department of Accounting, Finance and Economics, Griffith School Griffith University

First draft: December 15, 2017 Current version: February 25, 2018

Abstract

This paper presents the most comprehensive study to date on commodity futures in China. We find that passive long-only investments deliver poor economic returns. Among 12 long-short strategies examined, momentum and term structure strategies generate statistically significant economic profits in nearby and distant contracts, illiquid markets and randomly selected commodity sectors. Our results cannot be attributed to aggregate market risks, none-tradable macroeconomic risks, commodity specific risks, market sentiment, transactions costs and data- snooping. We show that liquidity, anchoring bias and regulation induced limits-to-arbitrage provide at least a partial explanation. Furthermore, our findings suggest that long-short strategies that exploit past returns and hedging pressure make excellent candidates for hedging against movements in traditional assets in China. This paper also highlights the urgency to establish a CFTC-type repository for positions data that distinguish hedgers and speculators. Such data are essential to assess the effectiveness of risk transfers in these markets.

JEL Classification: G13, G14, G15, G41, N25, Q02

Keywords: China, RQFII, Commodity Futures, Momentum, Term Structure, Hedging Pressure,

Open Interest, Liquidity, Inflation, Diversification, Limits-to-arbitrage

* Corresponding author. 170 Kessels Road, Nathan, Queensland 4111, Australia. Tel: +61 7 3735 3948; Fax: +61 7 3735 7760. E-mail: [email protected] (John Fan).

1 1 Introduction

Since the beginning of the millennium, commodity markets in China have seen tremendous developments. With the help of a maturing regulatory system, the trading volume of commodity futures market soared from a mere 3 trillion RMB in 2001 to 227 trillion RMB in 2010 (China Futures Association, 2012). The once extraneous market in the global commodities trade has become a significant force that influences the dynamics of international commodity prices (WSJ, 2016a; Bloomberg, 2018). Not only have the Chinese exchanges topped the biggest commodity exchanges globally, products such as soybean meal and steel rebar have become the world’s most actively traded instruments (FT, 2016; WSJ, 2016b). Although there has been a slowdown in recent years due to regulatory intervention, there is little doubt about the continuation of the growth, given the vast amount of resources needed to fulfil the economic transition from manufacturing to consumption (FT, 2017). Despite the unprecedented growth, the commodity futures market in China remains a “mystery” to practitioners and academics alike (Bloomberg, 2017a; Li, Zhang, & Zhou, 2017). Using a broad sample of 30 commodities traded across Dalian, and Zhengzhou commodities exchanges, this paper presents a comprehensive analysis of commodity futures in China.

While a vibrant literature on commodity futures exists in the U.S. and European markets, studies focusing on Chinese commodity futures appear to be of paucity. Given the immaturity of the market, early studies have primarily focused on qualitative market developments (Williams, Peck, Park, & Rozelle, 1998). As the market becomes more mature, the research attention turned to market efficiency and linkages with the US market. Fung et al. (2003) compare the pricing impact and information transmission between markets in China and the US and conclude that the less regulated products in China are affected more by the US market. Xin et al. (2006) find that copper and aluminium futures markets in China are efficient and they attribute the efficiency to improved regulations and skills of participants. Chan, Fung, and Leung (2004) find asymmetric responses in the Chinese market after analysing copper, mung beans, soybeans, and wheat, and also show that volatility is positively related to trading volume and negatively related to open interest.

Due to the growing importance of the Chinese economy on global trade and investments, the Chinese commodity markets have become increasingly informative for the global pricing dynamics. Recent studies have examined the broad market performance (Tu, Song, & Zhang, 2013), trend-following strategies (Li, Zhang, & Zhou, 2017) and pricing implications (He, Jiang, & Molyboga, 2017). Tu et al. (2013) argue commodity futures can be a diversifier for

2 Chinese asset management firms. They conclude that the correlation between the Chinese market and the US market has increased during the period 2000-2010. Li et al. (2017) find that trend following strategies outperform buy and hold strategies. He et al. (2017) document returns of 17.5% and 14.5% per annum for cross-sectional and time-series momentum strategies, respectively. Kang and Kwon (2017) find the momentum profits in Chinese markets are large and significant, with monthly returns ranging from 0.84% to 2.13%. Overall, the current state of the commodities literature on China suggest that the literature is in its infancy. Therefore, addressing this urgent gap is of interest to both academia and the investment management industry, as such investigation provides better insights into the pricing dynamics, which in turn aids in better understanding the risk and return trade-off.

In this paper, we aim to demystify the Chinese commodity futures markets by investigating the performance of long-only and 12 long-short investment strategies which are theoretically backed and empirically validated. To do this, we first take a deep dive into the unique institutional settings of the Chinese futures markets, then proceed to examine two categories of strategies. The first strand of strategies is motivated by the Hedging Pressure Hypothesis (Cootner, 1960) and the Theory of Storage (Kaldor, 1939). These strategies exploit information on roll-yield (Gorton & Rouwenhorst, 2006), past returns (Miffre & Rallis, 2007), net positions of hedgers/speculators (Basu & Miffre, 2013) and skewness (Fernandez-Perez, Frijns, Fuertes, & Miffre, 2018). The second group of strategies is inspired by economic and empirical intuitions which exploits information such as value (Asness, Moskowitz, & Pedersen, 2013), open interest ( & Yogo, 2012), currency beta, inflation beta, volatility and liquidity (Szymanowska, de Roon, Nijman, & van den Goorbergh, 2014).

The findings uncovered in this paper present four key contributions to the commodities literature. First, we document several institutional settings unique to the Chinese markets. One of the most striking characteristic is the fact that the vast majority of investors are individual or retail rather than institutional. For example, by the end of 2015, there were a total of 1.07 million investors, of which 1.04 million were individual investors. This is clearly in contrast with developed markets in the US. As institutional investors are known to be more informed than retail investors, this unique composition implies that the price discovery mechanism is likely very different. Furthermore, direct participations by foreign individual investors in the

3 Chinese futures market are currently restricted, although this is likely going to be loosen soon.1 Besides, strict price and position limits systematically constrain the price discovery and speculative practices. For example, with the common wheat contract, the maximum number of long or short positions that non-futures company members and retail investors can hold between the first trading day of the contract and the 15th calendar day of the month prior to the delivery month are limited to 2000. This quota then reduces to 600 for the remaining calendar days of that month. Only 200 long or short positions are granted to non-futures company members during the delivery month, while a retail investor is prohibited to trade any contract in the delivery month. A forced liquidation process will be triggered if an investor’s holding positions exceed the defined limit. This means that speculators are involuntarily pushed towards the more distant part of the futures curve.2

Second, we demonstrate that passive long-only type investments do not yield statistically significant economic profits in China, regardless of the time and sector specifications or weighting schemes employed. This may help explain the fact that no products are known to investors for benchmarking the performance of Chinese commodity markets.3 Specifically, we show that energies performed the worst whereas oilseeds have outperformed the rest of the market. In the meantime, metals, oilseeds and energies exhibit non-synchronized growth with the broad market from 2004 leading to the global financial crisis (GFC). This along with the recent decline mirrors the rapid growth and subsequent slowdown experienced by the Chinese economy before and after the GFC. Our findings suggest that the theory of normal backwardation (Keynes, 1930; Hicks, 1939) does not hold well in China. The theory of normal backwardation suggests that hedgers are net short and speculators are net long. Speculators receive risk premium as a compensation for taking on the price risk of hedgers. However, since there is no information available on the type of traders, and given that more than 95% of participants are individual investors who may act as speculators or hedgers, we

1 A new directive stipulates that individuals with foreign citizenship as eligible investors on the upcoming crude oil contract on the Shanghai International Energy Exchange (INE). This framework also entitles qualified foreign brokerage firms to directly trade in the exchange on behalf of their customers, instead of conducting trades via a domestic intermediary. 2 Consequently, these position limits create a regulation induced limits-to-arbitrage. As speculators cannot trade sufficient volumes in the front contracts, they are forced to move to more distinct contracts in order to gain continuous exposures. As a result, the limits-to-arbitrage implies that pricing dynamics can vary significantly from front contracts to the more distant contracts. 3 The COMPASS China Commodity Index (CCCI) is available from 2008 and covers 31 commodities as of March 2018 (see compassft.com/ccci/). The CFMMC China Commodity Futures Index (CCFI) was introduced by the China Futures Market Monitoring Center in 2015. The CCFI covers 20 commodities as of January 2018 (See www.cfmmc.com/main/views/index.html).

4 conjecture that the failure of long-only investments in the Chinese market may be partially explained by speculators overpowering hedgers thus reducing the risk premium.

Third, we find that five out of 12 long-short strategies generate returns on average of 14.69% per annum and statistically significant. These strategies include term structure, hedgers’ hedging pressure (HHP), cross-sectional and time-series momentum and volatility. Consistent with the previous studies, we find strong price persistency within the sample period. The success of the term structure and HHP strategies suggest that commodity futures prices in China behave in accordance with the predictions of the Hedging Pressure Hypothesis (Cootner, 1960) and the Theory of Storage (Working, 1949), even though our hedging pressure measure differs from those employed in the US literature.4 Strikingly however, we find that liquidity strategy generate economic losses which are almost significant at 10% level. We demonstrate that illiquid commodities underperform the more liquid ones, suggesting that investors do not receive compensation for bearing illiquid risk in this market, but instead they pay a price to own illiquid commodities. This implies that certain groups of market participants may be forced to hold illiquid commodities due to market frictions or limits-to-arbitrage. Meanwhile, contrary to Fernandez-Perez et al. (2018), we find that commodities with higher skewness significantly outperform commodities with lower skewness. The failure of skewness strategy implies that the link between inventory and skewness is currently weak in China.

Strategies that do not report significant profits also contain unique information about the Chinese market. First, although inspired by the same theoretical foundation, speculators’ hedging pressure (SHP) strategy fails to deliver a statistically significant profit. Since there are no positions data available on hedgers and speculators in the Chinese market, it is difficult to gauge the quality of the proxy for SHP. Second, the negative and insignificant results on value strategy suggest that long-term losers in this market do not become winners in the long-run. The lack of reversal could provide insights into the remarkable profits achieved by momentum strategies. Third, the failure of inflation and FX beta strategies may be attributed to the weaker than expected correlations between commodity futures and macroeconomic factors. Furthermore, the Chinese government has long been criticized for frequently intervening in the RMB currency markets to maintain tight control over the value of the RMB. These intervention activities may have artificially contaminated valuable trading signals embedded in the

4 We conduct an extensive suite of robustness tests including re-estimating strategy returns with distant contracts, in illiquid markets and randomly selected commodity sectors, and by varying portfolio breakpoints. Regardless of the parameters, term structure and momentum profits remain strong and statistically significant.

5 correlation between commodity prices and the RMB. Finally, contrary to Hong and Yogo (2012), our results on open interest strategy imply that the predictive power of open interest is questionable in China.5

Fourth, while long-only investments are not effective tools for hedging against movements in traditional assets in China, long-short strategies offer great potential for diversifications. The commodity market and the market boast a significant correlation of 42.2%, suggesting that commodity prices are closely intertwined with domestic . Although Basu and Miffre (2013) also estimate a positive correlation between long-only commodity portfolios and the S&P 500 Index, the magnitude is much lower than those in China. Overall, the superior diversification feature of long-short strategy relative to a long-only investment in the Chinese commodity futures market is consistent with findings in US market by Basu and Miffre (2013). More importantly, HHP strategy reveals a significantly negative correlation with the , which appears to be a more promising tool for diversification than in the US. These findings remain consistent when we consider time-varying correlations and performances in different market conditions.

To ensure that the observed profits are not masked by existing risk factors, we first investigate the link between strategy returns and commodity specific risks. Although they cannot fully explain the return dynamics of each other, term structure and momentum strategies are positively related. Nevertheless, momentum appears to be considerably stronger in China. Furthermore, we show that the Bakshi, Gao, and Rossi (2017) three-factor model cannot explain the dynamics of hedger’s hedging pressure and the volatility strategies. Our findings are supported by He et al. (2017), who find that the three-factor model are not able to fully explain the cross-sectional return variations in China. In addition, we demonstrate that the profitability of long-short strategies cannot be attributed to commodity sectors, aggregate market risks, none-tradable macroeconomic risks, changes in market sentiment, transactions costs and data-snooping. However, we show that illiquidity and anchoring bias provide a partial explanation. Finally, we argue that the observed profits on the nearest contracts are artificially inflated by the regulation induced limits-to-arbitrage. As strategy performance deteriorates

5 The core assumption made by Hong and Yogo (2012) is that open interests will rise due to increasing hedging demand. However, given that the Chinese market is heavily dominated by individual investors, we argue that hedging demand has less impact on the trading activities, as most individual investors are speculation-motivated which leads to more frequent trading rather than holding positions for a prolonged period of time.

6 from front towards further distant contracts, and since liquidity is more abundant, the mispricing is mitigated in the second nearest contracts and beyond.

Our paper presents several policy implications. First, we urge the State Council Securities Commission (CSCR) and other government authorities to re-classify traders in accordance with their business purposes. Currently, the commodity futures market in China is characterized as a retail market and the classification of trader types is non-existent. Second, we highlight the immediate need to collect positions data (i.e. long and short open interests for each commodity futures and options) classified by producers/processor/manufacturer, swap dealers and managed money, similar of those reported by the US Commodity Futures Trading Commission (CFTC). Such data are critical in assessing the effectiveness of futures markets as mechanisms of risk transfer. Last but not the least, we call for the exchanges to loosen and gradually eliminate position and price limits, and continue to encourage participation by institutional investors both in China and internationally (see Bloomberg, 2017b; Reuters, 2017). Speculators provide liquidity to hedgers and other speculators, thereby facilitating the information discovery in futures markets. Consequently, the constraints on prices and positions might impair the efficient functioning of futures markets and ultimately escalate the cost of hedging.

The remainder of the paper proceeds as follows. Section 2 discusses the unique institutional settings of futures markets in China, followed by the sample selection in Section 3. Section 4 details the portfolio construction procedures. Section 5 presents the empirical results which are further analysed in Section 6. The paper concludes in Section 7.

2 Institutional Settings

2.1 A brief walk down the great wall of commodity

As we attempt to demystify commodity futures in China, it is important to first understand its origin and history. The story began with a grain wholesale market established on 12th October 1990 in Zhengzhou, the capital city of Henan Province, a major agricultural production region. This market was the first trial of the Futures Research Group of the State Council, formed in 1988 when the Premier stressed the urgency of establishing a futures market to meet the new demand of the market-oriented economic reform that began in the 1970s. Initially, the grain wholesale market was for spot transactions, but quickly evolved to futures-type trading

7 (Williams et al., 1998). This first trial was the first step in developing hedging tools for farmers to reduce the risks associated with price fluctuations.

In 1991, the first commodity exchange was founded in Shenzhen, the Shenzhen Metal Exchange. Shenzhen was the first Special Economic Zone and the frontier of the economic reform in China.6 This provoked the first boom of futures exchanges. More than 50 exchange organisations were established within two years, which were accompanied by more than 50 commodities products and more than 1000 brokerages (Xin et al., 2006). This was a chaotic time, as same products were traded on different exchanges with varying trading rules. As a result, excessive speculation and fraud led to an overhaul initiated by the Chinese government, known as the first rectification. At the beginning of 1994, the State Council and the General Office of State Council announced stringent changes such as the implementation of a strict licencing system, capital requirements and the clarification of the regulatory bodies - State Council Securities Commission (SCSC) and the China Securities Regulatory Commission (CSRC). 7 As a result, the number of exchanges were reduced to 14 by 1996. Although a preliminary regulatory framework was established during the first rectification, excessive speculative activities remained. Market manipulations were undertaken by large speculators, which led to several notorious market events. The major type of manipulation during 1994-1998 was referred to as “long-squeezing” by Møllgaard (1997) and Weiner (2002), wherein influential traders simultaneously held more forward positions than the available supply of the underlying spot commodity. Consequently, both futures prices and spot prices were driven up, demand was artificially created, and short-side investors suffered significant floating losses in settlement and difficulties in delivery. These negative impacts considerably distorted the economic function of futures trading. In 1999, to deepen regulation, the number of futures exchanges was further reduced to three, with only 7 contracts remaining (as opposed to 35 contracts in 1998). Consequently, trading volumes decreased substantially. This temporary retreat gave the regulators the opportunity to standardise the market, which effectively improved the efficiency and

6 The special economic zone is a region in which business and trade laws are different from the rest of and serves as an experimental economic entity aiming to attract foreign investment and foster national economic reform. The currently legitimate special economic zones include Shenzhen, Zhuhai, Xiamen, Shantou, and Hainan, all of which are in the southeast coastal area. 7 The General Office of State Council is an administrative agency of the State Council, which assists the Council leaders with day to day operations. Although it is a parallel organisation to other ministry-level entities, announcements made by the General Office of State Council normally mark urgency at national level, while actions taken by the ministry are usually an industry-level matter. For detailed information, see http://english.gov.cn/. The CSRC is the SEC equivalent in China. In 1998 April, SCSC merged into CSRC which then became the primary regulation authority.

8 sustainability of the market. Trading activity first peaked around 1996, then plummeted during 1999-2000. This was followed by a rebound, with a steady growth until 2008. Notably, the Chinese futures market experienced remarkable growth during the Global Financial Crisis. This may be partly due to the “Four Trillion Chinese Economic Stimulus Plan” launched in 2008.

Given the history of a speculative “market culture” and the 10-year monotonical growth in commodity markets since 2000, the authorities executed several modifications to further tighten commodity trading. For example, the Dalian Commodity Exchange (DCE) temporarily suspended the discount on transaction fees.8 This was followed by Shanghai Futures Exchange (SHFE), which increased the margin requirements to 10 percent for copper, aluminium, steel wire rod, gold, and fuel oil futures contracts; 12 percent for zinc and steel rebar contracts; and 13 percent for natural rubber contracts. Moreover, the daily price limits of the most active contracts were confined within 6 percent of the previous settlement price (Zhao, 2015). Starting from 2011, newly launched contracts were mainly big-scale.9 This fundamentally constrained speculation, and it also made the Chinese market more compatible with international commodity markets.

After decades of consolidation, three exchanges survived the reshuffle, namely the Shanghai Futures Exchange (SHFE), Zhengzhou Commodity Exchange (ZCE), and Dalian Commodity Exchange (DCE). These exchanges, along with the new-born China Financial Futures Exchange (CFFEX), constitute the contemporary Chinese futures market. We now turn our attention to the three commodity futures exchanges and the products traded.

2.2 Exchanges and products

Originally established to meet farmers’ and producers’ hedging needs, commodity markets now function as an essential element in the Chinese economy. These exchanges have several unique features which differentiate them from each other and from their international peers. Firstly, all the futures exchanges are directly managed by the government in the form of non- profit organisations, rather than being a private or publicly listed organization. This means

8 For expansionary purposes, if one investor opens one position and then closes this position on the same day (which are two transactions), the total transaction fee will be halved. For contractionary purposes, this discount policy will be cancelled, to reduce the trading incentives of short-term investors and constrain the overall market activities. 9 Xin et al. (2006) mentioned that those small-scale commodities traded during 1994-1998 experienced over- speculation by big speculators, which drove up the market risk. Soybean meal, one of the early contracts, is set as 10 tons per contract, while the newly launched product thermal coal is designed as a big-scale contract of 100 tons per contract.

9 political risk or managerial risk are always embedded in the market. Secondly, each exchange specialises in a different area. ZCE and DCE cover a wide array of agricultural and chemical products, while SHFE focuses on metal and the energy sectors. Thirdly, the exchanges could frequently modify any contracts for either expansionary or contractionary deeds. For example, in 2012, the strong gluten wheat contract was altered twice to improve trading activities and liquidity, with the contract size increased from 10 to 50 tons per contract in January, and then reduced to 20 tons per contract in July. Overall, these characteristics mark the developmental nature of the market, and indicate a relatively high uncertainty. After more than a decade of development, there are now 48 commodity futures contracts traded across three exchanges. As of 2017, 14 different commodities contracts are listed on SHFE, including from oldest to latest: Aluminium, Copper, Natural Rubber, Fuel Oil, Zinc, Gold, Steel Rebar, Steel Wire Rod, Lead, Silver, Bitumen, Hot-Rolled Coil, Tin and Nickel. The DZE trades 16 products, namely Soybean Meal, No.1 Soybean, Corn, No.2 Soybean, Soybean Oil, Linear Low-Density Polyethylene (LLDPE), RBD Palm Olein, Polyvinyl Chloride (PVC), Metallurgical Coke, Block Board, Medium Density Fibreboard, Fresh Hen Egg, Ire Ore, Coking Coal, Polypropylene (PP) and Corn Starch. The remaining 17 futures contracts traded on the ZCE are Strong Gluten Wheat, Cotton No.1, White Sugar, Pure Terephthalic Acid (PTA), Rapeseed Oil, Common Wheat, Early Long-Grain Non-Glutinous Rice, Methanol, Rapeseed, Rapeseed Meal, Flat Glass, Thermal Coal, Japonica Rice, Ferrosilicon, Manganese Silicon, Late Indica Rice, Cotton Yarn and Apple. 10 Notably, commodity futures markets in China have expanded rapidly with approximately three new contracts being established per year since 2004, consistent with the enormous rise in trading volume presented in Figure 1. With increasing trading activities and improved regulations, the Chinese commodities futures market started to play a vital role in the domestic economy. For example, in 2001, the three exchanges in total delivered 151.11 million tonnes of materials. Particularly, a quarter of the copper spot trading volume was conducted through futures exchanges (Xin et al., 2006). Moreover, the copper futures price has been set as the base price to implement macroeconomic control. Likewise, the soybean and wheat futures prices have been referred to by the State Grain Administration to manage the grain reserve. The standardisation and reorganisation in 2000 led to much improved market efficiency and pricing dynamics, which enhanced the role that Chinese commodity markets played in the global economy. Evidence has shown that the interactions between the Chinese and US commodity markets have increased during 2000-2010

10 Initially launched as Hard White Wheat in 1993, it was modified and renamed as Common Wheat in 2012.

10 (Tu et al., 2013). In 2011, ZCE, SHFE and DCE were ranked as the 11th, 14th and 15th exchanges globally, respectively. Cotton and Steel Rebar futures contract topped the world No.1 agricultural and metal commodity futures respectively in terms of trading lots. More recently in 2015, all three exchanges were ranked in the top 10 exchanges worldwide (top 5 for agricultural and metal commodity futures).11 Cotton futures prices have become an important benchmark in the global cotton trades. Instead of directly investing in futures contracts, there are other instruments that enable investors to gain exposure on commodity assets. These include the listed stocks of commodity- producing companies, as well as other companies whose products or services are directly or indirectly related to pertinent commodities. However, these stocks do not indicate persistent synchronisation with the commodity market. These stocks are labelled as “conceptive shares” when they are in an uptrend driven by a bullish performance of the underlying commodities. For example, some famous gold-related companies listed in the Chinese equity market include Shandong Gold (ticker: 600547), Zhongjin Gold (600489), Hunan Gold (002155), and Gansu Ronghua Industry Group (600311). All of them once experienced boom and bust along with the international gold market. A mutual fund that invests in those commodity-related stocks is also an option to gain commodity exposure. However, not only is the pricing mechanism different from commodity futures, trading rules are not as flexible as in direct investment.12 Overall, there are vast amount of stocks and funds that are commodity-related in the broad Chinese financial market, which serve as indirect commodity vehicles. Although they may be partially driven by commodities, the price dynamics are very different, which makes these indirect investment vehicles less ideal for maintaining commodity exposures.

