International Review of Economics and Finance 61 (2019) 156–169

Contents lists available at ScienceDirect

International Review of Economics and Finance

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

The time-varying spillover effect between WTI crude oil futuresreturns and hedge funds

Yue-Jun Zhang a,b,*, Yao-Bin Wu c

aBusiness School, Hunan University, Changsha, 410082, ChinabCenter for Resource and Environmental Management, Hunan University, Changsha, 410082, Chinac School of Management, Fudan University, Shanghai, 200433, China

A R T I C L E I N F O

JEL classification:Q01G15O16

Keywords:Crude oil futuresDCOT reportsHedge fundsTime-varying granger causality

* Corresponding author. Center for Resource anE-mail address: zyjmis@126.com (Y.-J. Zhang)

https://doi.org/10.1016/j.iref.2019.02.006Received 18 April 2018; Received in revised formAvailable online 15 February 20191059-0560/© 2019 Elsevier Inc. All rights reserv

A B S T R A C T

To capture the relationship change between WTI crude oil returns and hedge funds in recent years,the linear and nonlinear Granger causality test approaches and Hong's time-varying informationspillover statistics are employed based on the data from 2006 to 2017. The empirical resultsindicate that, first, hedge funds' net positions may Granger cause crude oil futures returns in thelinear manner but the nonlinear relationship appears weaker. Second, both in the linear andnonlinear manners, the influence of crude oil futures prices has been evidently strengthened whenthey experience high volatility whilst the influence of hedge funds' net positions appears strongeronly when crude oil futures prices rise substantially. Finally, hedge funds play an important role inpushing up crude oil price bubbles which are responsible for the crash of crude oil prices in 2008,but the crash of oil prices in 2014 is more attributed to market fundamentals.

1. Introduction

In the past two decades, the pace of commodity financialization has been constantly accelerated (Zhang, Chevallier, & Guesmi,2017). The development of technology and the need for investment have made financial products based on commodities a gooddiversification tool for investors (Gkanoutas-Leventis & Nesvetailova, 2015). Meanwhile, the regulation has become more permissivesince the launch of the Commodities Future Modernization Act (CFMA) in 2000. As a result, institutional investors have built theirpositions rapidly in the commodity futures markets since 2004 (Basak & Pavlova, 2016), and a large number of institutional investorshave held commodities through commodity index futures. In fact, financialization provides more opportunities to spread risks duringthe periods of economic ascent but also multiplies the systematic risks during the descent (Gkanoutas-Leventis & Nesvetailova, 2015;Yu, Du, Fang, & Yan, 2018).

The last decade has witnessed striking volatility in commodities prices, of which crude oil is undoubtedly one of the most importantproducts. Since 2003, crude oil prices have experienced violent fluctuations. The USWest Texas Intermediate (WTI) crude oil spot pricesreached the highest at $145 per barrel in July 2008. Meanwhile, hedge funds’ share grew rapidly in global crude oil futures marketaccounting for $260 billion, which was nearly 20 times that in 2003 (Cho, 2008). However, WTI crude oil spot prices fell to $30 perbarrel in late 2008 due to the global financial crisis, but after 2009, WTI crude oil prices have experienced a gradual rising trend.However, after the second half of 2014, U.S. shale producers increased production whilst traditional oil producing countries such asLibya and Iraq began to resume production, which resulted in the supply of crude oil significantly exceeding demand (OPEC, 2014).

d Environmental Management, Hunan University, Changsha, 410082, China..

9 February 2019; Accepted 13 February 2019

ed.

Y.-J. Zhang, Y.-B. Wu International Review of Economics and Finance 61 (2019) 156–169

Additionally, Volcker Rule faced execution and U.S. dollar experienced continuous appreciation. As a result, WTI crude oil pricesdropped sharply and reached a record low price at $26 per barrel after the global financial crisis, in early 2016. With the limit crude oilproduction of OPEC and Russia, WTI crude oil prices have been on an upward trend afterwards.

As we all know, international crude oil market proves a typical complex system, where it is difficult to capture market changes simplybased on the supply and demand situation (Cifarelli & Paladino, 2010; Gallo, Mason, Shapiro, & Fabritius, 2010; Zhang &Wang, 2019;Zhang & Zhang, 2018). According to the statistics of the Commodity Futures Trading Commission (CFTC), the non-commercials held53% positions of crude oil futures and options in the New York Mercantile Exchange (NYME) by the end of 2016.1 Along with theincreasing proportion, financial positions aiming to speculate without physical support have increasing influence on crude oil prices (Li,Kim, & Park, 2016), of which hedge funds constitute the major part (Ding, Kim, & Park, 2014). Therefore, we also have to place muchemphasis on the financial positions.

Although there have beenmany studies on the relationship between crude oil prices and financial positions, most have regarded non-commercial positions as speculators' positions, which may lead to a misunderstanding of the speculators’ influence on crude oil prices.The judgment of speculators should be based on trading behaviors rather than registration (Kaufmann, 2011). Non-commercialsconstitute a part of speculators; however, speculative trading activities also exist in commercials (Ederington & Lee, 2002; Irwin &Sanders, 2010). Additionally, existing literature mainly focuses on the period of global financial crisis in 2008 but there is a lack ofstudies on the sharp fluctuations in crude oil prices observed since 2014. Meanwhile, the mainmethods used in previous relevant studieshave indicated the relationship in the mean and static sense rather than considering the time-varying characteristics, which may ignorethe relationship changes over time.

Under these circumstances, this paper focuses on the relationship between crude oil futures returns and hedge funds' net positions,which are the largest part of the non-commercial positions, and explores its time-varying characteristics over the last decade. Specif-ically, we first detect the general relationship from June 2006 to December 2017 using the Granger causality test both in linear andnonlinear forms. Then, the linear time-varying characteristics between crude oil futures returns and hedge funds' net positions areexplored using Hong's time-varying information spillover statistics. The rolling window is also introduced into DP Granger test toinvestigate their nonlinear time-varying characteristics. Through this study we hope to offer helpful information for investors, analystsand related market regulators.

The main contribution of this paper consists of two aspects. On the one hand, the dynamic interaction between crude oil futures andhedge funds over the past decade is analyzed. On the other hand, the comparison is made for the causes of the crashes of crude oil futuresprices in 2008 and 2014 according to the information spillover strength between crude oil futures returns and hedge funds’ netpositions.

The remainder of the paper is organized as follows. Section 2 reviews related literature. Section 3 introduces empirical methods anddata definitions. Section 4 presents empirical results and analyses, and Section 5 concludes the paper.

2. Literature review

As the main representative of commodities, crude oil has been widely studied by academia. In the 21st century, international crudeoil prices began to deviate from the fundamentals and bubbles gradually grew. Lammerding, Trede, and Wilfling (2013) introduce twoMarkov-regimes into the state-space representation to investigate the bubbles of WTI crude oil prices and find that they have existedsignificantly since 2005 and reached the maximum in mid-2008. Similarly, Zhang and Yao (2016) find that crude oil prices aresignificantly driven by bubbles from the end of 2001 to mid-2008, using the state-space model. Zhang and Wang (2015) also claim theexistence of significant bubbles in WTI crude oil prices during 2003–2012, and the price dynamics cannot be fully explained by thesupply and demand situation.