2.3 Participants and foreign investors

The participants of the Chinese futures market can be categorised into three groups. The top tier are exchanges which, according to the “Administrative Regulations on Futures Trading”13 and “Administrative Measures for Futures Exchanges” 14 , are responsible for organising trading-related matters, such as managing contracts and facilitating transactions. The middle

11 DCE ranked 8th, ZCE ranked 9th, and SHFE ranked 10th. The top 5 agricultural commodity futures are Soybean Meal, Rapeseed Meal, White Sugar, RBD Palm Olein, and Soybean Oil. The top 5 metal commodity futures are Steel Rebar, Iron Ore, Silver, Copper, and Nickel. 12 Short-sell are restricted in indirect investment approaches. Mutual funds also place constraints on liquidation, and the transaction fees are high. 13 Published by the State Council on 6th March 2007, with the newest version taking effect on first December 2012. See http://www.shfe.com.cn/regulation/exchangelaw/21135449.html 14 Published by CSRC on 15th April 2007, see htttp://www.shfe.com.cn/regulation/exchangelaw/211203689.html

11 tier are futures companies, which not only must comply with general corporation law but are subject to the approval by CSRC and relevant supervisory law. The futures companies serve as the intermediaries between the exchanges and investors. 15 In addition to specific futures companies, several other types of organisations, registered as membership holders of exchanges, also function as intermediaries, such as private futures funds, risk management companies, asset management companies, and securities brokerage companies. The bottom tier consists of investors that can be generally divided into professional and ordinary (retail) investors by the definition from “Administrative Measures for the Adequacy of Securities and Futures Investors”.16 Professional investors are qualified institutions and individuals. All others are ordinary investors. The two types of investors can convert to each other under certain circumstances. The purpose of this classification among investors is to manage risk and protect investors. For example, promoting a product with a risk level that is above the ordinary investors’ risk tolerance to an ordinary investor is prohibited, and is subject to legal penalty. Strikingly, the vast majority of the investors in these markets are individual or retail investors rather than institutions. By the end of 2015, there were a total of 1.07 million investors, of which 1.04 million were individual investors. Even though the number of institutional investors saw a 173.8% growth driven by the participation of private funds and asset management companies, it was still a mere 11,000 (ChinaFuturesAssociation, 2012). This has significant implications for commodities pricing and the performance of trading strategies, as individual and institutional investors are known to behave differently. For example, a study on the Chinese stock market found that institutional investors are momentum followers and individual investors with a lower level of wealth are contrarians, while the individual investors with higher level of wealth trade like institutional investors when buying and behave like less wealthy investors when selling (Ng & Wu, 2007). Li et al. (2017) also argue that the herding behaviour of individual investors in the Chinese futures market has two opposing effects on the strategy’s profitability. On one hand, this bias causes prices to reveal a trend, improving the trend-following strategy’s profitability. On the other hand, overuse of the strategy may reduce the overall profitability. As a foreign investor, investing in Chinese financial markets is rather difficult. Apart from interventions by the government and regulators, capital in-and-out of the country also

15 “Administrative Measures for Futures Companies Supervision” was published on 29th October 2014 by CSRC. See http://www.shfe.com.cn/regulation/exchangelaw/911322933.html 16 Published by CSRC on 12th December 2016, became effective since 1st July 2017. See http://www.csrc.gov.cn/pub/newsite/flb/flfg/bmgz/zhl/201701/t20170110_309253.html

12 faces strict barriers. Combined with the frequent policy modifications, uncertainties appear to be relatively high compared with investing in international markets such as the London Metal Exchange (LME), New York Mercantile Exchange (NYMEX) and Intercontinental Exchange (ICE). To meet the increasing foreign demand and deepen market liberalisation, QFII and RQFII programs were introduced to provide qualified foreign institutional investors with access to Chinese financial markets. Under the QFII (Qualified Foreign Institutional Investor) program, foreign institutional investors are required to apply for a quota of total investable capital.17 By April 2011, there were a total of 103 licensed QFII investors, with a combined quota of USD 20.7 billion.

The RQFII (RMB qualified foreign institutional investor) program was launched in response to the rapid growth of the RMB in offshore markets. In 2011, the People’s (PBOC) along with CSRC and SAFE jointly published the “Pilot Measures for the Securities Investment by RMB Qualified Foreign Institutional Investors of Fund Management Companies and Securities Companies” (the “Pilot Measures”) allowing qualified institutions, which include subsidiaries of domestic fund management companies and securities companies, to invest in markets in mainland China with RMB funds raised in Hong Kong.18 Notably, under the QFII and RQFII guideline, participants are only allowed to invest in Chinese stock index futures for hedging purposes. In addition, the trading activities and the value of stock index futures contracts held by QFII are subject to quota approval. In terms of commodity futures investment, the latest version of the relevant ordinance clarifies in “Article Eight” that members of QFII and RQFII schemes are eligible professional investors. The last five years have witnessed exponential growth in these programs with QFII quota increasing from USD 21 billion to 81 billion and RQFII quota rising from RMB 11 billion to 444 billion over the period 2011-2015.19 To speed up the internationalization of

17 Supervised by the CSRC and the State Administration of Foreign Exchange (SAFE), the QFII was initially established in 2002 in conjunction with the publication of the “Administrative Provisions for Securities Investment Made in China by Qualified Foreign Institutional Investors”. The latest version became effective on 1st September 2006. See http://www.csrc.gov.cn/pub/newsite/flb/flfg/bmgz/jjl/201012/t20101231_189872.html 18 This programme experienced an update on 1st March 2013, with a revised “Pilot Measure” that expanded the eligibility for pilot institutions and the range of investable RMB securities, such as stock index futures. Refer to “Pilot Measures for the Securities Investment Made in China by RMB Qualified Foreign Institutional Investors”, available at http://www.csrc.gov.cn/pub/newsite/flb/flfg/bmgz/jjl/201310/t20131021_236659.html 19 By the end of 2017, there were only four institutional investors from Australia participating in the QFII scheme: AMP Capital, Platinum Investment, Macquarie Bank, and FSS Trustee Corporation, with an approved quota of 0.5, 0.3, 0.8 and 0.5 billion USD respectively. In addition, the only two Australian institutional investors enrolled in the RQFII programme are Vanguard (with a granted quota of RMB 30 billion) and First State Super (with 3.2 billion).

13 China’s financial markets and currency, Chinese authorities are gradually loosening restrictions to meet increasing demands. Direct participation by foreign individual investors in the Chinese futures market are currently restricted. However, this could soon change, as indicated by a recent proposal where for the first time, the authorities clearly defined individuals with foreign citizenship as eligible investors on a specific product. 20 The crude oil contract (prescribed product) has undergone several rounds of across-the-market testing on the Shanghai International Energy Exchange (INE, a subsidiary of SHFE) and is expected to be officially launched in March 2018.21 Moreover, this framework also entitles qualified foreign brokerage firms to directly trade in the exchange on behalf of their customers, instead of conducting trades via a domestic intermediary. This is expected to spark more interest from international investors and boost the overall trading in Chinese commodity futures markets.

2.4 Regulations and trading rules

The primary regulator for commodity futures market in China is the China Securities Regulatory Commission (CSRC), which is an administrative organisation of the State Council of the People’s Republic of China with ministry-level jurisdiction. Market participants are required to comply with specific trading rules and administrative measures developed by local exchanges. In addition, all market participants must meet the obligations defined by both national and local associations, which are self-regulating organisations authorised by the “Administrative Regulations on Futures Trading”. If the investors are non-domestic organisations, they will face additional scrutiny by the State Administration of Foreign Exchange (SAFE). Not only does the quota of investment on specific assets need approval, but investors are also obligated to report trading activities and are subject to transaction limits. For instance, the maximum capital outflow in a typical month cannot exceed 20% of the total assets within the country as of the end of the prior year. The daily price limit mechanism is another factor that differentiates the Chinese market from its international counterparts. Using soybean futures as an example, the daily price limit is regularly reset twice a year in the US market, to reflect the market information of the previous period. However, in the Chinese market it will not be changed unless the exchanges believe it to be necessary for risk management purposes. Moreover, the daily price limit varies across the

20 Announced on 26th June 2015, the “Interim Administrative Measures for Domestic Futures Trading on Prescribed Futures Contracts by Foreign Investors and Foreign Brokerage Firms”. See www.csrc.gov.cn/pub/newsite/flb/flfg/bmgz/qhl/201507/t20150731_281991.html. 21 Details at www.csrc.gov.cn/pub/newsite/flb/flfg/bmgf/qh/qhjy/201510/t20151012_284999.html

14 contract life. According to the “Dalian Commodity Exchange Administrative Measures for Risk Management”, the daily price limit for all commodity futures contracts during non- delivery months is 4% within the settlement price of previous trading day, while it becomes 6% in the delivery month. 22 Price limits systematically constrain the price discovery and speculative practices, but it also prevents severe downside risk. Liu and An (2011) argue that the price limit mechanism adversely impacts the market in terms of reflecting new information in futures prices. Moreover, Li et al. (2017) posit that the structural differences between the Chinese and US markets cause weaker trend-following strategy performance in the Chinese futures market. Furthermore, the policy on price limit modification also vary across exchanges. Unlike DCE, which applies a relatively static mechanism on price limit and margin, SHFE adopts a market-oriented daily price limit adjustment process that resets the next day’s price limit and margin based on the previous day’s price fluctuation. For example, SHFE defines the following situation as a “one-side market” for a futures contract: On any trading day, there is only a buy (sell) request with a price that reaches the upper (bottom) price limit within five minutes of closing the market. If a “one-side market” occurs on any trading day, the price limit for that contract on the following day will be the previous day’s reference rate plus three percentage points; the required margin will also increase two percentage points in addition to the new daily price limit rate. For example, assuming a copper contract has a price limit of 5% and a margin of 4%, if the closing price increases by 5% relative to previous day settlement price, the price limit and margin requirement will automatically rise to 8% and 10% respectively on the next trading day. In addition to price limit restrictions, the number of positions an investor can hold is also regulated. The position limit rule defines the maximum number of long or short positions on a contract that an investor can hold at a particular time. For example, for a common wheat contract, the maximum number of long or short positions that non-futures company members and retail investors can hold between the first trading day of the contract and the 15th calendar day of the month prior to the delivery month is 2000.23 This quota then reduces to 600 for the

22 An exchange-level ordinance, drafted in accordance with the “Dalian Commodity Exchange Trading Rules”, periodically reviewed to reflect the current market condition, with the latest version taking effect on 24th July 2017. See http://www.dce.com.cn/dalianshangpin/fg/fz/jysgzhgz/2048658/index.html 23 Members of an exchange are the organisations who can directly trade on the exchange and collectively manage the exchange. They are separated into futures company members and non-futures company members. The latter includes security companies and risk management companies. Retail/individual investors can only trade through brokerages which must be an exchange member.

15 remaining calendar days of that month. Only 200 long or short positions are granted to non- futures company members during the delivery month, while a retail investor is prohibited to trade any commodity’s contract in the delivery month. If an investor’s holding positions exceed the defined limit, it will trigger a forced liquidation process. It is worth noting that commodities traded on the same exchange have different position limits at different periods, and exchanges define time-periods distinctively. Overall, the Chinese commodity futures markets have seen remarkable developments in recent years. A healthy and sustainable commodity futures market that facilitates risk transfer is gradually taking shape under a maturing regulatory system. While a vibrant literature on commodity futures exists in the U.S. and European markets, studies focusing on Chinese commodity futures appear to be of paucity. In light of unique institutional settings, we aim to shed light on this urgent gap in the literature by examining the performance of long-only and a suite of long-short investment strategies in this emerging commodity futures market.

3 Data

3.1 Sample selection

To examine the performance of investment strategies, we obtain the entire archive of commodity futures contracts traded in China from Datastream International. Over the period from January 1992 to June 2017, we download daily settlement prices (in RMB), trading volumes and open interests on 47 commodities with all contract maturities traded across three exchanges. This results in more than 4,000 contracts in total. We require each commodity to have at least four years of data and a minimum of 12 commodities in the cross-section to control the volatilities of portfolios.

The cleaning process resulted in a final sample of 30 commodities covering the period of February 2004 to May 2017: sugar, cotton No.1, pure terephthalic acid (PTA), methanol, flat glass, rapeseed oil, strong gluten wheat, common wheat, rapeseed meal, and rapeseed from the Zhengzhou Commodity Exchange (ZCE); aluminium, gold, copper, fuel oil, lead, steel rebar, natural rubber, steel wire rod, zinc, and silver traded on the Shanghai Futures Exchange (SHFE); and No.1 & No.2 soybean, corn, linear low density polyethylene (LLDPE), soybean meal, palm olein, polyvinyl chloride (PVC), soybean oil, metallurgical coke, and coking coal listed on the Dalian Commodity Exchange (DCE). Following Szymanowska et al. (2014), we

16 categorise commodities into five sectors in accordance with the Commodity Research Bureau (CRB) classification, which are industrial, grains and oilseeds, energy, and metal. Table 1 outlines the commodity tickers, contract size, delivery months, price limit, margin requirements, first- and end- contracts in the sample and their maturities.

3.2 Rolling futures contracts

To compile continuous futures returns, we closely follow Miffre and Rallis (2007) and Fernandez-Perez et al. (2018). We hold contracts until the last trading day of the month prior to expiration. The positions are then rolled over to the next nearest contract. For robustness, we also construct returns series based on distant contracts with mth maturity (m = 2, 3 and 4). These contracts represent the largest volume and open interests on the futures curves in China (as can be seen in Figure 1). When compiling the mth nearest-contract exposure, price changes of the mth maturity contract are held up to the last day of the month before the front contract matures.

A number of issues remain after performing the filtering procedure. Given the history of regulatory intervention, a number of commodities experienced variations in contract name, size, and symbol. For example, three commodities in the final sample have two corresponding codes in Datastream.24 Furthermore, mismatches between contract price and volume are also present. For instance, the fuel oil October 2016 contract has price information while the volume data is missing; the silver February 2015 contract shows the reverse. To fix this problem, a volume variable is created with no value assigned if the contract has price data, while the volume variable is removed if the corresponding price variable is absent. 25

Table 2 reports the annualized returns (Panel A), standard deviations (Panel B), trading volume (Panel C) and the Amihud illiquidity measures (Panel D) of commodities in the final sample. Panel A shows that individual commodities in China do not perform well in general, as the majority of the sample report large and negative returns. In general, it appears that the performance gradually improves towards more distant contracts. The returns range from 11.9% (soybean meal) to -13.7% (coking coal) on the nearest contracts and 9.6% (rapeseed meal) to - 8.7% (silver) on the 4th nearest contracts. The large ranges in returns are reflected in Panel B,

24 For rapeseed oil, “LFUTZRPD” 2007 July to 2013 May and LFUTZERD from July 2013 to May 2017. Strong gluten wheat “LFUTZWSD” May 2003 to May 2013 and “LFUTZGWD” afterwards. Common wheat, formerly called hard white wheat and assigned code “LFUTZWTD” for contracts from January 1994 to November 2012 and are labelled “LFUTZWHD” afterwards. 25 In summary, the problematic contracts are: fuel oil October 2016; natural rubber July and August 1995, January and December 1996, and January, March, April and May 1997; steel wire rod November 2014, May, August and December 2016, and April and June 2017; silver February 2015; rapeseed August 2016; and common wheat May 2017.

17 as the standard deviation ranges from as low as 9.5% (common wheat) to 36.8% per annum (coking coal). Turning to the trading volume, reflective of institutional settings discussed in the previous section, Panel C clearly shows that the number of contracts traded is concentrated in the more distant maturities. Correspondingly, the Amihud illiquidity measure in Panel D suggests that the majority of commodities present sufficient liquidity for the implementation of investment strategies.

3.3 Macroeconomic variables

To facilitate the implementation of trading strategies and the construction of risk adjustment models, several macroeconomic and financial variables in local currency are obtained. The RMB effective exchange rate index is attained from the Bank for International Settlements. The unexpected inflation rate and unexpected industrial production are computed as the difference between actual and forecasted figures estimated by Bloomberg. For movements in the stock market, four indices have been considered. Similar to the S&P 500, the CSI 300 consists of top 300 stocks traded on the Shanghai and Shenzhen stock exchanges. To capture the broader stock market movements, we also consider the Shanghai Composite Index and the Composite Index. Furthermore, to include large- and mid-cap Chinese stocks listed in Hong Kong, US and Singapore, we employ the MSCI All China Index. The Barclays China Aggregate Index is employed, as a proxy of the RMB-denominated fixed income market. The index covers fixed-rate treasury, government and corporate bonds. All macro and financial data are obtained from Bloomberg.

4 Portfolios Formation

4.1 Term Structure

Motivated by both the Hedging Pressure Hypothesis (Cootner, 1960) and the Theory of Storage (Working, 1949), the term structure strategy exploits roll-yield signals of commodities, by taking long (short) positions on contracts with higher (lower) roll-yields. The former theory argues that a higher roll-yield is driven by an over-supply by net short hedgers, which pushes down the futures price, resulting in a higher roll-yield. Conversely, a lower roll-yield is caused by an over-supply of net long hedgers, which give raise to futures prices (Fuertes, Miffre, & Rallis, 2010). The latter theory posits that a higher roll-yield is due to a higher convenience

18 yield, resulting from a lower level of inventories (Gorton, Hayashi, & Rouwenhorst, 2013). We compute roll-yield of commodity i at time t is as:

푅표푙푙푖,푡 = log(퐹푖푡,퐹푟표푛푡) − log(퐹푖푡,2) (1)

where 퐹푖푡,퐹푟표푛푡 represents the front contract price of commodity i at time t, and 퐹푖푡,2 denotes the price of the second nearest contract of commodity i at time t. We sort commodities into quartiles by roll-yields at the end of each month. This strategy takes long (short) positions on the commodities with the highest (lowest) quartile of roll-yields. These positions are held for one month, and then rebalanced based on roll-yields computed at the end of the following month. The term-structure strategy returns are computed as the difference between long and short portfolios.

4.2 Hedging Pressure

The Hedging Pressure Hypothesis by Hirshleifer (1990) postulates that risk premiums exist in both backwardated and contangoed markets. In the former market condition, hedgers are in net short positions, so speculators can earn premium by taking long positions with an expectation of price appreciation; while in the latter market condition, where hedgers are in net long positions, premium can still be accrued by speculators by taking short positions if the anticipation of price depreciation is realised (Basu & Miffre, 2013). Following this logic, hedging pressure strategy essentially employs the signals measuring hedgers’ and speculators’ net positions.

Several methods are proposed for computing hedging pressure. De Roon, Nijman, and Veld (2000) construct a ratio of hedgers’ net short positions to total number of hedge positions by extracting data from the U.S. Commodity Futures Trade Commission (CFTC). A similar hedging pressure measure created from the speculators’ perspective has been employed by other studies (Basu & Miffre, 2013; Dewally, Ederington, & Fernando, 2013).26 However, unlike the US commodity market, there is no CFTC-equivalent data repository for hedgers’ and speculators’ positions in the Chinese commodity futures market. Inspired by Bohl, Siklos, and Wellenreuther (2018), we apply the following ratios proposed by Garcia, Leuthold, and Zapata (1986) and Lucia and Pardo (2010) as proxies for hedgers’ hedging pressure (HHP) and speculators’ hedging pressure (SHP). We define the hedging ratio as:

26 Speculators’ hedging pressure is the ratio of speculators’ net long positions to total number of speculators’ positions.

19 퐻푒푑푔푒 ∆푂퐼푖,푡 푅푎푡푖표푖,푡 = (2) 푉표푙푖,푡 and the speculation ratio as:

푆푝푒푐푢푙푎푡푒 푉표푙푖,푡 푅푎푡푖표푖,푡 = (3) 푂퐼푖,푡 where 푂퐼푖,푡 and ∆푂퐼푖,푡 represent the monthly open interest and the change of monthly open interest for commodity i at time t, and 푉표푙푖,푡 denotes the total monthly volume of commodity i at time t. All measures are in the context of mth contract exposure. The core assumption behind the two ratios is that hedgers hold positions longer than speculators. Consequently, the two types of participants impact on different trading-related metrics, with speculators having more influence on trading volume as they are more likely to trade frequently, and hedgers’ impact primarily being reflected on open interest, as the outstanding contracts at the end of each month should be held by hedgers (Rutledge, 1979; Leuthold, 1983; Bessembinder & Seguin, 1993). Therefore, a higher hedging activity ratio indicates that hedgers dominate the market relative to speculators, whereas a higher speculation ratio suggests more trades are triggered by speculators. Based on such rationale, we construct the HHP strategy portfolio as the return difference between the lowest hedging activity portfolio and the highest hedging activity portfolio. For SHP strategy, we compute the return difference between the highest and the lowest speculative activity portfolios.27

4.3 Momentum

Momentum strategy in commodity futures has been extensively verified to generate persistent economic profits (Erb & Harvey, 2006; Miffre & Rallis, 2007; Fuertes et al., 2010; Asness et al., 2013; Szymanowska et al., 2014; Bianchi, Drew, & Fan, 2015, 2016; Bakshi et al., 2017). We follow the literature to examine whether past winners have the tendency to prevail in the subsequent period in the Chinese commodity futures market. We compute the sorting signal as:

27 Early studies have documented a connection between these two ratios and the commodity futures prices and volatilities. For example, Streeter and Tomek (1992) analyse the agricultural commodities in US markets and discovered a positive relationship between speculation ratio and the returns volatility of soybeans. Moreover, a Granger causality relationship between speculation ratio and the prices of wheat and rice futures was confirmed (Robles, Torero, & Von Braun, 2009). More recently, Bohl et al. (2018) conclude a positive impact of the speculation ratio on returns volatility and a negative impact of the hedging activity ratio on returns volatility in Chinese commodity futures market.