The existence of crude oil price bubbles implies strong impact of financial investors on crude oil prices over the past decade.2 DeLong, Shleifer, Summers, and Waldmann (1990) develop a bubble model to reveal the economic mechanisms by which speculation canaffect prices. Using these mechanisms, Tokic (2011) shows that crude oil price bubbles in 2008 were caused by commercial arbitrageursand speculators jointly. According to the contribution of macro-economic variables and financial speculation to crude oil price fluc-tuations, Morana (2013) finds that macro-economic shocks have been the main upward driver since mid-1980s whereas the influence offinancial shocks began to rise after 2000s, and the crash of crude oil price in 2007-2008 was caused by macro-economy and financialspeculation jointly. Using endogenously determined break tests, Fan and Xu (2011) find that speculation and episodic events were themain drivers of crude oil price changes from 2000 to 2004. However, as the influence of speculation appeared more prominent with theinflow of large amounts of speculative funds from 2004 to 2008, the supply and demand became comparatively less important. Besides,Yang, Han, Hong, and Wang (2016) prove that speculative activities have significant explanatory power for the dynamics of oil pricesduring 2003–2011 with an interval regression model.

In both linear and nonlinear ways, Zhang (2013) states that speculation has a robust linear effect on crude oil prices but weaknonlinearity, and has significant linear shocks on crude oil prices during the periods of high volatility. Besides, Li et al. (2016) use theGNNTS model to examine the nonlinear influence of commercial and non-commercial net long positions on WTI crude oil prices, and

1 The data are derived from a weekly report by CFTC, which are available at: http://www.cftc.gov/MarketReports/CommitmentsofTraders/HistoricalCompressed/index.htm.2 https://www.ftc.gov/sites/default/files/documents/public_comments/prohibitions-market-manipulation-and-false-information-subtitle-b-

energy-independence-and-security/535819-00038.pdf. Accessed 7 February 2019.

157

Y.-J. Zhang, Y.-B. Wu International Review of Economics and Finance 61 (2019) 156–169

conclude that hedgers and speculators both contributed to crude oil price bubbles in 2008 and the effect of them has become strongersince 2011; and speculators tend to extend crude oil price fluctuations but hedgers do not.

It should be noted that the above studies all identify non-commercial positions as speculators’ positions. This classification is simpleand non-detailed. In fact, non-commercials do not cover all the speculators and commercials also conduct some speculative tradingactivities (Ederington & Lee, 2002; Irwin & Sanders, 2010). Meanwhile, non-commercials contain different types of traders whoserelationships with crude oil prices are also different. By examining the Granger-causality between the prices of WTI crude oil futures anddifferent types of net position changes for the period of 2000–2009, Buyuksahin and Harris (2011) argue that during this period,non-commercial net position changes are driven by price changes in crude oil futures, but not vice versa. In addition, they dividenon-commercials into hedge funds (including commodity pool operators, commodity trading advisors, associated persons controllingcustomer accounts, and other managed money traders) and floor brokers & traders, and find that hedge funds remain in line with theabove argument, but floor brokers & traders contribute to price changes in crude oil futures.

Different from other studies, Irwin and Sanders (2012) identify the CFTC's Index Investment Data (IID) as the index funds positions,and find that index funds are not the cause of price changes in commodity futures during 2007-2011. Afterwards, Sanders and Irwin(2014) prove again at the firm level that fund positions are unrelated to price movements in WTI crude oil futures market.

Among the above studies on the interaction analysis, the Granger causality test is the most common method, and a number of newtest methods have been developed since Granger (1969) proposes the concept of Granger causality. For example, Sims (1972) introducesthe VAR model, which is now a very user-friendly method. Based on the work of Sims (1972), the VECM, GARCH and other models arealso introduced into the Granger causality test. In addition, Haugh (1976) first constructs the statistic based on cross-correlation co-efficients to examine the dependence of two time series which implies Granger causality. Haugh's statistic advances the cumulativeeffect of long-time information spillover but cannot determine the effect in different time periods. Accordingly, Hong (2001) solves thisproblemwith the kernel function. It should be noted that the methods above are mainly applied to the linear relationship (Baek& Brock,1992; Hiemstra & Jones, 1993); however, financial time series generally exists significant nonlinear characteristics (Hsieh, 1989, 1991;Zhang, Yao, He, & Ripple, 2019). Against this backdrop, Baek and Brock (1992) propose the nonlinear Granger causality test, and thenHiemstra and Jones (1994) modify Baek and Brock's version based on the non-parametric estimators and apply it to the relationshipresearch on stock price-volume. Besides, the Hiemstra-Jones test has been used in crude oil market as well (Moosa & Silvapulle, 2000).To avoid the over-rejection in the Hiemstra-Jones test, Diks and Panchenko (2006) propose a new statistic to test the nonlinear rela-tionship between two series.

However, the methods described above almost only investigate the general relationship during the whole sample period in the meansense, which produces a constant value, but do not accurately consider the relationship over time. In fact, a few studies have made someattempt on this aspect. For instance, Sato et al. (2006) expand a wavelet dynamic vector auto-regressive (DVAR) approach and apply it tothe time-varying Granger causality. Under the hypothesis that the VAR including time-varying coefficients is subject to a smooth breakin the coefficients of the Granger causal variables, Christopoulos and Le�on-Ledesma (2008) use the logistic smooth transition autore-gressive (LSTAR) model with time as the transition variable to examine the time-varying Granger causality. Besides, based on Haugh(1976) and Hong (2001), Lu and Hong (2012) construct the time-varying information spillover statistics by introducing a suitablerolling window, which have been used in crude oil market (Jammazi, Ferrer, Jare~no, & Shahzad, 2017). Additionally, Kolodzeij et al.(2014) also explore the time-varying correlation between returns to the spot prices of WTI and equities by the Kalman Filter model andCAPM model.

In summary, it is notable that existing relevant studies have provided a significant quantity of methods and empirical explorations forthe relationship between crude oil futures and financial positions, yet there are still many limitations that require further study. First, theactual relationship generally exists in linear and nonlinear forms but most studies have only focused on one form, such as the linear form.Second, existing studies have explored the general relationship over the whole sample period, or theymay identify the influence changesof financial positions on crude oil futures by dividing the whole sample period into several sub-periods but not in a continuous manner.Therefore, to a large extent, their results are still in the mean and static sense. Thirdly, in the commonly-used Commitments of TradersReport (COT) from CFTC, the concept of non-commercials is wider than actual speculators or hedge funds. Therefore, it is not reasonableor scientific to regard non-commercial positions as speculative positions. Finally, most existing studies focus on the collapse of crude oilprices in 2008, but fewer report that in 2014. Therefore, the main purpose of this paper is to explore the dynamic relationship betweencrude oil futures and speculative positions, and compare the causes of collapse of crude oil futures in 2008 and 2014, which may be ofpractical significance to the market analysts and regulators.

3. Methods and data definitions

3.1. Methods

3.1.1. Linear granger causality testThe traditional linear Granger causality test (Granger, 1969) has been used widely, and the most common test is generally based on

the VAR model, which is specified as Eq. (1), including two stationary time series fxtg and fytg.

xt ¼ A1ðLÞxt þ A2ðLÞyt þ Ux;t

yt ¼ A3ðLÞxt þ A4ðLÞyt þ Uy;t; t ¼ 1; 2;⋯; n (1)

where A1ðLÞ, A2ðLÞ, A3ðLÞ and A4ðLÞ are lag polynomials with roots outside the unit circle, and the error terms Ux;t and Uy;t are i.i.d.

158

Y.-J. Zhang, Y.-B. Wu International Review of Economics and Finance 61 (2019) 156–169

processes with zero mean and constant variance. Time series fytg may Granger cause fxtg if any coefficients in A2ðLÞ are significantlydifferent from zero. The bidirectional causality exists if both A2ðLÞ and A3ðLÞ joint tests are significantly different from zero.

3.1.2. Nonlinear granger causality testBefore the nonlinear Granger causality test, it should be determined whether the time series displays nonlinear characteristics ac-

cording to the Brock, Dechert and Scheinkman (BDS) test (Brock, Scheinkman, Dechert,& Lebaron, 1996). For filtering the linear effect,the BDS test is applied to the residuals after the Auto-Regressive and Moving Average(ARMA) model application. Then, according toDiks and Panchenko (2006), the null hypothesis that time series fytg does not Granger cause fxtg can be implied by Eq. (2):

q ¼ E�fX;Y ;Zðx; y; zÞfXðxÞ � fX;Y ðx; yÞfX;Zðx; zÞ

� ¼ 0 (2)

where fW ðwÞ is the joint probability function, Wt ¼ ðXt ;Yt ;ZtÞ is a Lx þ Ly þ 1-dimensional vector, including Xt � ðXt�Lxþ1;Xt�Lxþ2;⋯;

XtÞ, Yt � ðYt�Lyþ1;Yt�Lyþ2;⋯;YtÞ, and Zt ¼ Xtþ1. The estimator of q can be expressed as TnðεÞ, whose calculation can be simplified intoEq. (3):

TnðεÞ ¼ ð2εÞ2dXþdYþdZ

n

Xi

br0ðWiÞ (3)

with Wi ¼ ðXLxi ;YLy

i ;ZiÞ; i ¼ 1;⋯n, and

br0 ðWiÞ ¼ 13

� dfX;Y;Z ðXi;Yi; ZiÞbfX ðXiÞ �dfX;Y ðXi; YiÞcfX;Z ðXi;ZiÞ�

þ 13n

Xj

�IXij dfX;Y ;Z �Xj; Yj;Zj

�ð2εÞ�dX þ IXYZijbfX �Xj

�ð2εÞ�dX�dY�dZ

�IXYij dfX;Y �Xj;Yj

�ð2εÞ�dX�dZ � IXYij cfX;Z �Xj;Zj

�ð2εÞ�dX�dY

� (4)

where n ¼ N �maxðLX ; LY Þ is the effective sample size, dX , dY and dZ are the dimension of Xi, Yi and Zi, br0ðWiÞ is the estimator of r0ðWiÞwhich is an asymptotic value used to calculate TnðεÞ (see Appendix A in Diks& Panchenko, 2006), IWij is the indicator function and equals

one if����Wi �Wj

���� � ε and zero otherwise, cfW ðWiÞ is the estimator of a dW -variate random vector W at Wi as Eq. (5), and ε is thebandwidth whose asymptotically optimal choice is given by Eq. (6).

cfW ðWiÞ ¼ ð2εÞ�dW

n� 1

Xj;j 6¼i

IWij (5)

ε� ¼ minC*n�

27; 1:5

(6)

where C* is a constant determined by the distribution of time series.

The U-statistic TnðεÞ satisfies T ¼ ffiffiffin

p TnðεÞ�qσ d!Nð0;1Þ under the null hypothesis of i.i.d. For convenience, the estimator of r0ðWiÞ

proposed by Diks and Panchenko (2006) is used to calculate the approximate variance of TnðεÞ, i.e., bσ2 ¼ 9Varð br0ðWiÞÞ. Based on theupper-tailed test in the asymptotic result, we can say that fytg strictly nonlinearly Granger causes fxtgwhen the test statistic exceeds thecritical values.

3.1.3. Time-varying information spillover statistic

(1) White noises

Suppose two independent time series fvi;tg; i ¼ 1;2, are seen as stationary ARMA-GARCH processes. The residuals are standardizedbased on Eq. (7) after the ARMA-GARCH models used for each time series.(

ei;t ¼ ui;tffiffiffiffiffiffihi;t

p ); i ¼ 1; 2 (7)

where ui;t and hi;t are the residuals and variances of the ARMA-GARCH models, respectively.

159

Y.-J. Zhang, Y.-B. Wu International Review of Economics and Finance 61 (2019) 156–169

(2) Size of rolling window

Let M be the size of the rolling window, then ½t �M þ 1; t� is constructued as the window period. Hong's statistics follow anasymptoticNð0;1Þ distribution, so the value ofM needs to be sufficiently large. However, changes of information spillover are smoothedwhen M is too large. In such a situation, according to Lu and Hong (2012), M is singled out based on Eq. (8).

M ¼ 2�z1�α=2 þ z1�β=2

�2�μ0�μ1σ

�2 ¼ 2�z1�α=2 þ z1�β=2

�2Δ2 (8)

where z1�ϕ=2 is the critical value under the significance level ϕ in the Nð0; 1Þ distribution, α is the probability of type-one error, β is theprobability of type-two error and Δ ¼ μ0�μ1

σ is the standardized difference between mean values.

(3) Time-varying statistics

As for two standardized residual series in the window period, let rtðj;MÞ be the cross-correlation of lag order j as Eq. (9).

rtðj;MÞ ¼

8>>>>>>>>>><>>>>>>>>>>:

XM�j�1

i¼0e1;t�ie2;t�i�jffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiXM�1

i¼0e21;t�i

XM�1

i¼0e22;t�i

q ; j ¼ 0; 1;⋯;M � 1

XMþj�1

i¼0e1;t�iþje2;t�iffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiXM�1

i¼0e21;t�i

XM�1

i¼0e22;t�i

q ; j ¼ �1;�2;⋯; 1�M

(9)

Then, Hong's time-varying information spillover statistics are specified as Eqs. (10a) and (10b). The unidirectional statistic for in-formation spillover from time series fv2g to fv1g at time t is H1ðk;MÞ:

H1ðk;MÞ ¼�MXM�1

j¼1K2

�jk

r2t ðj;MÞ � D1

��ð2E1Þ

12 (10a)

and the bidirectional statistic is H2ðk;MÞ:

H2ðk;MÞ ¼�MXM�2

j¼2�MK2

�jk

r2t ðj;MÞ � D2

��ð2E2Þ

12 (10b)

where k is the effective lag order, D1, D2, E1 and E2 are constants determined by the kernel function KðaÞ. Since Lu, Hong, Wang, Lai, andLiu (2014) and Jammazi et al. (2017) state that lagged dynamic correlations usually tend to be zero with large lags in financial markets,the Bartlett kernel function is applied in this paper (Priestley, 1981), as shown in Eq. (11).