20 11 1 푀푂푀 = ( ) ∑ 푟 (4) 푖,푡 12 푖,푡−푗 푗=0

where 푟푖,푡−푗 denotes the front contract return of commodity i in the month t-j. At the end of each month t, all commodities are sorted into four portfolios based on based on 푀푂푀푖,푡. We hold long positions in the portfolio of past winners and short positions in the portfolio of past losers for one month. As an alternative momentum strategy, we follow Moskowitz, Ooi, and Pedersen (2012) in constructing the time-series momentum portfolio. We take long (short) positions in commodities with a positive (negative) 푀푂푀푖,푡.

4.4 Volatility

Generally, a risk-return trade-off features in the financial assets, which implies the higher the volatility, the higher the expected return. In the context of commodity futures investment, a higher expected return is associated with more uncertainties around future spot price, because futures contracts offer a hedge for spot price fluctuations. Dhume (2011) find holding highly volatile assets is rewarded in US commodity markets. 28 Gorton et al. (2013) and Szymanowska et al. (2014) report additional supporting evidence. Motivated by these studies, we define the volatility signal as:

2 휎 푖,푡 퐶푉푖,푡 = (5) |휇푖,푡|

The numerator denotes the front contract’s daily return variance of commodity i at time t over the previous 36-month, and the denominator represents the corresponding absolute value of the prior 36-month average daily return at time t. The volatility strategy takes long positions in portfolios with the highest volatilities, and short positions in portfolios with the lowest volatilities.

4.5 Open Interest

Hong and Yogo (2012) document a positive correlation between the movements in open interest and the movements in both futures and spot prices in commodity markets. They argue that the change in open interest is a more reliable variable of hedging demand than futures prices, as a higher hedging demand will only drive up the open interest but can either raise or

28 Dhume (2011) defines the spot price used in volatility measures as the near-month futures price, which is the front-contract-exposure price series in this study.

21 suppress the prices. Motivated by Hong and Yogo (2012), Szymanowska et al. (2014) sort commodities into quartiles by the open interest of the entire term structure, and achieve significant profits through taking long (short) positions in portfolios with higher (lower) changes in open interest. We define the open interest signal as:

∆푂퐼∆푖,푡 = 푂퐼푖,푡 − 푂퐼푖,푡−1 (6) where 푂퐼푖,푡 represents the total open interests of all available contracts for commodity i at time t. The sorting signal is the monthly change of total open interest along the term structure of each commodity. The strategy performance is computed as the return difference between the high changes of open interest commodities portfolio and the lower changes of open interest commodities portfolio.

4.6 Liquidity

Liquidity plays a vital role in pricing an asset. Intuitively, the less liquid an asset is, the more return an investor would require for holding it. Early studies have shown that stocks with higher expected returns are found to be more sensitive to aggregate liquidity measures (Pástor & Stambaugh, 2003), and that liquidity interacts with expected returns and co-moves with prices (Acharya & Pedersen, 2005). Marshall, Nguyen, and Visaltanachoti (2012) survey the relevance of the well-known stock market liquidity measures to the commodity futures market, and find that the Amihud measure reveals the highest correlation with commodity transaction costs.29 Szymanowska et al. (2014) develop a liquidity strategy using the Amivest measure, which takes long (short) positions in illiquid (liquid) commodities. Following Szymanowska et al. (2014), we define the liquidity signal as:

1 푉표푙푑 퐴푚푖푣푒푠푡푖,푡 = ∑ (7) 퐷 |푟푑| where 푉표푙푑 denotes the prior 2-month daily volume of front contract of commodity i at time t, and 푟푑 stands for corresponding daily return. Given that D is the number of days in the past 2 months, this signal can be interpreted as the average ratio of daily volume to absolute return in the past 2 months. The higher this ratio, the higher the liquidity. Consequently, we construct the liquidity strategy portfolio by examining the return difference between the portfolio that contains the commodities with lowest and highest 퐴푚푖푣푒푠푡푖,푡.

29 Amihud and Amivest measures are two different proxies for liquidity constructed by Amihud (2002).

22 4.7 Beta

We also examine the performance of two beta signals based on currency movements and inflation shocks, respectively. The economic intuition comes from the fact that commodities prices are negatively correlated with currency and positively correlated with inflation shock (Bodie & Rosansky, 1980; Erb & Harvey, 2006; Gorton & Rouwenhorst, 2006; Bhardwaj, Gorton, & Rouwenhorst, 2016). This relationship holds true in the Chinese market, as our data reveal a correlation of 0.05 between the aggregate commodity market and inflation shocks, and a correlation of -0.31 with the RMB exchange rate. This is consistent with the findings by Mallick and Sousa (2013), in which they show a persistent reduction in commodity price level is associated with a temporary appreciation of domestic currency and a decline of inflation. Following Szymanowska et al. (2014), we define the currency signal as the slope of regression of monthly commodity futures returns on effective RMB index return:30

푐푢푟푟푒푛푐푦 퐶표푣(푟푖,푡(42), 푟푐,푡(42)) 훽푖,푡 = (8) 푉푎푟(푟푐,푡(42))

with 푟푖,푡(42) denoting the prior 42 observations of monthly return of commodity i at time t,

푟푐,푡(42) representing the prior 42 observations of monthly return of effective RMB exchange index at time t, 퐶표푣(푟푖,푡(42), 푟푐,푡(42)) representing the covariance between the two variables and

푉푎푟(푟푐,푡(42)) representing the variance of 푟푐,푡(42) Likewise, the signal for inflation shock strategy can be expressed as:

푖푛푓푙푎푡푖표푛 푠ℎ표푐푘 퐶표푣(푟푖,푡(42), 푟푖푛푓,푡(42)) 훽푖,푡 = (9) 푉푎푟(푟푖푛푓,푡(42))

Currency strategy returns are computed as lower currency-beta portfolio minus high currency- beta portfolio, and inflation shock strategy profit being estimated as high inflation-beta portfolio minus low inflation-beta portfolio.

4.8 Skewness

Theory of Storage and the Theory of Normal Backwardation implicitly suggest that negative skewness coincides with a backwardated market driven by lower inventories (Fama & French, 1987; Deaton & Laroque, 1992; Symeonidis, Prokopczuk, Brooks, & Lazar, 2012; Gorton et

30 Szymanowska et al. (2014) uses a rolling window of 60 months. Given the limited observations of our data, this study applies the maximum rolling window that can be proceeded, which is 42 months.

23 al., 2013). Fernandez-Perez et al. (2018) investigate the profitability of a skewness-sorted long- short portfolio and find statistically significant profits. We follow Fernandez-Perez et al. (2018), and compute the skewness signal as:

1 3 ∑퐷 (푟 − 휇 ) 퐷 푑=1 푖,푑 푖,푡 (10) 푠푘푒푤푖,푡 = 3 휎푖,푡

The 푟푖,푑 denotes the daily return of commodity i at time d, 휇푖 represents the prior 12-month average daily return of commodity i at time t, 휎 푖 stands for the standard deviation of the past 12-month daily return at time t, and D is the number of days in the past 12 months. Consequently, the performance of the skewness strategy is measured by the return difference between the portfolio with the most negatively skewed commodities and the portfolio with the most positively skewed commodities.

4.9 Value

Asness et al. (2013) implement the “value” concept to the commodities market by adopting a “long-run” value to represent the “book” value for commodity futures studies. Essentially, the fundamental hypothesis regarding value strategy in the commodity futures market is that the long-term loser will subsequently turn into a winner. Following Fernandez-Perez, Fuertes, and Miffre (2017), we define the value signal as:

1 ∑퐷 푓 퐷 푑=1 푖푑,푓푟표푛푡 (11) 푣푎푙푢푒푖,푡 = ln 푓푖푡,푓푟표푛푡

푣푎푙푢푒푖,푡 is the log of the average daily price of front contract from 4.5 to 5.5 years prior divided by the front contract price at time t. A higher value signal implies the current price is lower than “long-run” value and is likely to rise in the following period. The performance of value strategy can be captured by taking long positions in the commodities with the highest value signal and short positions in commodities with the lowest value signal.

5 Strategy Performance

5.1 Passive long-only

24 Table 3 reports the performance of long-only investments in board markets as well as individual sectors. The overall market is represented by an equally-weighted portfolio consisting of 30 commodities in the sample. The individual sectors are energy, grains, industrial, metals and oilseeds. As discussed in the data section, the majority of the traded volume are on the third and fourth nearest contracts, therefore we construct market and sector portfolios from the first to fourth nearest contracts (also see Figure 1). Panel A reports the results over the sample period 2004-2017, whereas Panel B extends the sample to 1992 (contract inceptions). The findings in Table 3 suggest that regardless of time, sector and maturity specifications, long-only investments in China do not generate statistically significant economic profits. Most notably, the grains and energy sectors saw significant losses on the first and second nearest contracts. These results can be better visualized in Figure 2.

Figure 2 illustrates the cumulative performance of the broad market and individual sectors. In addition to equally weighting portfolios, we also employ alternative weighting schemes taking into account the impacts of trading volume and open interest. Regardless of weighting schemes, these plots confirm the findings in Table 3, that in general, passive long- only investments in the broad market or sectors deliver poor economic returns. Specifically, energies are the worst performer followed by industrials, grains and oilseeds commodities, on average losing more than half of their values since 2004. Notably, oilseeds portfolio (equal- weighted) appears to have outperformed the rest of the market. This may be attributable to the speculation mania portrayed by the media in recent years, as the volume growth in the sector significantly outpaced other sectors (see Table 2). However, when more weights are assigned to oilseeds with higher volume and open interest, the outperformance deteriorates. This implies that for rapeseeds meal, soybean meal and soybean oil, the relationship between trading volume and returns are likely negative. In the meantime, metals, oilseeds and energies exhibit non- synchronized growth with the broad market from 2004 leading to the global financial crisis (GFC). Over the last 5 years, all sectors have declined by varying degrees, although it is less pronounced in grains. This decline may be explained by the excess supply and the slowed demand experienced by the Chinese economy since the GFC.

Consistent with the literature on the US market by Working (1949), Telser (1958) and Dusak (1973), our findings clearly suggest the rejection of the null hypothesis of non-zero premium for long-only investors. The theory of normal backwardation suggests that hedgers are net short and speculators are net long. Speculators receive risk premium as a compensation for taking on the price risk of hedgers. However, since there is no information available on the

25 type of traders, and given that more than 95% of participants are individual investors who may act as speculators or hedgers, we conjecture that the failure of long-only investments in the Chinese market may be partially due to speculators overpowering hedgers thus reducing the risk premium. Overall, these findings suggest that the theory of normal backwardation (Keynes, 1930; Hicks, 1939) does not hold well in China. We now move on to the performance of long- short strategies.

5.2 Long-short strategies

Table 4 reports the performance of 12 long-short strategies on first (Panel A), second (Panel B), third (Panel C) and forth (Panel D) nearest contracts. Following the literature, we first focus on the nearest contracts, as these contracts are often regarded as proxies for spot returns.

While passive long-only strategies do not deliver economic returns, five out of 12 long- short strategies examined yield statistically significant profits on the first nearest contract. Most notably, the cross-sectional momentum strategy returns an astonishing 21.94 percent per annum on average. This is followed by the term structure strategy which generates an average of 15.91 percent per annum. Time-series momentum, hedgers’ hedging pressure (HHP) and volatility strategies rank as the third, fourth and fifth most profitable, with an annualized return of 14.08 percent, 10.99 percent and 10.53 percent, respectively. The success of these strategies is consistent with the findings in the US market (Miffre & Rallis, 2007; Moskowitz et al., 2012). The salient returns by momentum strategies indicate that price persistency is strong in the Chinese market. Moreover, the profitability of term structure (Fuertes et al., 2010) and HHP strategy (Basu & Miffre, 2013) implies that the commodity futures in China behave according to the predictions of the Hedging Pressure Hypothesis (Cootner, 1960) and the Theory of Storage (Working, 1949). Furthermore, the success of volatility strategy implies commodities with higher risk as measured by coefficient of variation outperform those with lower risks in the Chinese market (Szymanowska et al., 2014).

Strikingly, we find that the skewness strategy generate statistically significant losses. Contrary to Fernandez-Perez et al. (2018), who document a sizeable alpha earned in the US markets, we find that commodities with higher skewness actually outperform commodities with lower skewness in China. The failure of skewness strategy suggests that skewness is not (yet) priced in the cross-section. This finding also fails to support the extension of the Theory of Storage (Deaton & Laroque, 1992), which implies that the link between commodity inventory and skewness is currently weak in China. We cannot rule out the possibility that

26 inventories might be distorted by tight regulatory controls though. Meanwhile, we find that illiquid commodities significantly underperform the more liquid ones (almost statistically significant at 10%). This implies that investors do not receive compensation for bearing illiquid risk in this market, but instead they pay a price to own illiquid commodities. We put forward two possible explanations. First, Amivest measure may not be the best proxy of liquidity in the Chinese market. Second, certain groups of market participants may be forced to hold illiquid commodities due to market frictions or limits-to-arbitrage. These findings suggest that commodity futures prices in China behave very differently to the US.

Strategies that do not report significant profits also contain unique information about the Chinese commodity futures market. First, although inspired by the same theoretical foundation, the SHP strategy fails to deliver a significant profit, with an annualised mean return of merely 4.43 percent – statistically indifferent from zero. Second, the negative and insignificant results on value strategy suggest that long-term losers in this market continue to lose. The lack of reversal may also provide insights about the remarkable momentum profit. Moreover, strategies sorted on open interest and macroeconomic betas all deliver statistically insignificant returns. The economic intuitions behind the inflation and foreign exchange (FX) strategies reside in the fact that commodity futures are positively related to unexpected inflation rate and negatively related to foreign exchange rate, as has been documented in both the US and Chinese markets (Gorton & Rouwenhorst, 2006; Mallick & Sousa, 2013; Erb & Harvey, 2016). Intuitively, commodities with a higher inflation beta/lower currency beta should earn higher returns relative to commodities with lower inflation beta/higher currency beta. However, this is not the case in the Chinese commodity future market.

The failure of inflation and FX strategies may be due to two reasons. First, the correlation between commodity futures and the two macroeconomic factors is not strong.31 Second, the inaccuracy of inflation and FX data could have distorted the strategy performance. The Chinese government has been criticized for frequently intervening in the RMB currency markets to maintain tight control over the value of the RMB. These intervention activities, which have been proven as considerable forces on both the onshore and offshore RMB market, may have artificially contaminated valuable trading signals embedded in the correlation between commodity prices and the RMB. Zhang and Pan (2004) use adjusted foreign reserves as a proxy for exchange rate intervention by the government to estimate the “actual” exchange

31 We estimate a correlation between AVG and RMB effective exchange rate index return as -0.3094 and significantly different from zero at 5% significance level.

27 rate and find the RMB would have appreciated by 15-22% against US dollar if there was no government intervention during 1996-2003.

Hong and Yogo (2012) find that open interests not only predict commodity returns but the real economy in the US. However, our results do not support this finding, implying that the predictive power of open interest is questionable in the Chinese market. The core assumption made by Hong and Yogo (2012) is that open interests will rise due to increasing hedging demand. However, given that the Chinese market is heavily dominated by individual investors, we argue that hedging demand has less impact on the trading activities, as most individual investors are speculation-motivated which leads to more frequent trading rather than holding positions for longer time. This evidence once again highlights the distinction between futures markets in China and the US.

Furthermore, it has been argued that distant contracts on the futures curve contain valuable information which influences investment strategy returns (Fuertes et al., 2010; de Groot, Karstanje, & Zhou, 2014). Motivated by these studies, we re-evaluate the performance of all strategies on the second, third, and fourth nearest contracts in Panels B through to D. The findings indicate that strategy performance generally deteriorates as investors move from the front towards further distant contracts. HHP and volatility strategies are no longer statistically significant on all distant contracts. Although the cross-sectional momentum strategy maintains its state as the most profitable strategy across all distant contracts, its profitability has declined significantly as we move from the front contract towards the more distant contracts. This decline in profits offers crucial clues in explaining the extraordinary profits of momentum strategies in the front contracts. As stipulated in the trading rules across three exchanges, strict position limits apply in the nearest to maturity contracts. Consequently, these position limits create a regulation induced limits-to-arbitrage. As speculators cannot trade sufficient volumes in the front contracts, they are forced to move to more distinct contracts in order to gain continuous exposures. As a result, the profits to momentum strategies decline as the market corrects the mispricing in the second nearest contracts and beyond.

Overall, findings presented in Table 4 suggest that momentum and term structure strategies consistently deliver statistically significant but weaker profits along the futures curve. This return erosion impacts the momentum strategy more than the term structure strategy. Given the presence of limits-to-arbitrage in the front contracts, this difference implies that momentum is the most prominent strategy in Chinese commodity futures market.

28 Figure 3 illustrates the cumulative returns of one RMB invested in term structure, HHP, cross-sectional and time-series momentum and volatility strategies, benchmarked against the passive long-only portfolio (AVG). Momentum strategy finishes strongly with a terminal value of 11.89 RMB versus 6.50 and 4.88 for term structure and time-series momentum, respectively. We focus on the nearest contracts here as strategies perform the best on these contracts. Consistent with the US market and previous studies on Chinese markets (Li et al., 2017), Figure 3 shows that cross-sectional momentum suffered considerable losses following the GFC. However, the resilience observed post-2010 is in sharp contrasts with Bianchi et al. (2016), in which they show a clear declining trend in the cumulative return of various momentum strategies in the US.

To investigate the stability of strategy performance over time, Figure 4 illustrates the recursive performance of term structure, momentum, HHP and volatility strategies. Panel A shows the returns, Panel B reports standard deviations and Panel C plots the Sharpe ratios. The recursive statistics are computed on a monthly expanding window with an initial window size of 12 months. Overall, plots in Figure 4 suggest that the risk-adjusted performance are fairly stable over time, although less profitable over time. As can be seen from Panels A and B, both recursive mean return and standard deviation have declined from a historical peak. Inspired by the Adaptive Market Hypothesis (Lo, 2004), Bianchi et al. (2016) argue that the gradual profitability erosion that has occurred in recent years, as observed in various types of momentum strategies, is a result of the learning and competition pressures driven by increasing numbers of market participants, whose ultimate goal is to survive in a constantly changing and evolving market. Therefore, we conjecture that the reduced profitability in China may be due to increasing market participation, since the market experienced explosive growth in the past decade. We now proceed to examine the diversification potential of the long-short strategies.

5.3 Potential for diversification

We first compute the Pearson correlation among long-short strategies and then examine the level of co-movements between traditional assets and our strategy returns.

29 Table 5 reports the returns correlations among all strategies tested. Panel A reports results on the nearest contract and Panel B reports the third nearest contract.32 The results in Panel A reveal that among the five profitable strategies, HHP and volatility are not correlated with the other four strategies. Term structure, cross-sectional, and time-series momentum are positively correlated with each other, as has been confirmed by Moskowitz et al. (2012) and Fernandez- Perez et al. (2018). The former study acknowledges the distinction between cross-sectional and time-series momentum strategy by deconstructing their returns and conclude that their high correlation is driven by the positive auto-covariance in futures returns. The latter argues that the interaction between term structure and cross-sectional momentum strategy is due to the two strategies all implicitly long the backwardated and short the contangoed commodities. These findings seem to hold true in the Chinese market. However, the magnitude of correlation between term structure and cross-sectional momentum strategy suggests the drivers of their returns remain different.33

Furthermore, despite the lack of statistical significance, the unprofitable strategies exhibit significant correlations with each other, and with the profitable strategies. For example, value strategy is negatively correlated with both cross-sectional and time-series momentum strategies. This is consistent with Asness et al. (2013), in which they argue that the two strategies’ opposite exposures to liquidity risk may be the primary driver of the negative correlation. This finding is particularly interesting in China, because while momentum strategies deliver persistent economic returns, value strategy does not seem to work at all in these markets. This negative correlation further confirms the lack of reversals in China. In addition, although the FX strategy does not produce significant profit, it reports significant correlations with term structure, HHP, and volatility strategies, implying that the value of RMB may play an important role in the return dynamics of these strategies. Similarly, the liquidity strategy reveals a negative correlation of -0.4179 with SHP, suggesting that more liquid commodities somewhat overlap with highly speculated commodities. Finally, the skewness strategy is not correlated with any strategies, confirming the findings that skewness is not presently priced in the Chinese commodity futures market. When strategies are implemented on the third nearest contract, Panel B reports consistent results.

32 We choose the third nearest contracts performance as comparison against front contracts, since on average, these contracts report the highest trading volumes and open interests. 33 An average correlation of 0.3156 between cross-sectional momentum and term structure strategies has been estimated by Fuertes et al. (2010), which is slightly lower than our estimate of 0.4673.

30 We now proceed to the discussion on whether long-only and long-short strategies can serve as effective tools for risk diversification and inflation hedging in the Chinese market. Figure 5 illustrates the pairwise correlations between strategy returns and equity, returns and changes in unexpected inflation.34 First, correlations between the commodity market and the stock market is a significant 42.2%. Although Basu and Miffre (2013) also estimate a positive correlation between long-only commodity portfolios and the S&P 500 Index, the magnitude is much lower than that in the Chinese market. Moreover, 5.2% and -6.4% correlations are documented between the commodity market and unexpected inflation and the bond market respectively, yet neither of them is significantly different from zero. This clearly contradicts the notion that commodity futures can diversify traditional assets returns and hedge unexpected inflation (Erb & Harvey, 2006; Gorton & Rouwenhorst, 2006). It may be premature to conclude that the commodity futures in China fails to hedge inflation at the current stage, as Gorton and Rouwenhorst (2006) argued that correlation estimates are subject to rebalancing frequency and the diversification effect of commodity futures increases at longer horizons. However, it is clearly evident that passive long-only investments in broad commodity futures market presents a poor candidate for diversifying negative shocks in the stock market in China.

For most of the long-short strategies, a zero correlation with the macro and financial variables cannot be rejected at 5% level, highlighting their diversification advantage over long- only investments. There are a few exceptions. Term structure strategy shows a statistically significant correlation of 26.4% with the stock market, which is considerably lower than that between long-only commodity portfolio and stock market. Intriguingly, HHP strategy reveals a significantly negative correlation with CSI 300, which makes HHP strategy the best candidate to diversify movement in Chinese stocks. Overall, the superior diversification feature of long- short strategies in China is consistent with those in the US, but the HHP strategy exhibits superior diversification potential (Basu & Miffre, 2013). Overall, findings presented in Figure 5 suggest that long-short strategies present excellent candidate for hedging against movements of traditional assets in China. However, it is not the case for unexpected inflation.

Since correlations can be time-dependent, we compute dynamic correlations to address the concerns that unconditional correlations may lead to false conclusions about the

34 The unexpected inflation is the difference between actual and forecasted inflation estimated by Bloomberg. Equities are proxied by the CSI 300 index which consists of top 300 stocks traded on the Shanghai and Shenzhen stock exchanges. Bond returns are measured by the Barclays China Aggregate Index which covers fixed-rate treasury, government and corporate bonds. AVG denotes the equally-weighted portfolio of 30 commodities. We find consistent results by employing the Shanghai Stock Exchange Composite Index, the Shenzhen Stock Exchange Composite Index and the MSCI All China Index.

31 diversification benefits. Figure 6 exhibits the time-varying correlation between the five successful strategies and the Chinese stock market. The correlations are estimated using the asymmetric dynamic conditional correlation generalized autoregressive conditional heteroskedasticity (ADCC-GARCH) model developed by Cappiello, Engle, and Sheppard (2006). Consistent with previous discussions, HHP strategy exhibits relatively lower and negative correlations with the stock market compared to the other four strategies. Moreover, the correlation between cross-sectional momentum and stock market appears to be the most volatile throughout our studying period. Consistent with the previous literature, we find that correlations are generally higher in periods of market stress in China.35

To add another dynamic to the diversification potential, Figure 7 illustrates the average monthly return of the broad commodity market and long-short strategies during various market conditions. To do that, we rank all months into quintiles using the CSI 300 index returns. Plots in Figure 7 suggest that on average, most strategies appear to perform better (worse) during stock market growth (crisis) periods. Confirming with correlation results, the performance of the broad market is in perfect sync with the CSI300 from crisis to growth periods. In the meantime, the HHP strategy reveals the reversed pattern, highlighting its outstanding diversification benefits. Moreover, the cross-sectional momentum strategy delivers the strongest performance when the market condition is neutral, suggesting that commodities momentum in China is sensitive to market states.

5.4 Robustness checks

To gain further confidence in our results, we conduct an extensive suite of robustness tests including re-estimating strategy returns in the most liquid markets and randomly selected commodity sectors, by varying portfolio breakpoints and taking into consideration the transactions costs and the possibility of data-snooping.

To examine whether illiquidity impacts strategy performance, we construct an alternative sample by excluding the 10 least liquid commodities, sorted on the average trading volume over the sample period. Table 6 reports these results on the first nearest contracts (Panel A) and the third nearest contracts (Panel B). Since the liquidity strategy failed to deliver statistically significant returns (i.e. illiquid commodities should outperform more liquid commodities), we expect strategies to perform better or at least unaffected when implemented

35 In addition to the bear markets globally in 2008-2009 and 2011-2012, the Shanghai stock market fell 30 percent in July 2015. The index fell again on 24 August 2015 (“Black Monday”) by 8.48 percent. In January 2016, the market experienced another steep sell-off of 18 percent, and trading was halted on 4 January and 7 January 2016.

32 on more liquid contracts. As a result in Panel A, while the unsuccessful strategies remain insignificant, successful strategies (with the exception of volatility) indeed experience improvements in risk-adjusted returns. The failure of volatility strategy implies that more illiquid commodities may also be highly volatile instruments.

Furthermore, when strategies are implemented on the third nearest contracts, Panel B indicates that HHP and volatility strategies are no longer significant, and the performance of strategies deteriorates. While this is consistent with the unmodified sample, the economic profits of momentum strategy remain strong both quantitatively and statistically. This implies that on the more distant end of the futures curve, where relatively more volumes are traded, liquidity risk matters less to momentum strategy than to term structure strategy. This further suggests that momentum strategies returns are robust in different liquidity environments and on various part of the futures curve. Overall, the findings presented in Table 6 suggest that holding relatively illiquid commodities is not rewarded in China.

To examine whether the strategy performances are driven by specific a sector, we re- evaluate the performance of strategies by excluding an entire sector of commodities at a time. Table 7 reports the results with industrials (Panel A), metals (Panel B), grains (Panel C), oilseeds (Panel D) and energies (Panel E) excluded, respectively. While most of the unsuccessful strategies remain insignificant, several interesting dynamics emerge. First, volatility strategy fails to deliver significant return when grains or oilseeds are excluded from the sample. This suggests that the performance of volatility strategy is largely driven by grains or oilseeds commodities. Second, the HHP strategy also appears to be sensitive to sector specifications as profits range from 8.6% to 14.6%, though remain statistically significant. Third, term structure, cross-sectional and time-series momentum strategies appear not affected by sector specifications. Notably, cross-sectional momentum strategy reports extreme outperformance when grains are excluded from the sample. This implies that exposure to grains commodities may help shed light on the difference between momentum and volatility strategies. This also suggests that grains commodities are less prone to price continuation. Overall, the findings presented in Table 7 suggest that among all successful strategies, term structure and momentum profits are not due to concentrated allocations to a specific commodity sector. Besides, our findings suggest that hedging pressure and volatility risks are more sector-specific. These findings are better illustrated in Figure 8, which depicts the percentage of total trades each strategy assigns to every sector.

33 To examine whether the profitability is due to luck or the merits embedded in the methods, we conduct data-snooping test using the White (2000) Reality Check (RC) and Hansen (2005) Superior Predictive Ability (SPA) test. The null hypothesis is that the average performance of the benchmark is as small as the minimum average performance across the strategies tested, while the alternative hypothesis is that the minimum average loss across strategies is smaller than the average benchmark performance. Appendix A1 reports the RC and SPA test results. We perform seven groups of tests. Panel A demonstrates the results when running tests on the five strategies as a group against a zero-mean benchmark, while Panel B lists the results when the benchmark is replaced by our AVG factor. Further, we also test individual strategy against AVG. As a test of robustness, we consider a bootstrap block length of 2, 10 and 20 months. Within each block length, stationary and circular bootstraps are performed based on 10,000 replications. Overall, our results consistently suggest the rejection of the null hypothesis, implying the successes of five strategies are not a result of data mining.

To address the concerns that strategy profits are a compensation for high transactions costs, we compute the return net of transaction cost for the five profitable strategies in Table 8. Following Fuertes et al. (2010), we first compute the portfolio turnover for all quartile portfolios and long-short portfolios. We then apply a rather aggressive round trip transaction cost of 0.086% estimated by Marshall et al. (2012) to achieve the net return. It is evident that portfolio turnovers vary across different trading strategies, with the highest average turnover occurring in HHP strategy and the lowest average turnover being observed in time-series momentum. Interestingly, portfolio turnovers for momentum strategies appear considerably lower compared to the US market estimated by Fuertes et al. (2010). This implies that winner (loser) commodities remain as winners (losers) even over long periods of time. This helps to shed light on the failure of value strategies and the remarkable profitability of momentum strategies in China. Overall, the profits to long-short strategies are clearly too large to be subsumed by transaction costs. We now move on to investigate the sources of long-short strategies returns.

6 Strategy Returns Analysis

6.1 Decomposing long-short portfolios

34 Findings in the previous section have shown that long-short strategies are superior to long-only investing in capturing risk premiums in China. However, to what extent each of the long and short components contributes to this result? Table 8 reports the performance of quartile portfolios for term structure, hedgers’ hedging pressure, momentum, volatility and time-series momentum strategies. Since these strategies are proven to generate statistically significant profits, we focus on understanding the sources of returns for these strategies on the first and the third nearest contracts. The results in Table 8 present two key findings.

First, when strategies are implemented on the nearest contracts, we find that both long and short portfolios of term structure and momentum strategies contribute to the strategy returns. This suggest that the profits are not short sided, although the significance in short portfolios appears to be slightly stronger than that in the long portfolios. The findings on momentum is not consistent with Miffre and Rallis (2007), in which they show that momentum profits are dominated by short portfolios. In the meantime, the findings on term structure is consistent with Fuertes et al. (2010), where they document an even performance in the long and short portfolios. Although taking short positions in futures markets is as straight-forward as taking long positions, the regulatory environment in China may impede speculator’s ability to take short positions. This poses a challenge for investors if their strategies derive profits mainly from the short side.

Second, we find that profits to hedgers’ hedging pressure are almost entirely attributable to the short portfolio. This suggest that commodities with the highest hedging pressure significantly underperforms commodities with lower hedging pressure, albeit returns from other quartile portfolios are statistically indifferent from zero. This also can be explained from the perspective of volatility, as Bohl et al. (2018) state that the higher the hedging activity ratio, the lower the volatility. Similar to HHP, profits to volatility strategy are largely sourced from the short portfolio, suggesting that commodities with the lowest volatility risk significantly underperform. These findings are consistent when strategies are implemented on the third nearest contracts, although volatility and HHP lose significance and the profitability generally deteriorates. Overall, findings presented in Table 8 suggest that long positions are equally important as short positions, whereby highlighting the importance of long-short strategies in capturing risk premiums in commodity futures in China.

6.2 Commodity-specified risks

35 To examine whether the profitability of the long-short strategies can be explained by risk taking, we first employ the commodity specific risk factor model. Table 9 reports the regression results of the long (Panel A), short (Panel B), and long-short (Panel C) portfolios in a three-factor model proposed by Bakshi et al. (2017). Bakshi et al. (2017) demonstrate that AVG, CARRY and MOM factors are sufficient to describe the cross-sectional variation of commodity futures returns in the US. AVG represents the broad commodity market risk and CARRY is constructed by taking long positions in the five most backwardated commodities and short positions in the five most contangoed commodities. Identical to our momentum strategy, MOM is constructed by taking long positions in past 12-month winner commodities and short positions in past loser commodities. We report results on the first and third nearest contracts.

For long portfolios, we find that profits to time-series momentum can be well explained by the broad market risk and the momentum factor with a high R2 and insignificant intercept. Similarly, the three-factor model can explain the dynamics of long portfolios to the other strategies reasonably well, albeit that alphas remain statistically significant. Meanwhile, the loadings on AVG are uniformly positive and statistically significant, and ranges from 0.86 to 1.25. This suggests that the broad commodity market performance imposes considerable impacts on long portfolios. For short portfolios, it appears that term structure, time-series momentum and volatility strategies can be adequately explained by the three-factor model. Notably, the long and short portfolios of term structure and time-series momentum strategy respond to the MOM factor oppositely. This finding is inconsistent with Bakshi et al. (2017), who find that the long portfolio to carry strategy is not related to the momentum factor.

For the long-short portfolios, we first show that returns to term structure strategy is loading positively on MOM and AVG, but the intercepts remain significant. This suggest that term structure profits cannot be fully explained by bearing systematic risk such as the broad market and momentum factors. When momentum profits are regressed on CARRY and AVG, we find momentum loads positively on CARRY but not AVG, and the intercepts are larger and remain significant. This suggest that momentum is stronger compared to term structure, and a double-sort strategy will likely improve the risk-adjusted performance of single-sorted momentum and term structure strategies. Consistent with the results in Panels A and B, the time-series momentum strategy load positively on all factors with an R2 of 66.9%, although only MOM remains significant on the third nearest contract. This suggest that time-series momentum are compensations for bearing AVG, CARRY and MOM factor risks.

36 Furthermore, we find that the Bakshi et al. (2017) three-factor model cannot adequately explain the dynamics of hedger’s hedging pressure (HHP) and the volatility strategies on the nearest contracts. Interesting, while the HHP loads negatively on AVG and the volatility strategy loads positively on the AVG, neither strategies are related to CARRY or MOM. However, these loadings are not consistent on the third nearest contracts, suggesting that the return dynamics may be largely different between front and distant contracts. Overall, the Bakshi et al. (2017) three-factor model cannot fully explain the success of long-short strategies in the Chinese market. Our findings are supported by He et al. (2017), in which they demonstrate that a three-factor model are not able to fully explain the cross-sectional return variations in the Chinese market.36

6.3 Standard risk adjustments

Following Miffre and Rallis (2007), Fuertes et al. (2010) and Bianchi et al. (2016), we conduct additional risk adjustments by employing a six-factor model as shown in Table 10 Panels A and B report results on long and short portfolios, whereas Panel C reports the long-short portfolios. We augment the model to reflect the market dynamics in China by employing the CSI 300 as stock market factor (STOCK), Barclays China Aggregate index as a proxy for the bond market returns (BOND), the sample of commodities in this study as the representative of commodity market factor (AVG), as well as the China unexpected inflation rate (INF), RMB effective exchange rate index return (FX), and unexpected industrial production (UIP). An insignificant alpha, strong factor loadings and a high R2 suggest that the returns of trading strategies are rewards for bearing these risk factors. Again, we report results on both first and third nearest contracts.

Consistent across all strategies, the returns of long and short portfolios reveal positive and statistically significant loadings on AVG. Nevertheless, it appears that profits to long and short portfolios are not exposed to aggregate risks associated with stocks and bonds. Interestingly, we find that HHP, momentum and volatility strategies respond to shocks to industrial production in China. Although the dynamics of long and short portfolios can be reasonably described by financial and macro risks, these risks are poor at explaining the returns dynamics of long-short portfolios. While the AVG remains significant across the board, the intercepts remain largely significant and the R2s are generally low. Overall, the findings

36 He et al. (2017) employ a three-factor model similar to our study but replace the MOM factor with a time-series momentum factor (TSMOM). For robustness, they also test a three-factor model with the MOM factor, but do not attain any improvements in cross-sectional pricing.

37 presented in Table 10 suggest that the profitability of long-short strategies in the China cannot be attributed to broad markets or none-tradable macroeconomic risks.

6.4 Liquidity, behavioral and sentiment factors

Previous studies have demonstrated that liquidity risk explains the momentum profits across asset classes (Sadka, 2006; Asness et al., 2013; Daskalaki, Kostakis, & Skiadopoulos, 2014). Recently, Bianchi et al. (2016) document that momentum can be largely explained by the anchoring behaviour of investors (proxied by the 52-week high momentum). Motivated by these findings, we further employ the Amihud Illiquidity factor (AI) and the 52-week high momentum factor (HMOM) in search for an alternative source of observed profits.

Table 11 reports the regression results. The results in Panel A suggests that the AI factor alone is unable to explain any strategies, as most of the coefficients are indifferent from zero and the R2s are extremely low. However, we find that long portfolios of term structure, momentum and time-series momentum appear to be negatively related to illiquidity risk. This is not the case for the short portfolios. Nevertheless, we find that the long-short portfolios consistently exhibit negative exposures (by a varying degree) to liquidity risk. This striking result implies that a lower liquidity (or higher illiquidity) signals lower, as oppose to higher, expected returns of long-short strategies. This is in sharp contrasts with Asness et al. (2013), in which they document a positive relationship between momentum and liquidity risk. Nonetheless, the low R2s across the board suggest that liquidity cannot explain the return dynamics of long-short strategies returns.

Panel B of Table 11 reports the results on the behavioural factor. Consistent with Bianchi et al. (2016), we find that momentum and time-series momentum strategies are indeed related to the 52-week high momentum factor, with relatively high R2s. Since the intercepts remain large and significant, this suggests that anchoring bias provides at least a partial explanation for the remarkable profits generated by momentum strategies. This behaviour may be amplified in the Chinese market as market participants are primarily individuals who are more likely affected by behavioural biases, as discussed in Li et al. (2017). Interestingly, it appears that only the short portfolios are exposed to anchoring not the long portfolios. The profits to short portfolios can be completely subsumed by the 52-week high momentum. Furthermore, we find that the HHP strategy is also weakly related to anchoring behaviour, whereas term structure and volatility strategies are not exposed to anchoring bias. Overall, the findings presented in Table 11 suggest that illiquidity and behavioural factors provide at least

38 a partial explanation for the profitability of long-short strategies in the Chinese futures market. Strikingly, we find that higher liquidity (or lower illiquidity) leads to higher expected returns of long-short strategies, highlighting once again, the striking differences between the commodity markets in China and the US.

Finally, in addition to liquidity and behavioural factors, market sentiment has been found to be a significant component that influences the broad financial markets. Studies have documented the connections between sentiments and stock returns (Baker & Wurgler, 2007), and commodity futures returns (Gao & Süss, 2015; Bianchi et al., 2016). Motivated by these findings, we employ the CBOE Crude Oil Volatility Index (OVX) and AlphaShares Chinese Volatility Index (CHIX) as proxies for market sentiment. The OVX measures the market expectation of 30-day volatility of crude oil prices. CHIX measures the implied volatility of options on the FTSE Xinhua China 25 and Hang Seng indices.

Table 12 reports the regression results of long, short and long-short portfolios on OVX (Panel A) and CHIX (Panel B). Findings presented in Panel A reveal that, even though crude oil contract is not present in the sample, we find that long and short portfolios reveal statistically significant negative loadings on the OVX factor. Consistent with Bianchi et al. (2016), this findings imply that long and short portfolios respond to volatility risk symmetrically. However, the alphas of long-short portfolios remain large and statistically significant, suggesting that strategy profits cannot be subsumed by shifts in commodity markets sentiment. Furthermore, we find consistent results by employing CHIX in Panel B. With low R2s and significant alphas across the broad, our findings suggest that profits to long-short strategies cannot be explained by changes in stock market sentiment. Overall, findings presented in Table 12 suggest that profits to long-short strategies are not sourced from changes in market sentiments.

7 Conclusion

In conclusion, this paper extensively examined the performance of long-only and long-short investment strategies in the Chinese commodity futures markets. Our results suggest that passive long-only investments deliver poor economic returns and they do not appear to provide a hedge against inflation and movements in traditional assets. However, we found that momentum and term structure strategies generate persistent economic profits along the futures curve, in illiquid markets and randomly selected commodity sectors. We demonstrated that

39 long-short strategies that exploit past returns and hedging pressure make excellent candidates for hedging against movements in traditional assets in China.

The successes of these strategies could not be attributed to aggregate market risks, none-tradable macroeconomic risks, commodity specific risks, changes in market sentiment, transactions costs and data-snooping. However, we showed that illiquidity and behavioural bias provide at least a partial explanation for the profitability of long-short strategies in China. Furthermore, reflective of unique institutional settings, we argue that the observed profits are partially due to regulation induced limits-to-arbitrage. Our findings also highlighted the distinctiveness of the Chinese futures market compared to the US. Finally, we urge the CSCR and relevant authorities to reclassify trader types in accordance with their trading purposes and establish a CFTC-type repository for positions data on different trader types. Such data are essential to assess the effectiveness of the risk transfers in these markets.

References

Acharya, V. V., & Pedersen, L. H. (2005). Asset pricing with liquidity risk. Journal of Financial Economics, 77(2), 375-410. Amihud, Y. (2002). Illiquidity and stock returns: cross-section and time-series effects. Journal of Financial Markets, 5(1), 31-56. Asness, C. S., Moskowitz, T. J., & Pedersen, L. H. (2013). Value and momentum everywhere. The Journal of Finance, 68(3), 929-985. Baker, M., & Wurgler, J. (2007). Investor sentiment in the stock market. The Journal of Economic Perspectives, 21(2), 129-151. Bakshi, G., Gao, X., & Rossi, A. G. (2017). Understanding the sources of risk underlying the cross-section of commodity returns. Mangement Science, forthcoming. Basu, D., & Miffre, J. (2013). Capturing the risk premium of commodity futures: The role of hedging pressure. Journal of Banking & Finance, 37(7), 2652-2664. Bessembinder, H., & Seguin, P. J. (1993). Price volatility, trading volume, and market depth: Evidence from futures markets. Journal of financial and Quantitative Analysis, 28(1), 21-39. Bhardwaj, G., Gorton, G. B., & Rouwenhorst, K. G. (2016). Investor interest and the returns to commodity investing. The Journal of Portfolio Management, 42(3), 44-55. Bianchi, R. J., Drew, M. E., & Fan, J. H. (2015). Combining momentum with reversal in commodity futures. Journal of Banking & Finance, 59, 423-444. Bianchi, R. J., Drew, M. E., & Fan, J. H. (2016). Commodities momentum: A behavioral perspective. Journal of Banking & Finance, 72, 133-150. Bloomberg. (2017a). China's commodity beast enters hibernation. Retrieved on 10 December 2017 from https://www.bloomberg.com/gadfly/articles/2017-11-15/china-s-sated- commodity-beast-is-going-into-hibernation

40 Bloomberg. (2017b). China plans to open up commodity trading to foreigners to bring down prices. Retrieved on 20 June 2017 from https://www.bloomberg.com/news/articles/2017-05-18/global-pricing-power-the- target-for-china-s-big-derivatives-push Bloomberg. (2018). China sets new records for gobbling up the world’s commodities. Retrieved on 18 January 2018 from https://www.bloomberg.com/news/articles/2018- 01-12/world-s-commodity-engine-roars-to-another-record-with-xi-at-helm Bodie, Z., & Rosansky, V. I. (1980). Risk and return in commodity futures. Financial Analysts Journal, 36(3), 27-39. Bohl, M. T., Siklos, P. L., & Wellenreuther, C. (2018). Speculative activity and returns volatility of Chinese agricultural commodity futures. Journal of Asian Economics, 54(1), 69-91. Cappiello, L., Engle, R. F., & Sheppard, K. (2006). Asymmetric Dynamics in the Correlations of Global Equity and Bond Returns. Journal of Financial Econometrics, 4(4), 537-572. Chan, K. C., Fung, H.-G., & Leung, W. K. (2004). Daily volatility behavior in Chinese futures markets. Journal of International Financial Markets, Institutions and Money, 14(5), 491-505. China Futures Association. (2012). China Futures Markets Annual Report. Cootner, P. H. (1960). Returns to speculators: Telser versus Keynes. Journal of Political Economy, 68(4), 396-404. Daskalaki, C., Kostakis, A., & Skiadopoulos, G. (2014). Are there common factors in individual commodity futures returns? Journal of Banking & Finance, 40(1), 346-363. de Groot, W., Karstanje, D., & Zhou, W. (2014). Exploiting commodity momentum along the futures curves. Journal of Banking & Finance, 48(1), 79-93. De Roon, F. A., Nijman, T. E., & Veld, C. (2000). Hedging pressure effects in futures markets. The Journal of Finance, 55(3), 1437-1456. Deaton, A., & Laroque, G. (1992). On the behaviour of commodity prices. The Review of Economic Studies, 59(1), 1-23. Dewally, M., Ederington, L. H., & Fernando, C. S. (2013). Determinants of trader profits in commodity futures markets. The Review of Financial Studies, 26(10), 2648-2683. Dhume, D. (2011). Essays on measuring and explaining commodities returns. PhD thesis, Harvard University, Dusak, K. (1973). Futures trading and investor returns: An investigation of commodity market risk premiums. Journal of Political Economy, 81(6), 1387-1406. Erb, C. B., & Harvey, C. R. (2006). The strategic and tactical value of commodity futures. Financial Analysts Journal, 62(2), 69-97. Erb, C. B., & Harvey, C. R. (2016). Conquering misperceptions about commodity futures investing. Financial Analysts Journal, 72(4), 26-35. Fama, E. F., & French, K. R. (1987). Commodity futures prices: Some evidence on forecast power, premiums, and the theory of storage. Journal of Business, 60(1), 55-73. Fernandez-Perez, A., Frijns, B., Fuertes, A.-M., & Miffre, J. (2018). The skewness of commodity futures returns. Journal of Banking & Finance, 86, 143-158. Fernandez-Perez, A., Fuertes, A.-M., & Miffre, J. (2017). Harvesting Commodity Styles: An Integrated Framework. Working paper. Auckland University of Technology. FT. (2016). Speculators march into China commodities. Retrieved on 20 June 2017 from https://www.ft.com/content/592c8af4-0b94-11e6-b0f1-61f222853ff3