KðaÞ ¼�1� jaj; jaj � 10; otherwise

(11)

Under the null hypothesis that two time series are i.i.d., whenM → ∞, the two statistics show an asymptoticNð0; 1Þ distribution, i.e.,H1ðk;MÞ � Nð0;1Þ and H2ðk;MÞ � Nð0;1Þ. Then, if the statistic exceeds the upper-tailed Nð0; 1Þ critical values, we can say that thereexists significant spillover at time t.

3.1.4. Time-varying DP granger testTo capture the dynamic relationship between two time series, we also introduce a rolling window into statistic T in the DP Granger

test according to Hong's information spillover statistics. The statistic T� at time t is specified as:

T* ¼ ffiffiffiffim

p T*mðεÞ � q

σd!Nð0; 1Þ; σ2 ¼ 9Varðr0ðWiÞÞ; i 2 ½t � m; t � 1� (12)

wherem ¼ M �maxðLx; LyÞ is the effective size of rolling window. Similarly, the upper-tailed test is used for statistic T�. We can say thatfytg strictly nonlinearly Granger causes fxtg at time t when the test statistic exceeds the critical values.

3.2. Data descriptions

The postions of institutional investors in crude oil futures market can be reflected generally by the reports of CFTC on each Friday.These reports are divided into two categories: Commitments of Traders Report (COT) and Disaggregated Commitments of Traders

160

0 .0

0 .5

1 .0

1 .5

2 .0

2 .5

3 .0

0 20 40 60 80

100

120

140

160

Open Interest (Contracts of million lots)

Crude oil futures price(Dollar/Barrel)

P riceTO

Fig. 1. Crude oil futures prices (Price) and total open interests (TO).

Y.-J. Zhang, Y.-B. Wu International Review of Economics and Finance 61 (2019) 156–169

Report (DCOT). Both reports provide a breakdown of open interest on each Tuesday, which are divided into two categories: reportableand non-reportable positions.3 The difference between the two reports is the classification of reportable positions. Traders are classifiedas ‘commercial’ or ‘non-commercial’ in the COT reports. And the classification is more detailed in the DCOT reports, which involves‘Money Manager’, ‘Producer/Merchant/Processor/User’, ‘Swap Dealer’, and ‘Other Reportables’. Money Managers are defined astraders who are engaged in managing and conducting organized futures trading on behalf of clients, and generally regarded as hedgefunds (Buyuksahin & Harris, 2011). Hence, this paper defines the positions of Money Managers in the DCOT reports as the positions ofhedge funds.

As the prices of crude oil futures on the New York Mercantitle Exchange (NYMEX) are dominant in the world (Liu, Schultz, &Swieringa, 2014), thus the Tuesday to Tuesday closing prices of the Cushing, Oklahoma, Crude Oil Future Contract 1 (RCLC1) of WTItraded in the NYMEX are selected as crude oil futures prices in this paper, which are obtained from the website of Energy InformationAdministration (EIA) of United States. In addition, to study the relationship changes between crude oil futures returns and hedge funds’net positions during the collapse of crude oil prices in 2008 and 2014, this paper traces back to June 13th, 2006, when the earliest dataof the DCOT are avaliable. In the end, the sample period ranges from 6/13/2006 to 12/26/2017.

Fig. 1 presents the dynamics of crude oil futures prices (Price) and total open interest (TO) over the sample period. We can find thatprior to 2014, crude oil futures prices and open interest appeared to have similar trends. However, after 2014, although crude oil futuresprices experienced significant crashes and fluctuated at the relatively lower level, the total open interest remained an upward trend.

Fig. 2 shows the total open interests, the total positions of hedge funds and their proportions (i.e., TO, TO_HF and PTO_HF,respectively) increased prior to the boom of crude oil futures prices in 2008. Although both slightly decreased following the collapse in

3 According to the explanatory notes of CFTC, reportable positions show the futures and options positions of traders whose positions meet orexceed specific reporting levels set by CFTC regulations, and are generally regarded as companies and institutional investors, whereas non-reportablepositions involve traders whose positions do not meet the levels, and are generally regarded as small traders.

161

Fig. 2. Total open interests (TO), hedge funds' total positions (TO_HF) and their proportions to total open interests (PTO_HF).

Y.-J. Zhang, Y.-B. Wu International Review of Economics and Finance 61 (2019) 156–169

crude oil futures prices, they have increased steadily in recent years, which reflects the increasingly important role of hedge funds ininternational crude oil futures market.

Fig. 3 depicts the total positions and net positions of hedge funds as well as crude oil futures prices (i.e., TO_HF, HFNL and Price,respectively). We can see that the trend coordination between crude oil futures prices and hedge funds' net positions outweighs thatbetween crude oil futures prices and hedge funds' total positions. The hedge funds' total positions have shown an upward trend similar tocrude oil futures prices from early-2009 to mid-2011 but not after 2014. Comparatively, the trend of hedge funds' net positions isbasically consistent with crude oil futures prices in the whole sample period. This shows that hedge funds' net positions are moresensitive than hedge funds’ total positions to the price changes of crude oil futures.

Table 1 presents the descriptive statistics for crude oil futures returns and hedge funds' net positions. The JB statistic for the twotime series is statistically significant at the 1% significance level, indicating that they do not follow the normal distribution. Ac-cording to the skewness and kurtosis, we can say that crude oil futures returns and hedge funds' net positions both follow theleptokurtic and left-skewness distribution. The Ljung-Box Q-statistic of lag order 12 shows the null hypothesis of independent seriesis rejected at the 1% significance level for hedge funds’ net positions and 5% significance level for crude oil futures returns.Therefore, we can say that both series are auto-correlated. In addition, based on the ARCH LM-statistic, we can see that both seriesmay present nonlinear dependence due to their clustering effects (ARCH tests are based on the ARMA model). This is consistentwith the observations that returns in the futures markets generally show a fat tail, auto-correlation and volatility clustering features(Baum & Zerilli, 2016).

Moreover, the time series stationarity is tested based on the unit roots with the Augmented Dickey–Fuller (ADF), Phillips–Perron(PP) and Kwiatkowski-Phillips-Schmidt-Shin (KPSS) approaches. Table 2 shows that both crude oil futures returns and hedge funds' netpositions appear to be stationary, so the VAR model can be developed subsequently. Meanwhile, the BDS test results show that the null

162

Fig. 3. Crude oil futures prices (Price), hedge funds' total positions (TO_HF) and net long positions (HFNL).

Y.-J. Zhang, Y.-B. Wu International Review of Economics and Finance 61 (2019) 156–169

hypothesis of i.i.d. is rejected at the 1% significance level for crude oil futures returns and hedge funds’ net positions.4 It can therefore beinferred that a nonlinear relationship exists between them.