41 FT. (2017). China commodities slip as copper sell-off continues. Retrieved on December 15 2017 from https://www.ft.com/content/c31602a4-aabf-3eb2-9b97-80bc8811974b Fuertes, A.-M., Miffre, J., & Rallis, G. (2010). Tactical allocation in commodity futures markets: Combining momentum and term structure signals. Journal of Banking & Finance, 34(10), 2530-2548. Fung, H.-G., Leung, W. K., & Xu, X. E. (2003). Information flows between the US and China commodity futures trading. Review of Quantitative Finance and Accounting, 21(3), 267-285. Gao, L., & Süss, S. (2015). Market sentiment in commodity futures returns. Journal of Empirical Finance, 33(1), 84-103. Garcia, P., Leuthold, R. M., & Zapata, H. (1986). Lead‐lag relationships between trading volume and price variability: New evidence. Journal of Futures Markets, 6(1), 1-10. Gorton, G., & Rouwenhorst, K. G. (2006). Facts and fantasies about commodity futures. Financial Analysts Journal, 62(2), 47-68. Gorton, G. B., Hayashi, F., & Rouwenhorst, K. G. (2013). The fundamentals of commodity futures returns. Review of Finance, 17(1), 35-105. Hansen, P. R. (2005). A test for superior predictive ability. Journal of Business & Economic Statistics, 23(4), 365-380. He, C., Jiang, C., & Molyboga, M. (2017). Risk premia in Chinese commodity markets. Working paper. University of International Business and Economics. Hicks, J. R. (1939). Value and capital. Oxford University Press. Hirshleifer, D. (1990). Hedging pressure and futures price movements in a general equilibrium model. Econometrica: Journal of the Econometric Society, 58(2), 411-428. Hong, H., & Yogo, M. (2012). What does futures market interest tell us about the macroeconomy and asset prices? Journal of Financial Economics, 105(3), 473-490. Kaldor, N. (1939). Speculation and economic stability. The Review of Economic Studies, 7(1), 1-27. Kang, J., & Kwon, K. Y. (2017). Momentum in International Commodity Futures Markets. Journal of Futures Markets, 37(8), 803-835. Keynes, J. M. (1930). A Treatise on Money. London. Macmillan. Leuthold, R. M. (1983). Commercial use and speculative measures of the livestock commodity futures markets. Journal of Futures Markets, 3(2), 113-135. Li, B., Zhang, D., & Zhou, Y. (2017). Do trend following strategies work in Chinese futures markets? Journal of Futures Markets, forthcoming. 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. Lo, A. W. (2004). The adaptive markets hypothesis. The Journal of Portfolio Management, 30(5), 15-29. Lucia, J. J., & Pardo, A. (2010). On measuring speculative and hedging activities in futures markets from volume and open interest data. Applied Economics, 42(12), 1549-1557. Mallick, S. K., & Sousa, R. M. (2013). Commodity prices, inflationary pressures, and monetary policy: evidence from BRICS economies. Open Economies Review, 24(4), 677-694. Marshall, B. R., Nguyen, N. H., & Visaltanachoti, N. (2012). Commodity liquidity measurement and transaction costs. The Review of Financial Studies, 25(2), 599-638. Miffre, J., & Rallis, G. (2007). Momentum strategies in commodity futures markets. Journal of Banking & Finance, 31(6), 1863-1886.

42 Møllgaard, H. P. (1997). A squeezer round the corner? self-regulation and forward markets. The Economic Journal, 107(440), 104-112. Moskowitz, T. J., Ooi, Y. H., & Pedersen, L. H. (2012). Time series momentum. Journal of Financial Economics, 104(2), 228-250. Newey, W. K., & West, K. D. (1987). Hypothesis testing with efficient method of moments estimation. International Economic Review, 28(3), 777-787. Ng, L., & Wu, F. (2007). The trading behavior of institutions and individuals in Chinese equity markets. Journal of Banking & Finance, 31(9), 2695-2710. Pástor, Ľ., & Stambaugh, R. F. (2003). Liquidity risk and expected stock returns. Journal of Political Economy, 111(3), 642-685. Politis, D. N., & Romano, J. P. (1992). A circular block-resampling procedure for stationary data. Exploring the limits of bootstrap, 263-270. Politis, D. N., & Romano, J. P. (1994). The stationary bootstrap. Journal of the American Statistical association, 89(428), 1303-1313. Reuters. (2017). China to promote wealth management investment in commodities – regulator. Retrieved on 15 July 2017 from https://www.reuters.com/article/us-china- commodities/china-to-promote-wealth-management-investment-in-commodities- regulator-idUSKBN19A0M7 Robles, M., Torero, M., & Von Braun, J. (2009). When speculation matters. International Food Policy Research Institute Washington, DC. Rutledge, D. J. (1979). Trading volume and price variability: New evidence on the price effects of speculation. Futures markets: Their establishment and performance, 137-156. Sadka, R. (2006). Momentum and post-earnings-announcement drift anomalies: The role of liquidity risk. Journal of Financial Economics, 80(2), 309-349. Streeter, D. H., & Tomek, W. G. (1992). Variability in soybean futures prices: an integrated framework. Journal of Futures Markets, 12(6), 705-728. Symeonidis, L., Prokopczuk, M., Brooks, C., & Lazar, E. (2012). Futures basis, inventory and commodity price volatility: An empirical analysis. Economic Modelling, 29(6), 2651- 2663. Szymanowska, M., de Roon, F. A., Nijman, T. E., & van den Goorbergh, R. W. J. (2014). An anatomy of commodity futures risk premia. Journal of Finance, 69(1), 453-482. Telser, L. G. (1958). Futures trading and the storage of cotton and wheat. Journal of Political Economy, 66(3), 233-255. Tu, Z., Song, M., & Zhang, L. (2013). Emerging Impact of Chinese Commodity Futures Market on Domestic and Global Economy. China & World Economy, 21(6), 79-99. Weiner, R. J. (2002). Sheep in wolves’ clothing? Speculators and price volatility in petroleum futures. Quarterly Review of Economics and Finance, 42(2), 391-400. White, H. (2000). A reality check for data snooping. Econometrica, 68(5), 1097-1126. Williams, J., Peck, A., Park, A., & Rozelle, S. (1998). The emergence of a futures market: Mungbeans on the China Zhengzhou Commodity Exchange. The Journal of Futures Markets, 18(4), 427. Working, H. (1949). The theory of price of storage. The American Economic Review, 39(6), 1254-1262. WSJ. (2016a). Chinese traders roil commodity markets. Retrieved on 20 August 2017 from https://www.wsj.com/articles/chinese-traders-roil-commodity-markets-1470698606

43 WSJ. (2016b). Agricultural Commodities Egged On by China’s Futures Frenzy. Retrieved on 20 June 2017 from https://www.wsj.com/articles/agricultural-commodities-egged-on- by-chinas-futures-frenzy-1462172722 Xin, Y., Chen, G., & Firth, M. (2006). The efficiency of the Chinese commodity futures markets： development and empirical evidence. China & World Economy, 14(2), 79- 92. Zhang, F., & Pan, Z. (2004). Determination of China's long-run nominal exchange rate and official intervention. China Economic Review, 15(3), 360-365. Zhao, L. (2015). The development and challenges of China's futures markets. The Chinese Economy, 48(4), 279.

44 Table 1 Contracts specification This table reports symbols, contract size, delivery month, price limit, margin requirements on our sample of 30 commodities traded on DCE, SHFE and ZCE. For each commodity, the table outlines the first and last contract maturity and their corresponding dates in the sample. * indicates that gold delivery months: monthly contract of the recent 3 consecutive months and consecutive even months contracts within the recent 13 months. ** indicate that fuel oil delivery months are all excluding the month for Chinese New Year holiday. *** indicates that No.2 soybean delivery months are 1,3,5,7,9,11 (all months available after 2018 May contract).

Exchange Sector Commodity Symbol Contract Size Delivery Month Price Limit Margin First Contract First Price Date End Contract End Price Date DCE Grains No.1 Soybean A 10 Tons/Contract 1,3,5,7,9,11 ±4% 5% 2003 Mar 2002/03/29 2017 May 2017/04/28 DCE Grains No.2 Soybean B 10 Tons/Contract *** ±4% 5% 2005 July 2004/12/31 2017 May 2017/04/28 DCE Grains Corn C 10 Tons/Contract 1,3,5,7,9,11 ±4% 5% 2005 Jan 2004/09/30 2017 May 2017/04/28 DCE Industrial LLDPE L 5 Tons/Contract all ±4% 5% 2007 Oct 2007/07/31 2017 June 2017/05/31 DCE Industrial PVC V 5 Tons/Contract all ±4% 5% 2009 Sep 2009/05/29 2017 June 2017/05/31 DCE Industrial Metallurgical Coke J 100 Tons/Contract all ±4% 5% 2011 Sep 2011/04/29 2017 June 2017/05/31 DCE Industrial Coking Coal JM 60 Tons/Contract all ±4% 5% 2013 July 2013/03/29 2017 June 2017/05/31 DCE Oil Seeds Soybean Meal M 10 Tons/Contract 1,3,5,7,8,9,11,12 ±4% 5% 2000 Nov 2000/07/31 2017 May 2017/04/28 DCE Oil Seeds RBD Palm Olein P 10 Tons/Contract all ±4% 5% 2008 Jan 2007/10/31 2017 June 2017/05/31 DCE Oil Seeds Soybean Oil Y 10 Tons/Contract 1,3,5,7,8,9,11,12 ±4% 5% 2006 Mar 2006/01/31 2017 May 2017/04/28 SHFE Energy Fuel Oil FU 50 Tons/Contract ** ±5% 8% 2005 Jan 2004/08/31 2017 June 2017/05/31 SHFE Industrial Natural Rubber RU 10 Ton/Contract 1,3,4,5,6,7,8,9,10,11 ±3% 5% 1995 Sep 1995/07/31 2017 June 2017/05/31 SHFE Metal Aluminium AL 5 Tons/Contract all ±3% 5% 1994 Feb 1993/11/30 2017 June 2017/05/31 SHFE Metal Gold AU 1 Kilogram/Contract * ±3% 4% 2008 June 2008/01/31 2017 June 2017/05/31 SHFE Metal Copper CU 5 Tons/Contract all ±3% 5% 1993 June 1993/05/31 2017 June 2017/05/31 SHFE Metal Lead PB 5 Tons/Contract all ±4% 5% 2011 Sep 2011/03/31 2017 June 2017/05/31 SHFE Metal Steel Rebar RB 10 Tons/Contract all ±3% 5% 2009 Sep 2009/03/31 2017 June 2017/05/31 SHFE Metal Steel Wire Rod WR 10 Ton/Contract all ±5% 7% 2009 Sep 2009/03/31 2017 June 2017/05/31 SHFE Metal Zinc ZN 5 Tons/Contract all ±4% 5% 2007 July 2007/03/30 2017 June 2017/05/31 SHFE Metal Silver AG 15 Kilograms/Contract all ±3% 4% 2012 Sep 2012/05/31 2017 June 2017/05/31 ZCE Energy Methanol MA 10 Tons/Contract all ±4% 5% 2012 Mar 2011/10/31 2017 June 2017/05/31 ZCE Grains White Sugar SR 10 Tons/Contract 1,3,5,7,9,11 ±4% 6% 2006 May 2006/01/31 2017 May 2017/04/28 ZCE Grains Strong Gluten Wheat WH 20 Tons/Contract 1,3,5,7,9,11 ±4% 5% 2003 May 2003/03/31 2017 May 2017/04/28 ZCE Grains Common Wheat PM 50 Tons/Contract 1,3,5,7,9,11 ±4% 5% 1994 Jan 1993/12/31 2017 May 2017/04/28 ZCE Industrial Cotton No.1 CF 5 Tons/Contract 1,3,5,7,9,11 ±4% 5% 2004 Nov 2004/06/30 2017 May 2017/04/28 ZCE Industrial PTA TA 5 Tons/Contract all ±4% 5% 2007 Feb 2006/12/29 2017 June 2017/05/31 ZCE Industrial Flat Glass FG 20 Tons/Contract all ±4% 5% 2013 Mar 2012/12/31 2017 June 2017/05/31 ZCE Oil Seeds Rapeseed Oil OI 10 Tons/Contract 1,3,5,7,9,11 ±4% 5% 2007 July 2007/06/29 2017 May 2017/04/28 ZCE Oil Seeds Rapeseed Meal RM 10 Tons/Contract 1,3,5,7,8,9,11 ±4% 5% 2013 May 2012/12/31 2017 May 2017/04/28 ZCE Oil Seeds Rapeseed RS 10 Tons/Contract 7,8,9,11 ±4% 5% 2013 July 2012/12/31 2016 Nov 2016/10/31

45 Table 2 Summary statistics This table reports the summary statistics. The dataset spans the period from February 2004 to May 2017. We compute annualized returns (Panel A), standard deviations (Panel B), trading volume (Panel C) and Amihud illiquidity measure for returns on mth nearest to maturity contracts, where m = 1, 2, 3 and 4. Trading volume (expressed in thousands) is the monthly average number of contracts traded. The Amihud illiquidity (AI) measure is defined as Amihud (2002), where AI is the ratio of returns by turnover (=price x contract size x volume) in RMB. AI is expressed in basis point per one-million-RMB trade. mth nearest contract mth nearest contract st nd rd th st nd rd th Sectors Commodities 1 2 3 4 1 2 3 4 Panel A: Annualized returns Panel B: Annualized standard deviations Energy Methanol -12.8% -7.5% -9.2% -4.9% 27.1% 24.4% 22.9% 22.9% Fuel Oil -11.3% -5.9% -0.7% -7.2% 32.5% 28.8% 27.5% 28.0% Grains Sugar -0.9% -2.3% -4.2% -3.7% 19.7% 19.2% 20.2% 20.6% Strong Wheat -9.5% -6.9% -6.7% -5.0% 12.0% 10.5% 9.9% 10.3% Common Wheat -16.1% -9.3% -5.6% -3.3% 12.9% 10.1% 9.6% 9.5% No.1 Soybean -4.2% -0.1% 3.6% 1.7% 16.4% 14.8% 16.3% 16.8% No.2 Soybean 8.7% 3.7% 5.6% 3.0% 20.6% 15.8% 16.3% 18.3% Corn -4.6% -1.5% -0.3% -1.9% 13.1% 11.1% 10.1% 11.0% Oilseeds Rapeseed Oil -6.0% -4.4% -2.3% -5.5% 17.8% 19.7% 21.4% 22.0% Rapeseed Meal 11.5% 11.6% 11.4% 9.6% 27.9% 22.9% 22.5% 20.5% Rapeseed -3.9% -5.0% -10.5% -6.1% 12.2% 12.9% 14.9% 17.3% Soybean Meal 11.9% 10.5% 7.9% 6.4% 24.1% 21.5% 21.8% 21.3% Palm Olein -12.4% -16.9% -9.6% -6.0% 23.2% 27.2% 26.1% 25.9% Soybean Oil -1.1% -1.6% 0.9% 0.8% 23.5% 22.5% 22.4% 22.1% Industrial Cotton -2.0% -2.0% -2.4% -1.8% 18.4% 17.7% 18.0% 17.9% Flat Glass 9.2% 9.8% 3.1% -0.4% 25.9% 16.5% 17.7% 15.8% Natural Rubber -7.7% -5.0% -4.8% -5.6% 30.5% 31.5% 32.3% 32.2% LLDPE 3.6% -1.0% -2.1% -4.0% 28.5% 28.6% 28.8% 31.7% PVC -10.6% -3.9% -3.8% -5.0% 16.0% 17.4% 18.2% 17.2% PTA -7.6% -9.0% -9.3% -6.3% 26.0% 25.1% 25.2% 23.7% Coking Coke -13.7% -12.9% -13.0% -5.9% 36.8% 32.0% 31.9% 31.6% Coking Coal 10.5% -9.5% 3.8% -3.2% 32.0% 25.3% 29.2% 27.3% Metal Aluminium -2.1% -2.7% -3.2% -2.7% 15.5% 15.0% 15.0% 15.2% Gold 1.3% 1.0% -2.8% 6.4% 19.7% 19.3% 19.5% 19.0% Copper 10.3% 11.2% 9.2% 8.5% 28.4% 28.9% 29.5% 29.9% Lead -3.7% -3.1% -2.1% -2.1% 20.3% 20.2% 20.0% 19.8% Steel Rebar -7.8% -7.5% -6.3% -3.5% 28.3% 24.3% 23.3% 23.4% Steel Wire Rod -11.3% -5.3% -2.0% -7.6% 15.3% 17.8% 18.4% 16.5% Zinc -4.9% -6.2% -5.8% -5.6% 26.9% 27.2% 27.6% 27.8% Silver -13.0% -11.8% -10.6% -8.7% 23.9% 24.3% 24.6% 25.1%

Panel C: Trading volume Panel D: Amihud illiquidity Energy Methanol 391.0 3034.4 4301.4 4373.7 917.6 777.6 335.5 723.3 Fuel Oil 100.0 655.2 965.0 86.6 211.7 57.7 151.1 160.3 Grains Sugar 931.7 3739.2 7262.6 6004.9 1.8 1.8 2.2 2.4 Strong Wheat 69.6 363.6 829.8 680.7 43.2 20.7 31.6 33.7 Common Wheat 8.5 20.2 64.4 50.4 240.5 101.0 68.2 106.9 No.1 Soybean 286.6 1229.3 2252.9 1553.1 6.0 9.1 12.3 27.1 No.2 Soybean 2.0 3.4 15.7 11.0 476.3 265.7 197.8 328.3 Corn 477.2 1205.1 2438.7 1895.2 7.4 3.8 5.3 3.4 Oilseeds Rapeseed Oil 177.7 548.4 762.3 259.5 122.8 79.4 88.4 126.7 Rapeseed Meal 3313.7 11125.0 13785.7 8050.9 18.6 10.9 6.2 5.1 Rapeseed 1.8 2.4 64.3 0.3 293.8 484.1 89.2 442.4 Soybean Meal 808.8 3489.8 6477.6 6237.9 23.3 13.0 6.6 8.4 Palm Olein 71.2 833.4 1286.0 1915.6 283.0 243.3 111.8 117.5 Soybean Oil 193.4 1039.3 2573.9 2819.2 62.0 53.0 28.8 33.0 Industrial Cotton 246.0 1185.7 2199.3 1200.4 1.1 0.7 1.3 0.7 Flat Glass 206.2 1342.9 3580.9 3281.4 453.4 224.6 118.3 156.7 Natural Rubber 459.8 1985.4 3262.0 3439.2 1.1 1.3 1.4 2.1 LLDPE 180.9 1940.8 2558.2 2396.4 183.3 114.5 101.2 82.6 PVC 77.1 312.5 461.3 257.2 445.3 583.5 548.0 689.8 PTA 324.1 3093.7 3296.3 3344.0 28.1 22.6 22.9 44.7 Coking Coke 115.4 613.9 1089.5 2113.5 135.4 68.4 100.4 92.9 Coking Coal 90.8 797.7 1455.2 1524.4 446.9 987.5 618.7 427.6 Metal Aluminium 210.4 955.6 873.1 248.0 0.1 0.0 0.0 0.0 Gold 131.0 273.1 473.1 496.0 8.1 4.7 2.5 4.3 Copper 452.0 2488.9 3164.0 1452.5 0.0 0.0 0.0 0.0 Lead 107.4 177.6 30.7 4.0 0.5 0.7 41.2 38.2 Steel Rebar 503.1 4333.9 9063.0 12209.5 4.8 2.4 3.1 2.2 Steel Wire Rod 18.1 7.9 8.5 1.9 1038.5 1297.1 1454.6 747.0 Zinc 415.1 2904.3 3351.2 1282.2 0.1 0.0 0.0 0.2 Silver 1577.3 2764.2 3982.5 4189.4 0.9 2.3 2.9 5.5

46 Table 3 Performance of long-only investments

This table reports the performance of long-only investments on the broad market and sectors. Panel A reports the results over the sample period from February 2004 to May 2017 whereas Panel B reports the results over an extended sample from 1992. AVG represents the broad market of 30 commodities. Energy, Grains, Industrial, Metals and Oilseeds report the performance of each commodity sector as classified in Table 2. All portfolios are equally-weighted. For each portfolio, we report returns on mth nearest to maturity contracts, where m = 1, 2, 3 and 4. Not all commodities started trading prior to our sample period in 2004.

AVG Energy Grains Industrial Metals Oilseeds Panel A: 2004-2017 First nearest contract (m=1) Annualized arithmetic mean -0.0285 -0.0285 -0.0285 -0.0285 -0.0285 -0.0285 t-statistics -0.84 -0.84 -0.84 -0.84 -0.84 -0.84 Annualized volatility 0.1234 0.1234 0.1234 0.1234 0.1234 0.1234 Sharpe Ratio -0.2306 -0.2306 -0.2306 -0.2306 -0.2306 -0.2306 Second nearest contract (m=2) Annualized arithmetic mean -0.0219 -0.0219 -0.0219 -0.0219 -0.0219 -0.0219 t-statistics -0.61 -0.61 -0.61 -0.61 -0.61 -0.61 Annualized volatility 0.1311 0.1311 0.1311 0.1311 0.1311 0.1311 Sharpe Ratio -0.1673 -0.1673 -0.1673 -0.1673 -0.1673 -0.1673 Third nearest contract (m=3) Annualized arithmetic mean -0.0139 -0.0139 -0.0139 -0.0139 -0.0139 -0.0139 t-statistics -0.36 -0.36 -0.36 -0.36 -0.36 -0.36 Annualized volatility 0.1407 0.1407 0.1407 0.1407 0.1407 0.1407 Sharpe Ratio -0.0987 -0.0987 -0.0987 -0.0987 -0.0987 -0.0987 Fourth nearest contract (m=4) Annualized arithmetic mean -0.0124 -0.0124 -0.0124 -0.0124 -0.0124 -0.0124 t-statistics -0.31 -0.31 -0.31 -0.31 -0.31 -0.31 Annualized volatility 0.1449 0.1449 0.1449 0.1449 0.1449 0.1449 Sharpe Ratio -0.0855 -0.0855 -0.0855 -0.0855 -0.0855 -0.0855

Panel B: 1992-2017 First nearest contract (m=1) Annualized arithmetic mean -0.0299 -0.0299 -0.0299 -0.0299 -0.0299 -0.0299 t-statistics -1.24 -1.24 -1.24 -1.24 -1.24 -1.24 Annualized volatility 0.1182 0.1182 0.1182 0.1182 0.1182 0.1182 Sharpe Ratio -0.2527 -0.2527 -0.2527 -0.2527 -0.2527 -0.2527 Second nearest contract (m=2) Annualized arithmetic mean -0.0234 -0.0234 -0.0234 -0.0234 -0.0234 -0.0234 t-statistics -0.89 -0.89 -0.89 -0.89 -0.89 -0.89 Annualized volatility 0.1282 0.1282 0.1282 0.1282 0.1282 0.1282 Sharpe Ratio -0.1823 -0.1823 -0.1823 -0.1823 -0.1823 -0.1823 Second nearest contract (m=3) Annualized arithmetic mean -0.0172 -0.0172 -0.0172 -0.0172 -0.0172 -0.0172 t-statistics -0.62 -0.62 -0.62 -0.62 -0.62 -0.62 Annualized volatility 0.1347 0.1347 0.1347 0.1347 0.1347 0.1347 Sharpe Ratio -0.1278 -0.1278 -0.1278 -0.1278 -0.1278 -0.1278 Fourth nearest contract (m=4) Annualized arithmetic mean -0.0179 -0.0179 -0.0179 -0.0179 -0.0179 -0.0179 t-statistics -0.62 -0.62 -0.62 -0.62 -0.62 -0.62 Annualized volatility 0.1396 0.1396 0.1396 0.1396 0.1396 0.1396 Sharpe Ratio -0.1280 -0.1280 -0.1280 -0.1280 -0.1280 -0.1280

47 Table 4 Performance of long-short strategies

This table reports the performance of 12 long-short strategies. Panels A through to D report results based on returns on mth nearest to maturity contracts, where m = 1, 2, 3 and 4, respectively. For each strategy, we sort commodities into quartiles based on the respective signal. At the end of each month, we take long positions in commodities within the top quartile and short positions in the bottom quartile. All portfolios are equally weighted and rebalanced monthly. The term structure signal is based on the roll-yield from the front to the next nearest contract. Hedgers’ hedging pressure (HP) and speculators’ HP strategy sort commodities based on hedging and speculation ratio introduced by Bohl et al. (2018). The hedging ratio is the change of open interest divided by volume, while the speculation ratio is volume divided by open interest. Momentum and time-series momentum strategies use past 12-month returns as the sorting signal. The volatility signal employed is the coefficient of variation computed based on prior 3-year daily returns. The open interest signal is computed as the change of monthly open interests of the entire curve. The liquidity signal employed is the Amivest measure as defined by Amihud (2002), computed as the average ratio of monthly turnover to absolute return in the past 2 months using daily returns. The FX signal is the regression beta of commodity returns on the RMB effective exchange rate index, with a rolling window of 42 months. The inflation signal is the regression beta of commodity returns on unexpected inflation with a rolling window of 42 months. The skewness signal is computed as the Pearson skewness based on the past 12-month daily return. Finally, the value signal is obtained by taking the log of the average daily prices from 4.5 to 5.5 years ago divided by the price at each time t. The sample period covers February 2004 through to May 2017.

Term Speculators' Time-series Inflation Hedgers' HP Momentum Volatility Open Interest Liquidity FX Skewness Value Structure HP Momentum shocks Panel A: First nearest contract (m=1) Annualized arithmetic mean 0.1591 0.1099 0.0443 0.2194 0.1408 0.1053 -0.0367 -0.0684 0.0477 0.0222 -0.0958 -0.0414 t-statistics 3.24 2.68 1.02 4.13 3.24 2.43 -0.94 -1.43 0.83 0.44 -2.32 -0.77 Annualized volatility 0.179 0.1485 0.1577 0.1866 0.1527 0.1394 0.1421 0.1736 0.1805 0.1599 0.1453 0.1513 Sharpe Ratio 0.8891 0.7399 0.2808 1.1757 0.9221 0.7555 -0.2582 -0.3942 0.2645 0.1387 -0.6596 -0.2737 Sortino Ratio 1.0041 1.3026 0.3233 2.1373 1.2675 1.1409 -0.3579 -0.5722 0.289 0.2039 -0.9312 -0.3798 Skewness -1.7586 0.0045 -2.059 0.1996 -0.3448 -0.3185 -0.5605 0.0855 -1.5333 -0.1431 0.1775 -0.2503 Kurtosis 14.4057 3.0681 15.7433 4.1832 5.5877 4.6964 3.8109 4.2825 12.0713 4.4294 4.1239 5.8152 99%VaR(Cornish-Fisher) 0.2705 0.1398 0.2496 0.207 0.1576 0.1344 0.1092 0.1655 0.2329 0.1504 0.1375 0.1469

Panel B: Second nearest contract (m=2) Annualized arithmetic mean 0.1753 -0.0258 0.0237 0.1738 0.0817 -0.0019 -0.0558 0.0290 0.0104 -0.0005 0.027 0.0018 t-statistics 3.94 -0.68 0.46 3.59 2.12 -0.04 -1.52 0.6 0.17 -0.01 0.71 0.04 Annualized volatility 0.1614 0.1379 0.1869 0.1695 0.135 0.1536 0.1326 0.1759 0.1873 0.1514 0.1335 0.1303 Sharpe Ratio 1.086 -0.187 0.1268 1.0252 0.605 -0.0125 -0.421 0.1646 0.0555 -0.0031 0.2024 0.0139 Sortino Ratio 1.3365 -0.2294 0.1256 1.8392 0.8382 -0.0146 -0.5369 0.2523 0.0571 -0.0046 0.2984 0.0238 Skewness -1.2063 -0.938 -3.039 0.4011 -0.4513 -0.9381 -1.0737 0.0942 -2.2592 0.1091 -0.3505 0.3019 Kurtosis 11.6339 7.0103 24.3918 5.2736 6.2134 6.8622 9.9717 5.5141 14.1756 4.1316 4.0679 3.3946 99%VaR(Cornish-Fisher) 0.2152 0.129 0.4474 0.2067 0.1383 0.1444 0.1464 0.1910 0.2792 0.1487 0.1159 0.1284

Panel C: Third nearest contract (m=3) Annualized arithmetic mean 0.1379 -0.036 -0.0079 0.1671 0.0827 -0.0369 -0.0311 0.0088 0.0153 0.0477 0.0152 0.0288 t-statistics 3.01 -1.04 -0.18 3.1 2.19 -0.67 -0.95 0.22 0.26 1 0.42 0.58 Annualized volatility 0.166 0.125 0.1548 0.1882 0.1317 0.1745 0.1183 0.1457 0.1842 0.1475 0.1272 0.1377 Sharpe Ratio 0.8308 -0.2885 -0.0511 0.888 0.6279 -0.2112 -0.2628 0.0605 0.083 0.3232 0.1197 0.2091 Sortino Ratio 0.8823 -0.4561 -0.0585 1.5867 0.9742 -0.1945 -0.3922 0.1024 0.0919 0.5486 0.1707 0.3884 Skewness -2.3344 0.0762 -1.6019 0.2985 -0.1276 -4.0245 0.0817 0.2074 -1.6907 0.1588 -0.5253 0.0593 Kurtosis 20.2724 3.4498 12.4489 4.4897 4.0098 32.8041 5.8944 3.4807 10.1077 3.856 3.8952 3.0171 99%VaR(Cornish-Fisher) 0.3288 0.1129 0.2 0.2129 0.126 0.5908 0.1268 0.1415 0.2117 0.1482 0.1033 0.1254

Panel D: Fourth nearest contract (m=4) Annualized arithmetic mean 0.1169 0.0052 0.0186 0.1526 0.0641 0.0199 -0.0553 -0.0137 0.0081 0.02 -0.0098 -0.0061 t-statistics 2.57 0.14 0.54 2.75 1.6 0.51 -1.64 -0.42 0.15 0.41 -0.25 -0.12 Annualized volatility 0.1639 0.1294 0.1248 0.1932 0.1393 0.123 0.1211 0.1173 0.1718 0.1517 0.1377 0.1417 Sharpe Ratio 0.7129 0.0401 0.1491 0.7901 0.46 0.1621 -0.457 -0.1167 0.0471 0.1315 -0.0715 -0.0431 Sortino Ratio 0.7469 0.0627 0.2399 1.3665 0.6614 0.169 -0.6244 -0.1819 0.0517 0.1915 -0.1066 -0.0718 Skewness -2.0513 -0.0752 -0.1603 0.2071 -0.2966 -1.6195 -0.3862 -0.1116 -1.6516 -0.1599 -0.158 0.3003 Kurtosis 16.3461 3.3751 3.2635 4.1648 3.9546 9.9784 8.423 2.8749 9.4287 4.8733 3.6408 4.039 99%VaR(Cornish-Fisher) 0.2719 0.1138 0.107 0.2082 0.1247 0.1407 0.1331 0.0965 0.1887 0.1464 0.1191 0.145

48 Table 5 Correlations of trading strategies

This table presents the Pearson correlation among strategy returns. Panel A reports results on the first nearest to maturity contract whereas Panel B reports the third nearest contract. To construct strategy portfolios, we sort commodities into quartiles based on the respective signal. At the end of each month, we take long positions in commodities within the top quartile and short positions in the bottom quartile. All portfolios are equally weighted and rebalanced monthly. The term structure signal is based on the roll-yield from the front to the next nearest contract. Hedgers’ hedging pressure (HP) and speculators’ HP strategy sort commodities based on hedging and speculation ratio introduced by Bohl et al. (2018). The hedging ratio is the change of open interest divided by volume, while the speculation ratio is volume divided by open interest. Momentum and time-series momentum strategies use past 12-month returns as the sorting signal. The volatility signal employed is the coefficient of variation computed based on prior 3-year daily returns. The open interest signal is computed as the change of monthly open interests of the entire curve. The liquidity signal employed is the Amivest measure as defined by Amihud (2002), computed as the average ratio of monthly turnover to absolute return in the past 2 months using daily returns. The FX signal is the regression beta of commodity returns on the RMB effective exchange rate index, with a rolling window of 42 months. The inflation signal is the regression beta of commodity returns on unexpected inflation with a rolling window of 42 months. The skewness signal is computed as the Pearson skewness based on the past 12-month daily return. Finally, the value signal is obtained by taking the log of the average daily prices from 4.5 to 5.5 years ago divided by the price at each time t. The sample period covers February 2004 through to May 2017. *denotes significance at 5% level or better.

Term Structure Hedgers' HP Speculators' HP Momentum Time-series Momentum Volatility Open Interest Liquidity FX Inflation shocks Skewness

Panel A: First nearest contract (m=1) Hedgers' HP -0.1820 Speculators' HP 0.4142* -0.1945 Momentum 0.4673* -0.0135 0.2475 Time-series Momentum 0.4846* -0.0094 0.3513* 0.8008* Volatility 0.2639 -0.1460 0.3059* 0.1445 0.2625 Open Interest 0.1120 -0.0898 0.2074 -0.0697 0.076 0.1744 Liquidity -0.2184 0.2363 -0.4096* -0.2757* -0.2664 -0.2054 -0.1443 FX 0.4475* -0.4301* 0.4652* 0.0200 0.2366 0.4930* 0.3872* -0.5278* Inflation shocks -0.2077 -0.0848 -0.2181 -0.1587 -0.0653 0.1124 0.0833 0.0453 0.1304 Skewness -0.0315 -0.0723 -0.0887 -0.0505 -0.0575 0.0235 -0.0263 0.0277 -0.0968 0.0719 Value -0.2618 -0.0795 -0.0370 -0.4353* -0.3663* -0.0518 -0.0386 0.0421 0.2386 0.4762* 0.1834

Panel B: Third nearest contract (m=3) Hedgers' HP 0.0640 1 Speculators' HP 0.2206 0.1094 1 Momentum 0.4467* -0.0455 0.1091 1 Time-series Momentum 0.3489* 0.0344 0.0771 0.7967* 1 Volatility 0.6190* 0.0136 0.6246* 0.0687 0.0348 1 Open Interest 0.2139 0.0818 0.4233* -0.0724 -0.0486 0.3382* 1 Liquidity -0.1264 0.0137 -0.4855* -0.1361 -0.0391 -0.2504 -0.0610 1 FX 0.5206* 0.0660 0.4329* -0.051 0.0343 0.5544* 0.3967* -0.3999* 1 Inflation shocks 0.2643 -0.0735 0.3709* 0.0766 0.1766 0.2234 0.2636 -0.2731 0.5241* 1 Skewness -0.0408 -0.0940 -0.0088 0.0741 0.0496 0.0075 -0.0763 -0.0550 -0.0748 0.2209 1 Value -0.1277 -0.0775 -0.0512 -0.4197* -0.3526* -0.1128 -0.0389 0.0320 0.0724 0.2604 0.1596

49 Table 6 Market liquidity

This table reports the performance of long-short strategies executed on a restricted sample of top 20 most liquid commodities. We eliminate 10 most illiquid commodities with the lowest average trading volume. Panel A reports results on the first nearest to maturity contract whereas Panel B reports the third nearest contract. For each strategy, we sort commodities into quartiles based on the respective signal. At the end of each month, we take long positions in commodities within the top quartile and short positions in the bottom quartile. All portfolios are equally weighted and rebalanced monthly. The term structure signal is based on the roll-yield from the front to the next nearest contract. Hedgers’ hedging pressure (HP) and speculators’ HP strategy sort commodities based on hedging and speculation ratio introduced by Bohl et al. (2018). The hedging ratio is the change of open interest divided by volume, while the speculation ratio is volume divided by open interest. Momentum and time-series momentum strategies use past 12-month returns as the sorting signal. The volatility signal employed is the coefficient of variation computed based on prior 3-year daily returns. The open interest signal is computed as the change of monthly open interests of the entire curve. The liquidity signal employed is the Amivest measure as defined by Amihud (2002), computed as the average ratio of monthly turnover to absolute return in the past 2 months using daily returns. The FX signal is the regression beta of commodity returns on the RMB effective exchange rate index, with a rolling window of 42 months. The inflation signal is the regression beta of commodity returns on unexpected inflation with a rolling window of 42 months. The skewness signal is computed as the Pearson skewness based on the past 12-month daily return. Finally, the value signal is obtained by taking the log of the average daily prices from 4.5 to 5.5 years ago divided by the price at each time t. The sample period covers February 2004 through to May 2017.

Time-series Open Inflation Term Structure Hedgers' HP Speculators' HP Momentum Volatility Liquidity FX Skewness Value Momentum Interest Shocks Panel A: First nearest contract (m=1) Annualized arithmetic mean 0.1895 0.1302 0.0106 0.2193 0.1253 0.0704 -0.0338 0.0206 -0.0708 -0.0369 -0.0917 -0.0784 t-statistics 3.71 2.70 0.20 3.46 2.63 1.56 -0.72 0.39 -1.02 -0.60 -1.77 -1.38 Annualized geometric mean 0.1726 0.1151 -0.0085 0.1958 0.1115 0.0599 -0.0485 0.0029 -0.0961 -0.0554 -0.1085 -0.0909 Annualized volatility 0.1857 0.1749 0.1909 0.2223 0.1672 0.1454 0.1699 0.1901 0.2173 0.1931 0.1822 0.1588 Annualized downside volatility 0.1086 0.1067 0.1532 0.1227 0.1099 0.0961 0.1219 0.1190 0.1918 0.1350 0.1269 0.1025 Sharpe Ratio 1.0204 0.7446 0.0554 0.9864 0.7493 0.4838 -0.1988 0.1083 -0.3259 -0.1911 -0.5032 -0.4936 Sortino Ratio 1.9041 1.2956 0.0693 1.9791 1.2077 0.7561 -0.2728 0.1745 -0.3575 -0.2686 -0.6925 -0.7380 Skewness 0.1675 -0.0087 -1.0691 0.9417 0.0843 -0.0134 -0.5772 0.5855 -1.2530 0.2866 -0.2111 0.1979 Kurtosis 4.2658 4.5427 7.5253 8.8849 4.5566 3.8806 4.2150 5.7942 8.1232 7.9165 3.6290 3.2046 Max monthly gain 0.1982 0.1523 0.1333 0.3756 0.1850 0.1361 0.1152 0.2259 0.1692 0.2466 0.1236 0.1246 Max monthly loss -0.1734 -0.1914 -0.2988 -0.1840 -0.1296 -0.1053 -0.1813 -0.1650 -0.3128 -0.2264 -0.1966 -0.1202 99%VaR(Cornish-Fisher) 0.2030 0.1814 0.1868 0.3441 0.1783 0.1410 0.1355 0.2310 0.2123 0.2451 0.1483 0.1435 % of positive months 0.6352 0.5759 0.5409 0.6351 0.6081 0.5968 0.5127 0.4747 0.4831 0.4576 0.4324 0.4362 Maximum Drawdown -0.2917 -0.3844 -0.6410 -0.3399 -0.2566 -0.2035 -0.5346 -0.5363 -0.6604 -0.4667 -0.8342 -0.4950 Drawdown Length (months) 4 9 66 12 8 4 89 95 16 23 117 92 Max Run-up (consecutive) 0.5410 0.5576 0.0000 0.5717 0.3157 0.3430 0.0000 0.4068 0.0000 0.2306 0.3750 0.0000 Run-up Length (months) 9 7 0 3 5 5 0 3 0 2 7 0 Max 12M rolling return 0.6805 0.5666 0.3670 0.9828 0.5633 0.5451 0.2640 0.4851 0.6626 0.3541 0.6002 0.2542 Min 12M rolling return -0.2569 -0.4200 -0.4706 -0.3988 -0.2414 -0.1772 -0.3230 -0.3826 -0.7574 -0.3512 -0.5736 -0.3074

Panel B: Third nearest contract (m=3) Annualized arithmetic mean 0.1171 0.0323 -0.0292 0.2137 0.0738 -0.0025 0.0003 -0.0125 -0.0621 0.0686 -0.0195 0.0757 t-statistics 2.17 0.76 -0.59 3.38 1.61 -0.04 0.01 -0.26 -1.02 1.13 -0.40 1.12 Annualized geometric mean 0.0980 0.0207 -0.0453 0.1903 0.0611 -0.0228 -0.0143 -0.0276 -0.0814 0.0511 -0.0337 0.0588 Annualized volatility 0.1955 0.1535 0.1781 0.2205 0.1600 0.1943 0.1671 0.1739 0.1896 0.1891 0.1691 0.1873 Annualized downside volatility 0.1326 0.0948 0.1352 0.1318 0.1027 0.1730 0.1308 0.1166 0.1728 0.1205 0.0911 0.0920 Sharpe Ratio 0.5989 0.2105 -0.1641 0.9691 0.4617 -0.0130 0.0018 -0.0721 -0.3276 0.3629 -0.1156 0.4042 Sortino Ratio 0.9319 0.3459 -0.2132 1.7902 0.7440 -0.0146 0.0022 -0.1070 -0.3494 0.5876 -0.2125 0.8522 Skewness -0.2542 0.2636 -0.3944 0.5856 -0.0877 -1.6842 -1.1542 0.2026 -1.6807 0.3151 0.1314 0.6653 Kurtosis 5.3689 4.3947 5.9453 6.9385 3.5361 11.4625 11.1395 4.7034 8.4204 6.1396 2.8691 4.5141 Max monthly gain 0.2008 0.1615 0.1835 0.3409 0.1604 0.1444 0.1656 0.1989 0.0993 0.2129 0.1622 0.2005 Max monthly loss -0.2362 -0.1322 -0.2045 -0.1632 -0.1082 -0.3200 -0.2925 -0.1322 -0.2678 -0.1957 -0.1106 -0.1218 99%VaR(Cornish-Fisher) 0.1979 0.1626 0.1700 0.3008 0.1475 0.2395 0.2026 0.1810 0.1898 0.2274 0.1512 0.2185 % of positive months 0.5796 0.5192 0.5192 0.6301 0.5685 0.5000 0.4872 0.4872 0.4741 0.5086 0.5274 0.5217 Maximum Drawdown -0.3455 -0.4584 -0.5533 -0.3480 -0.3544 -0.4343 -0.5079 -0.6080 -0.5535 -0.3788 -0.6835 -0.2676 Drawdown Length (months) 18 47 54 13 18 3 59 101 16 56 81 20 Max Run-up (consecutive) 0.3705 0.1419 0.0000 1.0239 0.2699 0.2341 0.1830 0.1337 0.0000 0.3498 0.2469 0.3094 Runup Length (months) 2 3 0 11 2 4 3 1 0 3 3 3 Max 12M rolling return 0.7017 0.3406 0.3080 1.0862 0.4844 0.5264 0.3537 0.2946 0.2598 0.6594 0.4692 0.5841 Min 12M rolling return -0.3323 -0.2847 -0.3992 -0.3937 -0.3752 -0.3082 -0.3964 -0.5407 -0.4657 -0.2279 -0.3155 -0.2123