4. Empirical results and analyses

4.1. The Granger causality results

The Granger causality test is based on the VARmodel with the lag order 1, which is selected by the SIC values. According to the linearand nonlinear Granger causality tests mentioned in section 3.1, the results are displayed in Table 3.

The results in Table 3 show that hedge funds' net positions may Granger cause crude oil futures returns in the linear form but otherresults are rather weak. According to the linear results, hedge funds' net positions may Granger cause crude oil futures returns at the 1%significance level, but crude oil futures returns may Granger cause hedge funds' net positions just at the 10% significance level. In thenonlinear results, hedge funds' net positions may Granger cause crude oil futures returns at the 10% significance level with lag order

4 Detailed results can be obtained upon request from authors.

163

Table 1Descriptive statistics for hedge funds’ net positions and crude oil returns.

Statistic LNHFNL Return

Mean 11.4107 �0.0003Std. Dev. 2.5267 0.0504Skewness �7.1997 �0.1442Kurtosis 58.9897 4.9912JB test 83972.4002 *** 101.7105 ***Q(12) 1473.3832 *** 21.7774 **ARCH-LM(5) 19.6544 *** 73.2752 ***

Note: LNHFNL is equal to ln(hedge funds net positions) or -ln(|hedge funds net positions|) when

hedge funds' net positions are negative. Returnt ¼ lnPtPt�1

, where Pt denotes the crude oil futures

price at time t. ** and *** indicate the significance at the 5% and 1% levels, respectively.

Table 2The unit root test results.

Variables ADF PP KPSS

LNHFNL �6.1679 *** �7.6224 *** 0.1083Return �25.0422 *** �25.1346*** 0.0490

Note: *** denotes the significance at the 1% level. The null hypotheses of ADF and PP tests are that the variable has a unit root,while for KPSS test no unit root.

Table 3Granger causality test results.

Panel A: Linear Granger causality test

H0 χ21 P-value

LNHFNL does not Granger cause Return 12.2306 0.0005Return does not Granger cause LNHFNL 3.4796 0.0621

Panel B: Nonlinear Granger causality testH0’ : LNHFNL does not Granger cause Return H0

’ : Return does not Granger cause LNHFNLLx¼Ly T P-value T P-value

1 1.4845 0.0688 1.1445 0.12622 1.2117 0.1128 1.0206 0.15373 0.8433 0.1995 1.6189 0.05274 0.5318 0.2974 1.5148 0.06495 0.4633 0.3216 1.3459 0.08926 0.4721 0.3184 0.9965 0.15957 0.8651 0.1935 0.8286 0.20378 0.8166 0.2071 0.551 0.2908

Note: As suggested by Diks and Panchenko (2006), we set the constant C� � 8 in Eq. (6), then the optimal bandwidth ε� ¼ 1:29 and the number of lagsLx ¼ Ly ¼ 1; 2;⋯8 in the nonlinear Granger test.

Y.-J. Zhang, Y.-B. Wu International Review of Economics and Finance 61 (2019) 156–169

Lx ¼ Ly ¼ 1, and crude oil futures returns may Granger cause hedge funds' net positions at the 10% significance level with lag order Lx ¼Ly ¼ 3; 4;5. Moreover, according to the P-value, the linear effect of hedge funds’ net positions on crude oil futures returns is strongerthan the nonlinear effect, which is consistent with Zhang (2013) but in contrast to Irwin and Sanders (2012) and Sanders and Irwin(2014). The reason for the divergence may be the use of different positions.

4.2. The time-varying information spillover test results

Hong's time-varying information spillover tests (see Eq. (10)) are applied to the ARMA-GARCH standardized residuals correspondingto crude oil futures returns and hedge funds' net positions.5 According to the ARCH LM test,6 the null hypothesis of all regression co-efficients in the ARCH model being jointly zero can be accepted at the 5% significance level. It can therefore be inferred that the twoARMA-GARCH models basically eliminate the clustering effects.

5 Detailed results of the ARMA-GARCH model can be obtained upon request from authors.6 The ARCH-LM (5) of LNHFNL is 0.0238, and the ARCH-LM (5) of Return is 0.8539, with the significance probabilities 1.0000 and 0.9735,

respectively.

164

0

5

10

15

20

Ln(1/p)

LNHFNL——>Return LNHFNL<——Return LNHFNL<——>Return

p = 0.01

p = 0.05

p = 0.10

Fig. 4. Hong's time-varying information spillover test resultsNote: The ordinate values that are not shown in some time periods in the figure are larger than 20.

Y.-J. Zhang, Y.-B. Wu International Review of Economics and Finance 61 (2019) 156–169

According to the Monte Carlo simulation results of Lu and Hong (2012), the efficiency of Hong's time-varying information spillovertests is silghtly affected by the selection of lag order k and rolling window sizeM. Thus we follow Lu and Hong (2012) to set k ¼ 10 in Eq.(10). According to the construction of statistics in Eq. (10), Hong's time-varying test cannot present the changes of correlations betweentwo time series during sample period t 2 ½1; M� 1�. To capture the relationship between crude oil futures returns and hedge funds' netpositions before the outbreak of financial crisis in 2008, we choose a smaller rolling window size M ¼ 64 when α ¼ 0:05, β ¼ 0:2 andΔ ¼ 0:5 in Eq. (8) proposed by Lu and Hong (2012).7 The results of Hong's time-varying information spillover tests are shown in Fig. 4.

To show the result more clearly, the simple formula lnð1=pÞ is applied in Fig. 4. Based on Fig. 4, it can be found that the relationshipbetween crude oil futures returns and hedge funds' net positions has evident time-varying characteristics. For most of the sample period,the P-values of bidirectional statistic (see Eq. (10b)) are below 1%, which reflects the significant bidirectional spillover between crudeoil futures returns and hedge funds' net positions. Meanwhile, the P-values of the unidirectional information spillover statistic (see Eq.(10a)) from crude oil futures returns to hedge funds' net positions are below 5% during September and December 2010, February toNovember 2014 and April to December 2017, and below 1% during February to October 2014 and May to December 2017. Conversely,the information spillover from hedge funds’ net positions to crude oil futures returns clusters during April to July 2008, January toFebruary 2016, and June to November 2017 when the P-values are below 5%, and during July and October 2017 when the P-values arebelow 1%.

Prior to the global financial crisis in 2008, WTI crude oil bubbles were pushed by speculative behaviors and had increased rapidlysince 2007 (Zhang & Yao, 2016). The information spillover from hedge funds' net positions to crude oil futures returns became sig-nificant during April to July 2008. In order to address the financial crisis, since the end of 2008, most countries including Europemembers and USA have implemented a series of expansionary fiscal policies and loosed monetary policies, which have attracted hedgefunds to flood into crude oil futures markets (Chen, Filardo, He, & Zhu, 2016; Cukierman, 2013; Grammatikos, Lehnert, & Otsubo,2015), and hedge funds' net positions have increased again. However, in 2010, the Dodd-Frank Act's Title IV required financial in-stitutions including hedge funds to offer more transparent disclosure (US Congress, 2010), which is a disadvantage to the excessivespeculation of hedge funds. Based on these grounds, the information spillover from hedge funds' net positions to crude oil futures returnsstrengthened significantly in 2009 but weakened sharply in 2010. Since the limit crude oil production of OPEC and Russia from 2016,the increasing crude oil prices have attracted hedge funds to flood into crude oil futures markets again. According to the mechansim ofpositive feedback (Tokic, 2011), hedge funds pushed the crude oil prices up further, which strengthened the effect of hedge funds' netpositions on crude oil futures returns again.