50 Table 7 Commodity sectors

This table reports the performance of long-short strategies after removing one commodity sector at a time. Panels A through to E reports the strategy returns on non-industrial, non-metal, non-grains, non-oil seeds and non-energy sample of commodities, respectively. The results are based on the first nearest to maturity contracts. For each strategy, we sort commodities into quartiles based on the respective signal. At the end of each month, we take long positions in commodities within the top quartile and short positions in the bottom quartile. All portfolios are equally weighted and rebalanced monthly. The term structure signal is based on the roll-yield from the front to the next nearest contract. Hedgers’ hedging pressure (HP) and speculators’ HP strategy sort commodities based on hedging and speculation ratio introduced by Bohl et al. (2018). The hedging ratio is the change of open interest divided by volume, while the speculation ratio is volume divided by open interest. Momentum and time-series momentum strategies use past 12-month returns as the sorting signal. The volatility signal employed is the coefficient of variation computed based on the prior 3-year daily returns. The open interest signal is computed as the change of monthly open interests of the entire curve. The liquidity signal employed is the Amivest measure as defined by Amihud (2002), computed as the average ratio of monthly turnover to absolute return in the past 2 months using daily returns. The FX signal is the regression beta of commodity returns on the RMB effective exchange rate index, with a rolling window of 42 months. The inflation signal is the regression beta of commodity returns on unexpected inflation with a rolling window of 42 months. The skewness signal is computed as the Pearson skewness based on the past 12-month daily return. Finally, the value signal is obtained by taking the log of the average daily prices from 4.5 to 5.5 years ago divided by the price at each time t. The sample period covers February 2004 through to May 2017. Term Speculators' Time-series Open Inflation Hedgers' HP Momentum Volatility Liquidity FX Skewness Value Structure HP Momentum Interest Shocks Panel A: Excluding industrial sector Annualized arithmetic mean 0.1490 0.1466 0.0514 0.1979 0.1338 0.1310 -0.0481 -0.0926 0.0713 0.0299 -0.0629 -0.0796 t-statistics 3.10 3.18 1.24 3.60 2.98 2.70 -1.08 -1.83 1.38 0.59 -1.33 -1.41 Annualized volatility 0.1748 0.1673 0.1508 0.1929 0.1576 0.1559 0.1614 0.1834 0.1618 0.1588 0.1665 0.1583 Sharpe Ratio 0.8525 0.8762 0.3405 1.0255 0.8489 0.8406 -0.2980 -0.5049 0.4407 0.1882 -0.3776 -0.5031 Sortino Ratio 1.2010 1.5981 0.4703 1.7109 1.1278 2.0495 -0.3706 -0.7031 0.5621 0.2790 -0.5185 -0.6625 Skewness -0.6112 0.5060 -0.6005 -0.0305 -0.4028 0.6377 -0.9620 -0.1451 -0.9402 -0.2832 -0.0154 -0.6243 Kurtosis 5.3305 5.4669 3.7725 4.7128 6.2672 3.4483 6.2343 3.9249 7.0441 3.9459 3.5595 5.7054

Panel B: Excluding metal sector Annualized arithmetic mean 0.1143 0.0894 0.0726 0.2020 0.1166 0.0959 -0.0116 -0.0427 0.0603 -0.0324 -0.1501 -0.0358 t-statistics 2.19 2.03 1.42 3.54 2.43 1.99 -0.27 -0.99 1.23 -0.56 -3.53 -0.54 Annualized volatility 0.1897 0.1596 0.1860 0.2006 0.1687 0.1552 0.1557 0.1561 0.1541 0.1808 0.1491 0.1844 Sharpe Ratio 0.6026 0.5602 0.3903 1.0073 0.6910 0.6180 -0.0746 -0.2737 0.3913 -0.1795 -1.0062 -0.1944 Sortino Ratio 0.7033 0.8677 0.4842 1.7856 0.9283 1.0934 -0.1011 -0.3523 0.5252 -0.2414 -1.3976 -0.2559 Skewness -1.6390 -0.1890 -1.1744 0.0552 -0.6509 -0.0500 -0.5333 -0.5779 -0.6040 -0.3039 0.2101 -0.5057 Kurtosis 11.7901 3.7767 8.8887 3.8173 6.1754 3.7822 3.5153 4.9509 7.0171 4.1607 4.0029 5.2092

Panel C: Excluding grains sector Annualized arithmetic mean 0.1511 0.1059 -0.0119 0.2826 0.1431 0.0469 -0.0224 -0.0070 -0.0755 -0.0332 -0.1123 -0.0985 t-statistics 2.82 2.17 -0.22 4.51 2.91 0.78 -0.44 -0.11 -1.05 -0.52 -1.95 -1.43 Annualized volatility 0.1952 0.1770 0.1926 0.2199 0.1726 0.1922 0.1832 0.2253 0.2251 0.1974 0.2025 0.1922 Sharpe Ratio 0.7741 0.5986 -0.0616 1.2850 0.8292 0.2442 -0.1224 -0.0313 -0.3356 -0.1684 -0.5545 -0.5126 Sortino Ratio 1.3825 1.0283 -0.0795 2.8781 1.1838 0.2757 -0.1554 -0.0448 -0.3435 -0.2659 -0.7226 -0.7375 Skewness 0.3703 0.0385 -0.7343 1.0107 -0.4838 -1.2760 -0.9247 0.0169 -1.2765 0.1421 -0.3877 0.0745 Kurtosis 5.3088 6.2298 6.7237 7.8091 6.0049 7.0641 6.6164 8.0951 7.5985 3.2981 3.9091 4.0712

Panel D: Excluding oilseeds sector Annualized arithmetic mean 0.1279 0.0861 0.0194 0.1973 0.1224 0.0742 0.0090 -0.0635 0.0555 0.0468 -0.0674 -0.0322 t-statistics 2.12 1.84 0.34 3.53 2.92 1.52 0.22 -1.16 0.90 0.85 -1.67 -0.53 Annualized volatility 0.2196 0.1698 0.2068 0.1962 0.1472 0.1565 0.1475 0.1986 0.1935 0.1729 0.1417 0.1691 Sharpe Ratio 0.5822 0.5072 0.0938 1.0055 0.8316 0.4740 0.0608 -0.32 0.2870 0.2704 -0.4754 -0.1904 Sortino Ratio 0.6225 0.8681 0.1105 1.6111 1.0639 0.7140 0.0863 -0.4845 0.3331 0.3661 -0.7672 -0.2789 Skewness -2.5593 0.1634 -1.5646 -0.0405 -0.7411 -0.4020 -0.3952 0.7427 -1.5169 -0.5414 0.5542 -0.1891 Kurtosis 21.0226 4.2619 12.2398 4.1638 7.4556 4.0643 4.7446 7.6727 12.0799 4.3068 5.0154 4.4834

Panel E: Excluding energy sector Annualized arithmetic mean 0.1761 0.1119 0.0389 0.2320 0.1475 0.1186 -0.0332 -0.0478 0.0413 0.0310 -0.1068 -0.0446 t-statistics 3.82 2.66 0.92 4.35 3.36 2.56 -0.83 -1.05 0.76 0.66 -2.66 -0.93 Annualized volatility 0.1676 0.1523 0.1537 0.1872 0.1543 0.1488 0.1452 0.1656 0.1700 0.1474 0.1412 0.1338 Sharpe Ratio 1.0506 0.7343 0.2532 1.2393 0.9560 0.7972 -0.2289 -0.2885 0.2427 0.2100 -0.7564 -0.3334 Sortino Ratio 1.0975 1.3363 0.2808 2.3952 1.4242 1.2563 -0.3267 -0.4263 0.2945 0.2828 -1.0170 -0.5342 Skewness -2.2467 0.1024 -2.2677 0.4056 -0.0178 -0.2494 -0.4913 0.1157 -1.1403 -0.5719 -0.0872 0.4202 Kurtosis 18.2023 3.3439 17.2174 4.4570 5.4266 4.3625 3.8379 4.3355 8.3086 5.8068 3.7910 4.2016

51 Table 8 Decomposing long-short strategy returns

This table reports the performance of quartile portfolios for term structure, hedgers’ hedging pressure, momentum, volatility and time-series momentum strategies. We report results on both the first and third nearest to maturity contracts. The term structure signal is based on the roll-yield from the front to the next nearest contract. Hedgers’ hedging pressure (HP) and speculators’ HP strategy sort commodities based on hedging and speculation ratio introduced by Bohl et al. (2018). The hedging ratio is the change of open interest divided by volume, while the speculation ratio is volume divided by open interest. Momentum and time-series momentum strategies use past 12-month returns as the sorting signal. The volatility signal employed is the coefficient of variation computed based on prior 3-year daily returns. This table also reports portfolio turnover and net return for each strategy. Portfolio turnover is estimated following Fuertes et al. (2010). The transaction cost employed in computing net return is based on Marshall et al. (2012) who document a 0.086% average transaction cost in the commodity futures market. The sample period covers February 2004 through to May 2017. First nearest contract (m=1) Third nearest contract (m=3) Q1 Q2 Q3 Q4 L-S Q1 Q2 Q3 Q4 L-S Term Structure Annualized arithmetic mean -0.0902 -0.0846 -0.0187 0.0689 0.1591 -0.0627 -0.0561 -0.0130 0.0752 0.1379 t-statistics -2.77 -2.13 -0.43 1.28 3.24 -1.71 -1.34 -0.29 1.32 3.01 Annualized volatility 0.1185 0.1448 0.1594 0.1961 0.1790 0.1327 0.1518 0.1606 0.2066 0.1660 Sharpe Ratio -0.7611 -0.5844 -0.1172 0.3516 0.8891 -0.4724 -0.3696 -0.0808 0.3640 0.8308 Skewness 0.1385 -1.5115 -0.2338 -2.2903 -1.7586 0.3370 -1.5845 -0.5048 -2.3509 -2.3344 Kurtosis 3.0586 10.1581 5.8269 17.9179 14.4057 4.8093 10.7171 5.5657 19.0136 20.2724 Portfolio Turnover 5.8929 7.7477 7.1403 6.4591 6.176 5.8929 7.7477 7.1403 6.4591 6.1760 Net Return 0.1485 0.1273 Hedgers' Hedging Pressure Annualized arithmetic mean 0.0090 -0.0363 -0.0113 -0.1009 0.1099 -0.0425 -0.0079 -0.0264 -0.0064 -0.0360 t-statistics 0.25 -0.88 -0.27 -2.05 2.68 -1.13 -0.18 -0.48 -0.15 -1.04 Annualized volatility 0.1308 0.1493 0.1510 0.1784 0.1485 0.1358 0.1620 0.1966 0.1516 0.1250 Sharpe Ratio 0.0690 -0.2435 -0.0746 -0.5653 0.7399 -0.3130 -0.0486 -0.1341 -0.0425 -0.2885 Skewness -1.0288 0.0193 -0.0027 -1.6624 0.0045 -0.5792 -0.4952 -2.8599 -0.1144 0.0762 Kurtosis 9.1119 5.1117 5.7508 12.6008 3.0681 6.3438 4.7162 23.3577 4.4182 3.4498 Portfolio Turnover 9.2171 9.6915 9.0970 10.5896 9.9033 8.5685 9.2255 8.7876 10.2087 9.1976 Net Return 0.0929 -0.0518 Momentum Annualized arithmetic mean -0.0992 -0.0632 -0.0075 0.1202 0.2194 -0.0654 -0.0013 -0.0135 0.1017 0.1671 t-statistics -2.61 -1.95 -0.16 2.13 4.13 -1.55 -0.03 -0.31 1.69 3.10 Annualized volatility 0.1337 0.1136 0.1684 0.1983 0.1866 0.1473 0.1429 0.1528 0.2094 0.1882 Sharpe Ratio -0.7420 -0.5562 -0.0447 0.6063 1.1757 -0.4439 -0.0093 -0.0882 0.4859 0.8880 Skewness -0.1082 0.2815 -2.9240 -0.2729 0.1996 -0.5550 -0.8747 -1.7779 -0.2653 0.2985 Kurtosis 5.0310 3.7031 25.1664 5.5433 4.1832 7.2530 8.6591 12.8271 5.9367 4.4897 Portfolio Turnover 2.3491 5.0439 4.4620 2.2355 2.2923 2.6832 5.3246 4.6689 2.3516 3.7571 Net Return 0.2155 0.1606 Volatility Annualized arithmetic mean -0.0711 -0.0195 -0.0234 0.0343 0.1053 -0.0002 0.0184 0.0080 -0.0371 -0.0369 t-statistics -2.26 -0.38 -0.47 0.72 2.43 -0.01 0.39 0.15 -0.59 -0.67 Annualized volatility 0.1009 0.1640 0.1601 0.1526 0.1394 0.1173 0.1502 0.1651 0.2005 0.1745 Sharpe Ratio -0.7042 -0.1190 -0.1464 0.2245 0.7555 -0.0020 0.1227 0.0482 -0.1850 -0.2112 Skewness 0.5579 -0.5986 -0.6136 -0.4562 -0.3185 0.7342 0.1104 -1.6149 -3.5883 -4.0245 Kurtosis 3.9551 6.4202 6.5832 5.6247 4.6964 4.6004 3.4690 10.9244 27.6213 32.8041 Portfolio Turnover 1.4510 4.1783 4.5678 3.1365 2.2937 2.2487 5.1349 5.0120 3.7893 4.0462 Net Return 0.1014 -0.0439 Time-series Momentum Annualized arithmetic mean -0.0813 0.0595 0.1408 -0.0353 0.0483 0.0827 t-statistics -2.50 1.17 3.24 -0.94 1.03 2.19 Annualized volatility 0.1142 0.1786 0.1527 0.1301 0.1624 0.1317 Sharpe Ratio -0.7123 0.3330 0.9221 -0.2716 0.2972 0.6279 Skewness -0.2559 -1.2190 -0.3448 -1.4921 -1.1316 -0.1276 Kurtosis 9.0021 12.0045 5.5877 13.3250 9.0604 4.0098 Portfolio Turnover 1.1994 1.6602 1.4298 1.5098 2.1900 1.8499 Net Return 0.1383 0.0795

52 Table 9 Commodity-specified risk adjustments

This table reports regression results on the (Bakshi et al., 2017) three-factor model. Panels A, B and C present the regression results of long, short and long-short portfolio, respectively. We report results on mth nearest to maturity contracts, where m=1 and 3. AVG is an equally-weighted portfolio of 30 commodities in our sample. CARRY is constructed by taking long positions in the five most backwardated commodities and short positions in the five most contangoed commodities. MOM is constructed by taking long positions in past 12-month winner commodities and short positions in past loser commodities. The t-statistics (in parentheses) are estimated using the Newey and West (1987) procedure (*** p<0.01, ** p<0.05, * p<0.1). The sample period covers February 2004 through to May 2017. Term Structure Hedgers' HP Momentum Time-series Momentum Volatility m=1 m=3 m=1 m=3 m=1 m=3 m=1 m=3 m=1 m=3 Panel A: Long Portfolio AVG 1.211*** 1.252*** 0.866*** 0.881*** 1.163*** 1.117*** 1.022*** 0.870*** 0.983*** 1.115*** (7.25) (10.37) (23.31) (21.10) (9.47) (9.81) (16.27) (22.94) (15.17) (9.90) CARRY -0.014 -0.085** 0.190*** 0.231** 0.091 0.029 -0.101* 0.255** (-0.40) (-2.58) (2.75) (2.41) (1.55) (0.67) (-1.89) (2.37) MOM 0.248*** 0.200*** 0.026 0.023 0.352*** 0.303*** 0.089 -0.062 (3.44) (3.85) (0.56) (0.66) (7.94) (8.98) (1.59) (-0.94) Constant 0.004 0.005** 0.004** -0.001 0.010** 0.007* -0.001 0.000 0.006** -0.003 (1.56) (2.48) (1.99) (-0.77) (2.51) (1.88) (-0.46) (0.11) (2.17) (-1.12) Observations 148 146 148 146 148 146 148 144 124 122 Adj. R2 0.733 0.847 0.689 0.769 0.608 0.680 0.828 0.853 0.657 0.803

Panel B: Short Portfolio AVG 0.740*** 0.768*** 1.244*** 0.933*** 0.897*** 0.927*** 0.873*** 0.883*** 0.531*** 0.676*** (5.98) (6.54) (14.95) (17.14) (11.29) (12.04) (25.20) (24.82) (5.10) (7.24) CARRY 0.090 -0.145** -0.149*** -0.140** -0.031 -0.016 -0.137** -0.242*** (1.62) (-1.99) (-3.55) (-2.38) (-1.45) (-0.32) (-2.07) (-2.93) MOM -0.108*** -0.091** -0.062 0.060 -0.240*** -0.240*** 0.066 0.108* (-3.44) (-2.18) (-1.63) (1.12) (-5.00) (-6.02) (1.01) (1.86) Constant -0.003 -0.003 -0.006** 0.002 -0.004* -0.004* -0.000 0.001 -0.004 0.002 (-1.61) (-1.38) (-2.58) (0.82) (-1.84) (-1.69) (-0.12) (0.65) (-1.24) (1.13) Observations 148 146 148 146 148 146 148 145 124 122 Adj. R2 0.563 0.653 0.753 0.700 0.609 0.697 0.821 0.850 0.409 0.612

Panel C: Long-Short Portfolio AVG 0.471* 0.484** -0.378*** -0.051 0.267 0.190 0.149* -0.020 0.452*** 0.439** (1.67) (2.10) (-4.46) (-0.73) (1.47) (1.07) (1.94) (-0.48) (3.55) (2.31) CARRY -0.104 0.060 0.339*** 0.371** 0.121** 0.040 0.036 0.497*** (-1.38) (0.67) (3.48) (2.56) (2.10) (0.83) (0.38) (2.82) MOM 0.356*** 0.291*** 0.088 -0.037 0.592*** 0.550*** 0.023 -0.171 (4.35) (3.79) (1.26) (-0.56) (11.71) (16.04) (0.24) (-1.64) Constant 0.007** 0.008** 0.010*** -0.003 0.014** 0.010** -0.000 -0.001 0.009** -0.006 (2.00) (2.33) (2.68) (-1.04) (2.50) (2.06) (-0.17) (-0.60) (2.20) (-1.31) Observations 148 146 148 146 148 146 148 146 124 122 Adj. R2 0.310 0.355 0.110 -0.013 0.138 0.131 0.669 0.629 0.173 0.407

53 Table 10 Standard risk adjustments

This table reports the regression results on standard risk adjustments. Panels A, B and C present the regression results of long, short and long-short portfolio, respectively. We report results on mth nearest to maturity contracts, where m=1 and 3. STOCK and BOND denote Chinese equity and bond markets, proxied by CSI 300 and Barclays China aggregate index, respectively. AVG is an equally-weighted portfolio of 30 commodities in our sample. FX denotes returns on the RMB effective exchange rate index. INFSHOCK and UIP denote unexpected inflation rate and unexpected industrial production in China, computed as the difference between the actual and consensus figures from Bloomberg. The t-statistics (in parentheses) are estimated using the Newey and West (1987) procedure (*** p<0.01, ** p<0.05, * p<0.1). The sample period covers February 2004 through to May 2017.

Term Structure Hedgers' HP Momentum Time-series Momentum Volatility m=1 m=3 m=1 m=3 m=1 m=3 m=1 m=3 m=1 m=3 Panel A: Long Portfolios STOCK 0.028 0.040** -0.005 0.002 -0.017 -0.025 0.003 0.006 -0.022 0.003 (0.98) (2.02) (-0.28) (0.11) (-0.42) (-0.59) (0.09) (0.25) (-0.59) (0.11) BOND 0.025 0.069 0.245 0.031 0.664 1.034* 0.196 0.188 0.001 -0.125 (0.05) (0.17) (0.93) (0.11) (1.54) (1.93) (0.61) (0.73) (0.00) (-0.25) AVG 1.256*** 1.283*** 0.893*** 0.862*** 1.220*** 1.253*** 1.151*** 0.993*** 1.041*** 1.162*** (7.58) (11.51) (16.90) (17.05) (9.63) (12.78) (11.90) (15.96) (12.82) (7.17) INFSHOCK 0.449 0.235 0.166 -0.229 0.987 0.111 1.202 -0.063 -0.284 0.361 (0.44) (0.32) (0.32) (-0.46) (0.90) (0.16) (1.17) (-0.10) (-0.40) (0.41) FX -0.064 0.088 -0.057 -0.048 -0.252 -0.057 -0.224 -0.086 0.120 -0.299 (-0.27) (0.44) (-0.38) (-0.39) (-0.71) (-0.16) (-0.85) (-0.48) (0.80) (-1.34) UIP 0.116 0.093 -0.271** -0.221** 0.151 -0.017 0.041 -0.217 -0.335** 0.115 (0.65) (0.64) (-2.46) (-1.99) (0.56) (-0.10) (0.20) (-1.18) (-2.49) (0.53) Constant 0.009** 0.007** 0.003 -0.002 0.011*** 0.006 0.007*** 0.004* 0.005** -0.000 (2.47) (2.58) (1.38) (-1.08) (2.81) (1.49) (2.68) (1.69) (1.99) (-0.01) Observations 148 146 148 146 148 146 148 144 124 122 Adj. R2 0.675 0.814 0.696 0.763 0.585 0.661 0.679 0.731 0.650 0.773

Panel B: Short Portfolios STOCK -0.025 -0.030 0.012 -0.057** 0.001 0.017 -0.017 -0.012 -0.018 -0.051* (-1.07) (-1.50) (0.50) (-2.05) (0.03) (0.67) (-0.96) (-0.64) (-0.73) (-1.93) BOND -0.297 -0.146 0.065 0.239 -0.479 -0.338 -0.616 -0.761 -0.251 0.114 (-0.77) (-0.37) (0.22) (0.61) (-0.99) (-0.97) (-1.03) (-1.50) (-0.83) (0.33) AVG 0.720*** 0.763*** 1.237*** 0.947*** 0.791*** 0.823*** 0.770*** 0.777*** 0.511*** 0.668*** (6.43) (7.28) (14.34) (11.52) (9.72) (11.37) (16.08) (12.23) (4.06) (4.87) INFSHOCK 0.112 0.522 -0.216 0.351 -0.052 0.139 -0.265 0.003 0.184 -0.086 (0.19) (1.19) (-0.37) (0.56) (-0.07) (0.26) (-0.44) (0.01) (0.23) (-0.13) FX 0.034 0.115 0.034 0.037 -0.013 -0.058 0.051 -0.104 -0.031 0.154 (0.21) (0.63) (0.21) (0.26) (-0.08) (-0.31) (0.42) (-0.75) (-0.18) (1.06) UIP -0.063 0.040 0.236* 0.076 0.363** 0.251* 0.052 0.009 -0.189* -0.027 (-0.47) (0.28) (1.88) (0.48) (2.53) (1.67) (0.53) (0.08) (-1.72) (-0.19) Constant -0.004* -0.003 -0.006*** 0.001 -0.005* -0.004* -0.003 0.000 -0.004 0.001 (-1.84) (-1.44) (-2.73) (0.28) (-1.72) (-1.72) (-1.04) (0.17) (-1.36) (0.25) Observations 148 146 148 146 148 146 148 145 124 122 Adj. R2 0.530 0.636 0.746 0.690 0.597 0.683 0.675 0.747 0.366 0.539

54 Panel C: Long-Short Portfolios STOCK 0.053 0.070** -0.017 0.059 -0.018 -0.042 0.020 0.014 -0.003 0.054 (1.30) (2.41) (-0.48) (1.36) (-0.32) (-0.77) (0.46) (0.36) (-0.09) (1.34) BOND 0.322 0.215 0.179 -0.208 1.143 1.372* 0.812 0.913 0.252 -0.239 (0.41) (0.29) (0.35) (-0.45) (1.44) (1.68) (1.00) (1.46) (0.51) (-0.35) AVG 0.535** 0.520*** -0.344*** -0.085 0.429** 0.430*** 0.381*** 0.209* 0.531*** 0.495* (2.05) (2.66) (-3.74) (-0.96) (2.40) (2.75) (2.84) (1.89) (3.71) (1.70) INFSHOCK 0.337 -0.287 0.382 -0.580 1.039 -0.028 1.467 -0.053 -0.469 0.447 (0.24) (-0.30) (0.46) (-0.61) (0.64) (-0.03) (1.02) (-0.06) (-0.46) (0.45) FX -0.098 -0.028 -0.091 -0.085 -0.239 0.001 -0.275 0.021 0.152 -0.453 (-0.29) (-0.09) (-0.34) (-0.37) (-0.50) (0.00) (-0.78) (0.07) (0.77) (-1.48) UIP 0.178 0.052 -0.507** -0.298 -0.212 -0.268 -0.011 -0.264 -0.146 0.142 (0.69) (0.22) (-2.55) (-1.32) (-0.58) (-0.92) (-0.04) (-0.88) (-0.89) (0.54) Constant 0.013*** 0.011** 0.010** -0.002 0.016*** 0.010* 0.010** 0.004 0.009** -0.001 (2.68) (2.39) (2.33) (-0.84) (2.70) (1.71) (2.03) (0.84) (2.23) (-0.13) Observations 148 146 148 146 148 146 148 146 124 122 Adj. R2 0.164 0.243 0.112 -0.007 0.063 0.070 0.118 0.040 0.158 0.262

55 Table 11 Liquidity and behavioural factors

This table reports factor loadings of strategy returns on liquidity and behavioural factors. Panel A reports regression results on Amihud illiquidity (AI), computed as the ratio of monthly return to RMB volume in absolute term based on the front-contract. The RMB volume is scaled to the nearest 1 million RMB. Panel B reports loadings on the 52-week high momentum (HMOM), a proxy of anchoring bias as proposed by Bianchi et al. (2016). The 52-week high momentum signal is computed as the ratio of current price to the highest price in the past 12 months. HMOM is constructed by taking long positions in commodities that are nearest to their 52-week highs and short positions in commodities which are furthest away from their 52-week highs. We report results on mth nearest to maturity contracts, where m=1 and 3. The t-statistics (in parentheses) are estimated using the Newey and West (1987) procedure (*** p<0.01, ** p<0.05, * p<0.1). The sample period covers February 2004 through to May 2017. Term Structure Hedgers' HP Momentum Time-series Momentum Volatility m=1 m=3 m=1 m=3 m=1 m=3 m=1 m=3 m=1 m=3 Panel A: Amihud illiquidity factor Long Portfolios AI -0.387 -0.245 -0.199 -0.180 -0.360 -0.298* -0.398* -0.050 -0.349* -0.085 (-1.54) (-1.09) (-1.13) (-0.89) (-1.29) (-1.76) (-1.88) (-0.32) (-1.94) (-0.35) Constant 0.010* 0.008 0.003 -0.002 0.015** 0.011* 0.010* 0.004 0.008 -0.002 (1.75) (1.28) (0.74) (-0.56) (2.14) (1.71) (1.82) (0.87) (1.36) (-0.25) Observations 159 157 158 156 148 146 148 144 124 122 R2 0.012 0.004 0.007 0.005 0.010 0.006 0.016 0.000 0.016 0.001 Short Portfolios AI -0.033 0.172 -0.366 0.062 -0.229 0.055 -0.098 0.083 -0.047 -0.087 (-0.16) (0.71) (-1.63) (0.27) (-1.20) (0.18) (-0.53) (0.31) (-0.22) (-0.30) Constant -0.007** -0.007* -0.004 -0.001 -0.005 -0.006 -0.006 -0.004 -0.005 0.001 (-1.99) (-1.71) (-0.94) (-0.23) (-1.13) (-1.22) (-1.59) (-0.85) (-1.37) (0.24) Observations 159 157 158 156 148 146 148 145 124 122 R2 0.000 0.005 0.013 0.001 0.009 0.000 0.002 0.001 0.001 0.002 L-S Portfolios AI -0.354* -0.417** 0.166 -0.242** -0.131 -0.353 -0.300* -0.131 -0.302* 0.002 (-1.90) (-2.41) (0.85) (-2.59) (-0.57) (-1.28) (-1.82) (-0.67) (-1.74) (0.01) Constant 0.017*** 0.015*** 0.007** -0.001 0.020*** 0.017*** 0.015*** 0.008** 0.013*** -0.003 (2.90) (3.28) (1.99) (-0.39) (2.62) (3.04) (3.39) (2.35) (2.65) (-0.43) Observations 159 157 158 156 148 146 148 146 124 122 R2 0.012 0.019 0.004 0.011 0.002 0.011 0.012 0.003 0.014 0.000

Panel B: Behavioural factor Long Portfolios HMOM -0.142 -0.165 -0.145 -0.123 0.080 0.200 -0.026 0.042 -0.237* -0.438 (-0.75) (-0.72) (-1.25) (-1.08) (0.45) (0.90) (-0.13) (0.23) (-1.89) (-1.58) Constant 0.007 0.008* 0.003 -0.001 0.009* 0.006 0.005 0.003 0.005 -0.000 (1.48) (1.89) (0.91) (-0.43) (1.85) (1.40) (1.17) (0.93) (1.05) (-0.02) Observations 147 145 147 145 147 145 147 143 124 122 R2 0.018 0.023 0.044 0.029 0.006 0.032 0.001 0.002 0.070 0.138 Short Portfolios HMOM -0.139** -0.109 -0.285* -0.169* -0.340*** -0.420*** -0.288*** -0.318*** -0.057 -0.035 (-2.52) (-1.49) (-1.85) (-1.79) (-3.87) (-3.81) (-4.19) (-2.92) (-1.13) (-0.47) Constant -0.006 -0.004 -0.006 0.002 -0.005 -0.002 -0.005 -0.000 -0.006 0.000 (-1.59) (-0.97) (-1.61) (0.40) (-1.39) (-0.48) (-1.44) (-0.10) (-1.63) (0.07) Observations 147 145 147 145 147 145 147 144 124 122 R2 0.049 0.025 0.087 0.045 0.217 0.286 0.217 0.215 0.009 0.003 L-S Portfolios HMOM -0.002 -0.057 0.139* 0.047 0.420*** 0.619*** 0.262* 0.360*** -0.180 -0.402 (-0.01) (-0.30) (1.81) (0.66) (3.24) (4.70) (1.88) (4.22) (-1.41) (-1.63) Constant 0.013*** 0.012*** 0.009*** -0.003 0.014*** 0.008** 0.009*** 0.004 0.010*** -0.000 (2.81) (3.55) (2.98) (-1.06) (3.42) (2.43) (3.39) (1.54) (2.78) (-0.08) Observations 147 145 147 145 147 145 147 145 124 122 R2 0.000 0.004 0.032 0.005 0.171 0.382 0.099 0.262 0.049 0.154

56 Table 12 Market sentiment

This table reports the regression results of strategy returns on market sentiment. Panel A reports the results on OVX, which denotes the CBOE crude oil volatility index that measures the market expectation of 30-day volatility of crude oil prices. Panel B reports loadings on CHIX, which denotes the AlphaShares Chinese volatility index that measures the implied volatility of options on the FTSE Xinhua China 25 and Hang Seng indices. We report results on mth nearest to maturity contracts, where m=1 and 3. The t-statistics (in parentheses) are estimated using the Newey and West (1987) procedure (*** p<0.01, ** p<0.05, * p<0.1). The sample period covers February 2004 through to May 2017. Term Structure Hedgers' HP Momentum Time-series Momentum Volatility m=1 m=3 m=1 m=3 m=1 m=3 m=1 m=3 m=1 m=3 Panel A: OVX factor Long Portfolios OVX -0.079 -0.090* -0.057** -0.085*** -0.054 -0.067* -0.072* -0.080** -0.073** -0.097 (-1.49) (-1.66) (-2.01) (-3.30) (-1.50) (-1.71) (-1.78) (-2.23) (-2.30) (-1.55) Constant 0.002 0.003 0.002 -0.003 0.006 0.004 0.003 0.002 0.004 -0.003 (0.36) (0.46) (0.43) (-0.88) (1.00) (0.77) (0.62) (0.45) (0.76) (-0.47) Observations 120 118 120 118 120 118 120 116 120 118 R2 0.045 0.053 0.052 0.100 0.024 0.035 0.044 0.068 0.066 0.065 Short Portfolios OVX -0.074*** -0.067*** -0.119** -0.070*** -0.102*** -0.100*** -0.099*** -0.102*** -0.034** -0.039** (-4.45) (-3.10) (-2.54) (-2.68) (-3.50) (-2.81) (-3.43) (-3.04) (-2.24) (-2.18) Constant -0.007* -0.005 -0.009 -0.000 -0.009* -0.005 -0.007* -0.004 -0.005 0.000 (-1.78) (-1.18) (-1.59) (-0.04) (-1.85) (-1.04) (-1.81) (-0.84) (-1.57) (0.06) Observations 120 118 120 118 120 118 120 117 120 118 R2 0.109 0.071 0.114 0.066 0.142 0.110 0.185 0.152 0.032 0.031 L-S Portfolios OVX -0.005 -0.023 0.061** -0.015 0.048 0.033 0.027 0.021 -0.039 -0.057 (-0.11) (-0.54) (2.01) (-0.97) (1.53) (1.28) (0.87) (0.85) (-1.32) (-1.02) Constant 0.009 0.008* 0.010*** -0.003 0.015*** 0.009** 0.010*** 0.006* 0.009** -0.003 (1.62) (1.76) (2.87) (-1.08) (2.78) (2.15) (2.72) (1.81) (2.31) (-0.60) Observations 120 118 120 118 120 118 120 118 120 118 R2 0.000 0.006 0.052 0.005 0.026 0.012 0.011 0.008 0.022 0.031

Panel B: CHIX factor Long Portfolios CHIX -0.070** -0.067* -0.040** -0.040* -0.043 -0.047 -0.042 -0.052 -0.064*** -0.097** (-2.09) (-1.95) (-2.09) (-1.93) (-1.36) (-1.30) (-1.42) (-1.63) (-3.18) (-2.21) Constant 0.006 0.006 0.001 -0.004 0.010* 0.008 0.005 0.004 0.002 -0.004 (0.98) (1.13) (0.17) (-1.08) (1.71) (1.55) (1.00) (0.95) (0.52) (-0.56) Observations 159 157 158 156 148 146 148 144 124 122 R2 0.052 0.044 0.038 0.037 0.019 0.021 0.023 0.042 0.073 0.098 Short Portfolios CHIX -0.044*** -0.050*** -0.073** -0.059*** -0.046** -0.059*** -0.039** -0.044* -0.023* -0.038*** (-3.36) (-2.97) (-2.39) (-2.93) (-2.29) (-2.88) (-2.05) (-1.97) (-1.96) (-2.92) Constant -0.008** -0.005 -0.009* -0.001 -0.008** -0.006 -0.007** -0.003 -0.006* -0.000 (-2.30) (-1.50) (-1.91) (-0.19) (-2.01) (-1.36) (-1.98) (-0.81) (-1.84) (-0.08) Observations 159 157 158 156 148 146 148 145 124 122 R2 0.057 0.059 0.070 0.063 0.047 0.066 0.047 0.047 0.022 0.043 L-S Portfolios CHIX -0.026 -0.017 0.033* 0.018 0.002 0.012 -0.004 -0.007 -0.041** -0.060 (-1.01) (-0.78) (1.66) (1.32) (0.09) (0.43) (-0.19) (-0.41) (-2.05) (-1.42) Constant 0.013** 0.011*** 0.009*** -0.003 0.018*** 0.014*** 0.012*** 0.007** 0.008** -0.003 (2.59) (2.83) (2.87) (-1.14) (3.32) (2.93) (3.28) (2.22) (2.29) (-0.62) Observations 159 157 158 156 148 146 148 146 124 122 R2 0.008 0.004 0.021 0.009 0.000 0.002 0.000 0.001 0.035 0.049

57 Table A1 Data-snooping test

This table reports results on the Reality Check (White, 2000) and SPA (Hansen, 2005) test. The parameter q is the geometric distribution that determines the block-length in the bootstrap process, where the block-length is computed as 1/q. For each test, the bootstrap is replicated 1000 times. The stationary and circular bootstraps are based on Politis and Romano (1994) and Politis and Romano (1992) respectively. Only the five statistically profitable strategies are tested. Panel A reports all strategies as a group against a zero-mean return benchmark. Panel B reports all strategies against the passive long-only benchmark (AVG), an equally weighted portfolio of 30 commodities in the sample. Panels C through to G report results of each strategy against AVG. Significant p-values indicate that the strategy outperforms the benchmark.

Bootstrap Bootstrap Reality check SPA test dependence method Consistent p-values Consistent p-values Panel A: All strategies versus zero-mean-return benchmark q=0.05 Stationary 0.0010 0.0020 Circular 0.0000 0.0000 q=0.1 Stationary 0.0020 0.0050 Circular 0.0000 0.0020 q=0.5 Stationary 0.0020 0.0020 Circular 0.0000 0.0000 Panel B: All strategies versus AVG q=0.05 Stationary 0.0010 0.0010 Circular 0.0020 0.0000 q=0.1 Stationary 0.0050 0.0000 Circular 0.0030 0.0010 q=0.5 Stationary 0.0020 0.0010 Circular 0.0050 0.0040 Panel C: Term structure strategy versus AVG q=0.05 Stationary 0.0000 0.0000 Circular 0.0000 0.0000 q=0.1 Stationary 0.0000 0.0000 Circular 0.0000 0.0000 q=0.5 Stationary 0.0000 0.0000 Circular 0.0000 0.0000 Panel D: Hedgers' hedging pressure strategy versus AVG q=0.05 Stationary 0.0250 0.0220 Circular 0.0270 0.0230 q=0.1 Stationary 0.0410 0.0330 Circular 0.0380 0.0370 q=0.5 Stationary 0.0630 0.0490 Circular 0.0610 0.0650 Panel E: Momentum strategy versus AVG q=0.05 Stationary 0.0000 0.0000 Circular 0.0000 0.0000 q=0.1 Stationary 0.0000 0.0000 Circular 0.0000 0.0000 q=0.5 Stationary 0.0000 0.0000 Circular 0.0000 0.0000 Panel F: Time-series momentum strategy versus AVG q=0.05 Stationary 0.0000 0.0010 Circular 0.0010 0.0000 q=0.1 Stationary 0.0010 0.0020 Circular 0.0020 0.0000 q=0.5 Stationary 0.0010 0.0010 Circular 0.0010 0.0000 Panel G: Volatility strategy versus AVG q=0.05 Stationary 0.0000 0.0000 Circular 0.0000 0.0000 q=0.1 Stationary 0.0000 0.0000 Circular 0.0010 0.0010 q=0.5 Stationary 0.0020 0.0000 Circular 0.0010 0.0020

58 7 250 No. of contracts (total) No. of contracts (1st-4th nearest) Trading volume (RMB)

6 200

5

150 4

3

100

RMB (Trillions) RMB No. of contracts (Billions) contracts of No. 2

50 1

0 0

Figure 1 Annual trading volumes

This figure illustrates the annual trading volume of commodity futures in China. The line plot exhibits the total trading volume (expressied in trillions of RMB) of 48 commodity futures contracts which progressively enterred the market. The solid bar plot represents the total number of contracts traded across the futures curve based on our sample of 30 commodities wheraeas the patterned bars plot the number of contracts from 1st to 4th nearest to maturity contracts. Data on annual RMB trading volume (turnover) are obtaiend from the China Futures Association.

59

Figure 2 Performance of passive long-only investments

This figure illustrates the cumulative performance of the passive long-only commodity portfolios in China. Panel A exhibits the broad market performance (AVG). Panels B through to F illustrate the performance of industrials, metals, grains, oilseeds and energy sectors, respectively. Within each panel, we construct equal-weighted, open interest and volume weighted portfolios, respectively. The unit value index (UVI) at time t is computed as UVIt

= UVIt-1 (1+ Rt), with an initial value of 100. Rt denotes the index return at time t. For equally-weighted portfolio, 1 Rt = ∑ 푟 where N is the number of commodities at time t and 푟 is the return of commodity i at time t. For open 푁 푡,푖 푡,푖 푎푡,푖 interest/volume weighted portfolio, the Rt = ∑ 푟푡,푖 where 푎푡,푖 denotes the open interest/volume of commodity i at 퐴푡 time t and 퐴푡 is the open interest of all commodities at time t. All the portfolios are constructed based on the nearest to maturity contracts. The sample period covers 2004 through to 2017.

60 AVG TERM HHP MOM TSMOM VOLA 11.89

12

9

6.50

6

4.88

3

1

0

2004 2006 2009 2012 2014 2017

Figure 3 Cumulative returns

This figure illustrates the future value of 1 RMB invested in long-only and long-short strategies. AVG denotes the equally-weighted portfolio of 30 commodities. To construct strategy portfolios, we sort commodities into quartiles based on the respective signal. At the end of each month, we take long positions in commodities within the top quartile and short positions in the bottom quartile. All portfolios are equally weighted and rebalanced monthly. The term structure signal (TERM) is based on the roll-yield from the front to the next nearest contract. Hedgers’ hedging pressure (HHP) strategy sorts commodities based on hedging and speculation ratio introduced by Bohl et al. (2018). The hedging ratio is the change of open interest divided by volume, while the speculation ratio is volume divided by open interest. Momentum (MOM) and time-series momentum (TSMOM) strategies employ past 12-month returns as the sorting signal. The volatility (VOLA) signal employed is the coefficient of variation of prior 3-year daily return. The sample period covers February 2004 through to May 2017.

61

Figure 4 Recursive performance

The figure illustrates the recursive monthly mean returns, standard deviations and Sharpe ratios of long-only and long-short strategy portfolios. The recursive statistics are computed on a monthly basis with an initial window of 12 months. The ending data point illustrates the full period performance. AVG denotes the equally-weighted portfolio of 30 commodities. TERM denotes the term structure strategy, HHP represents the hedgers’ hedging pressure. MOM and TSMOM denote the cross-sectional and time-series momentum strategies, respectively. VOLA denotes the volatility strategy.

62 50% AVG, 42.2% AVG TERM HHP MOM TSMOM VOLA 40%

TERM, 26.4% 30% VOLA, 20.2% TSMOM, 18.0% 20% TSMOM, 15.8% TERM, 6.0% MOM, 9.7% HHP, 12.0% MOM, 8.8% MOM, 8.9% 10% AVG, 5.2% VOLA, 8.8% TERM, 5.7% TSMOM, 3.7% 0%

-10% HHP, -7.7% AVG, -6.4%

-20% VOLA, -13.4% -30% HHP, -27.9% -40% Figure 5 Correlations

This figure illustrates the pairwise correlations between strategy returns and equity, bond returns and changes in unexpected inflation. The inflation shock is the difference between actual and forecasted inflation estimated by Bloomberg. The CSI 300 consists of top 300 stocks traded on the Shanghai and Shenzhen stock exchanges. The Barclays China Aggregate Index covers fixed-rate treasury, government and corporate bonds. AVG denotes the equally-weighted portfolio of 30 commodities. TERM denotes the term structure strategy, HHP represents the hedgers’ hedging pressure. MOM and TSMOM denote the cross-sectional and time-series momentum strategies, respectively. VOLA denotes the volatility strategy. The sample period covers February 2004 through to May 2017. The bold numbers indicate significance at 5% level or better.

100% TERM HHP MOM TSMOM VOLA

80%

60%

40%

20%

0%

-20%

-40%

-60%

-80%

Figure 6 Time-varying correlation

This figure illustrates the dynamic correlations between long-short strategies returns and the CSI 300 index. The correlations are estimated using the ADCC-GARCH (1,1) model. TERM denotes the term structure strategy, HHP represents the hedgers’ hedging pressure. MOM and TSMOM denote the cross-sectional and time-series momentum strategies, respectively. VOLA denotes the volatility strategy. The shaded areas represent periods of market stress. The sample period covers February 2004 through to May 2017.

63 4%

3%

2%

1%

0%

-1%

-2%

AVG TERM HHP MOM TSMOM VOLA -3% Figure 7 Extreme market conditions

This figure demonstrates the performance of long-only and long-short strategies during different market conditions. We classify market returns into quintiles using the CSI 300. The crisis period is the phrase when the stock market delivers the lowest average monthly return (the lowest quintile), while the growth period has the highest average monthly return (the highest quintile). AVG denotes the equally-weighted portfolio of 30 commodities. TERM denotes the term structure strategy, HHP represents the hedgers’ hedging pressure. MOM and TSMOM denote the cross-sectional and time-series momentum strategies, respectively. VOLA denotes the volatility strategy. The sample period covers February 2004 through to May 2017.

Panel A: Long portfolio Panel B: Short porfolio 40% 40%

30% 30%

20% 20%

10% 10%

0% 0% energy grains industrial metal oilseeds energy grains industrial metal oilseeds Figure 8 Percentage of total trades

This figure illustrates the percentage of total trades of long-short strategies. Panel A exhibits long portfolios whereas Panel B exhibits short portfolios. TERM denotes the term structure strategy, HHP represents the hedgers’ hedging pressure. MOM and TSMOM denote the cross-sectional and time-series momentum strategies, respectively. VOLA denotes the volatility strategy. The sample period covers February 2004 through to May 2017.

64