7 The result with lower probability of type-two error β ¼ 0:05 and M ¼ 100 is similar to the result in this paper.

165

Y.-J. Zhang, Y.-B. Wu International Review of Economics and Finance 61 (2019) 156–169

Besides, there exists significant unidirectional information spillover from crude oil futures returns to hedge funds' net positionswhen crude oil futures prices are highly volatile, and there is significant unidirectional information spillover in turn when crudeoil futures prices continue to rise. Combining Figs. 3 and 4, we can find that crude oil price experienced rapid rise and sharpdecline during the whole 2008 and June 2010 to October 2011 when the unidirectional information spillover from crude oilfutures returns to hedge funds' net positions has been strengthened, and the P-values are even below 1% during the crash in 2014and the continuous rise of crude oil prices in the second half of 2017. On the other hand, the hedge funds' net positions increasedsharply during April to May 2008, June to November 2009, January to March 2016 and July to October 2017 whilst crude oilfutures prices experienced rising continually. During most of these periods, the P-values of the unidirectional information spilloverstatistic from hedge funds' net positions to crude oil futures returns are below 5%, which means hedge funds have significantinfluence on crude oil futures prices and tend to put them up. Based on the analysis above, high volatility of crude oil futuresprices strengthens the influence of crude oil futures returns on hedge funds' net positions, which is in agreement with the results ofBuyuksahin and Harris (2011). And the result also indicates that hedge funds’ net positions have evident effect on crude oil futuresreturns when crude oil futures prices continue to rise. This proves the positve feedback mechanism of De Long et al. (1991) andTokic (2011).

Finally, comparing the significance of unidirectional information spillover from hedge funds' net positions to crude oil futures returnsin 2008 and 2014, we find that the bubbles in crude oil futures prices in which hedge funds play an important role contribute to the crashof crude oil futures prices in 2008, but hedge funds have no significant effect on the crude oil futures price crash in 2014. According toFigs. 3 and 4, the information spillover from hedge funds' net positions to crude oil futures returns remained significant when crude oilfutures prices rose rapidly in 2008, meaning that hedge funds' net positions have linear shock on crude oil futures returns during thisperiod. For instance, in the first half of 2008, hedge funds increased net positions and pushed up the bubbles in crude oil futures marketunder excessive supply of crude oil. This validates the explanation of Tokic (2011) that the speculators play an important role in pushingup crude oil prices bubbles in 2008.However, when crude oil futures prices dropped again in the second half of 2014, the informationspillover from hedge funds' net positions to crude oil futures returns is significant only after the crash of crude oil futures prices. Afterweakening in 2009, the hedge funds’ net positions have no significant influence on crude oil futures returns until 2016.

4.3. The time-varying DP granger test results

To caputure the dynamic relationship between crude oil futures returns and hedge funds’ net positions in the nonlinear form, therolling window is introduced to the DP Granger test. As denoted by Yang and Zhang (2014), the DP Granger test can be more accuratewhen the window size reaches at least 200 and the lag order is equal to one. Thus, we setM ¼ 200, Lx ¼ Ly ¼ 1 and ε* ¼ 1:5 accordingto Eq. (6) in the time-varying DP Granger test, as shown in Fig. 5.

0

2

4

6

8

10

12

Ln(1/p)

LNHFNL——>Return LNHFNL<——Return

p = 0.01

p = 0.05p = 0.10

Fig. 5. Time-varying DP Granger test results.

166

Y.-J. Zhang, Y.-B. Wu International Review of Economics and Finance 61 (2019) 156–169

Due to the large window size, the results are more smooth than the linear results and only present the relationship changes betweencrude oil futures returns and hedge funds' net positions after 2010. However, the nonlinear characteristics are similar to the linearresluts. First, the crude oil futures returns have significant influence on hedge funds' net positions during high volatility of crude oilfutures prices. When the crude oil futures prices experience drastic fluctuations after 2014, the P-values of the DP Granger test fromcrude oil futures returns to hedge funds' net positions keep below 1%. Second, the hedge funds' net positions have significnat influenceon crude oil futures returns when the crude oil futures prices rise continuously. The P-values of the DP Granger test from hedge funds' netpositions to crude oil futures returns remain below 5% after the crude oil futures prices bottom out and rise in 2016. Finally, we also findthat hedge funds are not the main reason for the crush of crude oil futures prices in 2014, since hedge funds’ net positions have nosignificant influence on crude oil futures returns during this period.

Both linear and nonlinear results prove that hedge funds are not the main reason for the crush of crude oil futures prices in 2014.According to the study of Zhang and Yao (2016), market fundamentals were themain driver ofWTI crude oil price drop during 2014 and2015. In 2014, due to increasing production in shale gas fields, American crude oil production significantly exceeded expected output. Inaddition, traditional crude oil exporting countries such as Libya and Iraq resumed production. In October 2014, OPEC's crude oilproduction reached 30.25 million barrels per day, which was higher than the production ceiling of 30 million barrels per day (OPEC,2014). In the case of slow growth in demand, the substantial increase of crude oil supply and dramatic appreciation of the US dollarjointly contributed to the drop of crude oil prices.

5. Concluding remarks

In this paper, the linear and nonlinear Granger causality tests as well as Hong's time-varying information spillover test and time-varying DP Granger test are used to explore the relationship between crude oil futures returns and hedge funds' net positions from2006 to 2017. Some main conclusions are soundly drawn as follows:

First of all, hedge funds' net positions have significant effect on crude oil futures returns in the linear form but their nonlinear in-fluence is much weaker. In general, hedge funds’ net positions may Granger cause crude oil futures returns at the 5% significance level inthe linear form, but not vice versa. Additionally, the results of DP Granger test are not significnat at the 5% level.

Second, both the linear and nonlinear relationship between crude oil futures returns and hedge funds' net positions has time-varyingcharactistics, and the influence of crude oil futures prices has been evidently strengthened when they experience high volatility whilstthe influence of hedge funds' net positions has been strengthened only when crude oil futures prices rise substantially. Besides, thecorrelation between crude oil futures returns and hedge funds’ net positions has become increasingly stronger since the crude oil futuresprice crash in 2014.

Finally, hedge funds play an important role in pushing up bubbles in crude oil futures prices, which contribute to the crash of crudeoil futures prices in 2008, but in 2014 crude oil price drop is mainly due to market fundamentals.

These conclusions may help crude oil futures investors to judge price tendency and related departments to develop appropriatepolicies. For example, considering the increasing influence of hedge funds in recent years, investors and policymakers should pay moreattention to the positions of hedge funds as well as market fundamentals to forcast crude oil futures prices.

There is still much relevant work to do in the future. For instance, this paper analyzes the weekly data, and we can make furtherresearch if the daily data are avaialble in the future. In addition, this paper examines the effect of hedge funds on the violent fluctuationsin crude oil futures prices during 2008 and 2014, there are still some key factors which need to be considered in the future, such aseconomic, finance and credit.

Acknowledgments

We gratefully acknowledge the financial support from National Natural Science Foundation of China (nos. 71273028, 71322103,71774051), National Program for Support of Top-notch Young Professionals (no. W02070325), Changjiang Scholars Program of theMinistry of Education of China (no. Q2016154), Hunan Youth Talent Program.

References

Baek, E. G., & Brock, W. A. (1992). A general test for nonlinear granger causality: bivariate model, Working paper. Iowa State University and University of WisconsinMadison.

Basak, S., & Pavlova, A. (2016). A model of financialization of commodities. The Journal of Finance, 71(4), 1511–1556. https://doi.org/10.1111/jofi.12408.Baum, C. F., & Zerilli, P. (2016). Jumps and stochastic volatility in crude oil futures prices using conditional moments of integrated volatility. Energy Economics, 53,

175–181. https://doi.org/10.1016/j.eneco.2014.10.007.Brock, W. A., Scheinkman, J. A., Dechert, W. D., & Lebaron, B. (1996). A test for independence based on the correlation dimension. Econometric Reviews, 15(3),

197–235. https://doi.org/10.1080/07474939608800353.Buyuksahin, B., & Harris, J. H. (2011). Do speculators drive crude oil futures prices? Energy Journal, 32(2), 167–202. https://doi.org/10.5547/issn0195-6574-ej-vol32-

no2-7.Chen, Q., Filardo, A., He, D., & Zhu, F. (2016). Financial crisis, US unconventional monetary policy and international spillovers. Journal of International Money and

Finance, 67, 62–81. https://doi.org/10.1016/j.jimonfin.2015.06.011.Cho, D. (2008). A few speculators dominate vast market for oil trading,Washington Post. Available at: http://www.washingtonpost.com/wp-dyn/content/article/2008/

08/20/AR2008082003898.html?noredirect=on.Christopoulos, D. K., & Le�on-Ledesma, M. A. (2008). Testing for Granger (non-)causality in a time-varying coefficient VAR model. Journal of Forecasting, 27(4),

293–303. https://doi.org/10.1002/for.1060.Cifarelli, G., & Paladino, G. (2010). Oil price dynamics and speculation: A multivariate financial approach. Energy Economics, 32(2), 363–372. https://doi.org/10.1016/

j.eneco.2009.08.014.

167

Y.-J. Zhang, Y.-B. Wu International Review of Economics and Finance 61 (2019) 156–169

Congress, U. S. (2010). Dodd-Frank wall street reform and consumer protection act, 2010, Sec. 401- Sec. 416. Available at: https://www.congress.gov/111/plaws/publ203/PLAW-111publ203.pdf.

Cukierman, A. (2013). Monetary policy and institutions before, during, and after the global financial crisis. Journal of Financial Stability, 9(3), 373–384. https://doi.org/10.1016/j.jfs.2013.02.002.

De Long, J. B., Shleifer, A., Summers, L. H., & Waldmann, R. J. (1990). Noise trader risk in financial markets. Journal of Political Economy, 98(4), 703–738. https://doi.org/10.1086/261703.

Diks, C., & Panchenko, V. (2006). A new statistic and practical guidelines for nonparametric Granger causality testing. Journal of Economic Dynamics and Control,30(9–10), 1647–1669. https://doi.org/10.1016/j.jedc.2005.08.008.

Ding, H., Kim, H. G., & Park, S. Y. (2014). Do net positions in the futures market cause spot prices of crude oil? Economic Modelling, 41, 177–190. https://doi.org/10.1016/j.econmod.2014.05.008.

Ederington, L., & Lee, J. H. (2002). Who trades futures and how: Evidence from the heating oil futures market. Journal of Business, 75(2), 353–374. https://doi.org/10.2139/ssrn.252331.

Fan, Y., & Xu, J.-H. (2011). What has driven oil prices since 2000? A structural change perspective. Energy Economics, 33(6), 1082–1094. https://doi.org/10.1016/j.eneco.2011.05.017.

Gallo, A., Mason, P., Shapiro, S., & Fabritius, M. (2010). What is behind the increase in oil prices? Analyzing oil consumption and supply relationship with oil price.Energy, 35(10), 4126–4141. https://doi.org/10.1016/j.energy.2010.06.033.

Gkanoutas-Leventis, A., & Nesvetailova, A. (2015). Financialisation, oil and the great recession. Energy Policy, 86, 891–902. https://doi.org/10.1016/j.enpol.2015.05.006.

Grammatikos, T., Lehnert, T., & Otsubo, Y. (2015). Market perceptions of US and European policy actions around the subprime crisis. Journal of International FinancialMarkets, Institutions and Money, 37, 99–113. https://doi.org/10.1016/j.intfin.2015.02.007.

Granger, C. W. J. (1969). Investigating causal relations by econometric models and cross-spectral methods. Econometrica, 37(3), 424–438. https://doi.org/10.2307/1912791.

Haugh, L. D. (1976). Checking the independence of two covariance-stationary time series: A univariate residual cross-correlation approach. Journal of the AmericanStatistical Association, 71(354), 378–385. https://doi.org/10.2307/2285318.

Hiemstra, C., & Jones, J. D. (1993). Monte Carlo results for a modified version of the Baek and Brock nonlinear Granger causality test. Working paper, University ofStrathclyde and Securities and Exchange Commission.

Hiemstra, C., & Jones, J. D. (1994). Testing for linear and nonlinear Granger causality in the stock price-volume relation. The Journal of Finance, 49(5), 1639–1664.https://doi.org/10.2307/2329266.

Hong, Y. M. (2001). A test for volatility spillover with application to exchange rates. Journal of Econometrics, 103(1–2), 183–224. https://doi.org/10.1016/S0304-4076(01)00043-4.

Hsieh, D. A. (1989). Testing for nonlinear dependence in daily foreign exchange rates. Journal of Business, 62(3), 339–368. https://doi.org/10.1086/296466.Hsieh, D. A. (1991). Chaos and nonlinear dynamics: Application to financial markets. The Journal of Finance, 46(5), 1839–1877. https://doi.org/10.1111/j.1540-6261.

1991.tb04646.x.Irwin, S. H., & Sanders, D. R. (2010). The impact of index and swap funds on commodity futures markets: Preliminary results. In OECD Food, Agriculture & Fisheries

Papers, No. 27, OECD Publishing, Paris. https://doi.org/10.1787/5kmd40wl1t5f-en.Irwin, S. H., & Sanders, D. R. (2012). Testing the masters hypothesis in commodity futures markets. Energy Economics, 34(1), 256–269. https://doi.org/10.1016/j.

eneco.2011.10.008.Jammazi, R., Ferrer, R., Jare~no, F., & Shahzad, S. J. H. (2017). Time-varying causality between crude oil and stock markets: What can we learn from a multiscale

perspective? International Review of Economics & Finance, 49, 453–483. https://doi.org/10.1016/j.iref.2017.03.007.Kaufmann, R. K. (2011). The role of market fundamentals and speculation in recent price changes for crude oil. Energy Policy, 39(1), 105–115. https://doi.org/10.

1016/j.enpol.2010.09.018.Kolodziej, M., Kaufmann, R. K., Kulatilaka, N., Bicchetti, D., & Maystre, N. (2014). Crude oil: Commodity or financial asset? Energy Economics, 46, 216–223. https://doi.

org/10.1016/j.eneco.2014.09.006.Lammerding, M., Trede, M., & Wilfling, B. (2013). Speculative bubbles in recent oil price dynamics: Evidence from a Bayesian Markov-switching state-space approach.

Energy Economics, 36(3), 491–502. https://doi.org/10.1016/j.eneco.2012.10.006.Li, H., Kim, M. J., & Park, S. Y. (2016). Nonlinear relationship between crude oil price and net futures positions: A dynamic conditional distribution approach.

International Review of Financial Analysis, 44, 217–225. https://doi.org/10.1016/j.irfa.2016.01.022.Liu, W. M., Schultz, E., & Swieringa, J. (2014). Price dynamics in global crude oil markets. Journal of Futures Markets, 35(2), 148–162. https://doi.org/10.1002/fut.

21658.Lu, F. B., & Hong, Y. M. (2012). Time-varying information spillover tests and their application to financial markets. Journal of Management Sciences in China, 15(4),

31–39. https://doi.org/10.3969/j.issn.1007-9807.2012.04.005.Lu, F. B., Hong, Y. M., Wang, S. Y., Lai, K. K., & Liu, J. (2014). Time-varying Granger causality tests for applications in global crude oil markets. Energy Economics, 42,

289–298. https://doi.org/10.1016/j.eneco.2014.01.002.Moosa, I. A., & Silvapulle, P. (2000). The price–volume relationship in the crude oil futures market: Some results based on linear and nonlinear causality testing.

International Review of Economics & Finance, 9, 11–30. https://doi.org/10.1016/S1059-0560(99)00044-1.Morana, C. (2013). Oil price dynamics, macro-finance interactions and the role of financial speculation. Journal of Banking & Finance, 37(1), 206–226. https://doi.org/

10.1016/j.jbankfin.2012.08.027.OPEC. (2014). OPEC monthly oil market report, Austria, November.Priestley, M. B. (1981). Spectral analysis and time series. London: Academic Press.Sanders, D. R., & Irwin, S. H. (2014). Energy futures prices and commodity index investment: New evidence from firm-level position data. Energy Economics, 46(S1),

S57–S68. https://doi.org/10.1016/j.eneco.2014.09.005.Sato, J. R., Junior, E. A., Takahashi, D. Y., de Maria Felix, M., Brammer, M. J., & Morettin, P. A. (2006). A method to produce evolving functional connectivity maps

during the course of an fMRI experiment using wavelet-based time-varying Granger causality. NeuroImage, 31(1), 187–196. https://doi.org/10.1016/j.neuroimage.2005.11.039.

Sims, C. A. (1972). Money, income, and causality. The American Economic Review, 62(4), 540–552. https://www.jstor.org/stable/1806097.Tokic, D. (2011). Rational destabilizing speculation, positive feedback trading, and the oil bubble of 2008. Energy Policy, 39(4), 2051–2061. https://doi.org/10.1016/j.

enpol.2011.01.048.Yang, W., Han, A., Hong, Y. M., & Wang, S. Y. (2016). Analysis of crisis impact on crude oil prices: A new approach with interval time series modelling. Quantitative

Finance, 16(12), 1917–1928. https://doi.org/10.1080/14697688.2016.1211795.Yang, Z. H., & Zhang, Y. L. (2014). Simulation analysis for power and finite-sample properties of nonlinear Granger causality tests. Statistical Research, 31(5), 107–112.

https://doi.org/10.19343/j.cnki.11-1302/c.2014.05.017.Yu, H. H., Du, D. L., Fang, L. B., & Yan, P. P. (2018). Risk contribution of crude oil to industry stock returns. International Review of Economics & Finance, 58, 179–199.

https://doi.org/10.1016/j.iref.2018.03.009.Zhang, Y. J. (2013). Speculative trading and WTI crude oil futures price movement: An empirical analysis. Applied Energy, 107(4), 394–402. https://doi.org/10.1016/j.

apenergy.2013.02.060.Zhang, Y. J., Chevallier, J., & Guesmi, K. (2017). De-financialization" of commodities? Evidence from stock, crude oil and natural gas markets. Energy Economics, 68,

228–239. https://doi.org/10.1016/j.eneco.2017.09.024.Zhang, Y. J., & Wang, J. (2015). Exploring the WTI crude oil price bubble process using the Markov regime switching model. Physica A Statistical Mechanics & Its

Applications, 421(1), 377–387. https://doi.org/10.1016/j.physa.2014.11.051.

168

Y.-J. Zhang, Y.-B. Wu International Review of Economics and Finance 61 (2019) 156–169

Zhang, Y. J., & Wang, J. L. (2019). Do high-frequency stock market data help forecast crude oil prices? Evidence from the MIDAS models. Energy Economics, 78,192–201. https://doi.org/10.1016/j.eneco.2018.11.015.

Zhang, Y. J., & Yao, T. (2016). Interpreting the movement of oil prices: Driven by fundamentals or bubbles? Economic Modelling, 55, 226–240. https://doi.org/10.1016/j.econmod.2016.02.016.

Zhang, Y. J., Yao, T., He, L. Y., & Ripple, R. (2019). Volatility forecasting of crude oil market: Can the regime switching GARCH models beat the single-regime GARCHmodels? International Review of Economics & Finance, 59, 302–317. https://doi.org/10.1016/j.iref.2018.09.006.

Zhang, Y. J., & Zhang, J. L. (2018). Volatility forecasting of crude oil market: A new hybrid method. Journal of Forecasting, 37(8), 781–789. https://doi.org/10.1002/for.2502.

Dr. Yue-Jun Zhang is a Professor at Business School, Hunan University, China, as well as the Director of Center for Resource and Environmental Management, HunanUniversity. He got his PhD degree in Energy Economics from Chinese Academy of Sciences in 2009. His research interests include energy economics and policy modelling.Up to now, Dr. Zhang has published more than 80 articles in peer-reviewed journals, such as Energy Economics, Journal of Forecasting, Quantitative Finance, EconomicModelling, Energy Policy, Journal of Policy Modelling, Resources Policy. Dr. Zhang was a visiting scholar at Energy Studies Institute of National University of Singapore,and Lawrence Berkeley National Laboratory, USA.

Mr. Yao-Bin Wu is a graduate student at School of Management, Fudan University, China. His research interests focus on oil market modelling.

169