Ol Prce Formaton Revsted: The Increasng Role of the Non- Commercals By Doulos Dmtrs, Dept.of Economcs Deree College,ddoulos@acg.edu Katsats Odysseas, Dept.of Economcs Deree College,katsa@acg.edu Margarts Kostas, BA Economcs Deree College, K.Margarts@acg.edu Merka Anna Dept.of Economcs Deree College,merkas@acg.edu Abstract We study the role of commercal versus non-commercal traders n the ol market. We fnd that over the perod 1993-2016 the mpact of non-commercals ncreased drastcally whle that of the commercals nearly faded out. Tryng to nvestgate further we run quantle regresson under heteroscedastcty and found that even though the mpact of non-commercals on ol prce formaton s stronger and remans consstent as we move to hgher quantles over the 2008-2016 perod, t s the commercals that prmarly nfluence and explan ol prce volatlty n a postve manner. Further, noncommercals at the hgher quantle of 0.9 are found to have a sgnfcant but reducng mpact on volatlty. Ths fndng mpnges drectly upon the nature of ths commodty and llustrates the mpact that unexpected geopoltcal events can have on the prce of ol. It seems that non-commercals have a stablzng effect on the market and commercals account for the prce volatlty. Our results have mportant mplcatons for all stakeholders nvolved. Keywords: Commercal ol traders, Non-commercal ol traders, ol market, quantle regresson analyss. 1
1. Introducton The sharp ncrease n ol prces n the late 2000s and the subsequent recent declne has drawn the attenton of academcs, market partcpants and polcy makers wth subsequent research to focus on the underlyng market dynamcs. The exploraton of ol prce formaton not only helps to explan the causes of ths recent prce volatlty but t may also have mportant polcy mplcatons. More specfcally, f speculaton s found to be a sgnfcant contrbutor n the recent ups and downs n ol markets, ths could be a reason for strcter regulaton and supervson n dervatve markets. If, on the other hand, t s fundamentals that prmarly affect ol prces, then the arena s set for polcy makers to act accordngly. There s no doubt that ol markets underwent sgnfcant structural changes n the last two decades. Fan and Hu (2011) argue that ol markets became more globalzed durng ths perod, manly due to the ncrease n nternatonal trade and advances n technology. Furthermore, they are ncreasngly related to other fnancal markets as well as macroeconomc developments. In other words, prces are no longer determned prmarly by demand and supply forces but they are more and more affected by developments n other fnancal markets such as currency, stock and futures markets. There has been an extensve debate over whether the recent fluctuatons n the prce of crude ol are due to changes n macroeconomc fundamentals or to ncreasng speculaton n ol futures and other fnancal markets actvty. Lu et. al. (2016) break down the effects of fundamental and dervatve market speculatve shocks on ol prces. Usng a structural VAR wth sgn restrctons they found that speculatve shocks explan only 10% of the ol prce varance whereas US and Chna demand shocks explan almost 70%, wth the effect from Chna beng sgnfcantly stronger. Hamlton (2009a) confrmed the conventonal vew that ol prce shocks before the 2000s were due to sgnfcant dsruptons n ol producton caused by geopoltcal events whereas the recent run up n ol prces before the fnancal crss was caused by demand condtons. In Hamlton (2009b), the author argued that strong growth n demand for ol by newly ndustralzed countres and the falure of global producton to ncrease have trggered sgnfcant commodty speculaton. Fan and Hu (2011) explored the man factors that caused ol prce fluctuatons durng the 2000-2009 perod consderng both fundamentals and fnancal markets. Usng an endogenously determned break test they dentfed three sub-perods: 2000-2004, 2004-2008 and 2008-2009. They found that 2
speculaton and strong demand drove ol prces before the fnancal crss and supply/demand fundamentals rght after the crss. They conclude that the effects of non-fundamentals have ncreased over the perod examned. Fnally, the exstence of ol prce bubbles was nvestgated by Zhang and Yao (2016) for the 2001-2015 perod. They dentfed postve bubbles n brent and WTI durng the 2000-08 perod concludng that these bubbles were responsble for the overshootng of ol prces before the fnancal crss. Ol markets have been ncreasngly fnancalzed as more and more dfferent types of nvestors (hedge funds, speculators, portfolo managers) partcpate usng several dfferent fnancal tools such as futures, optons, ndex funds etc. The growng mpact on ol prces from tradng actvty n other fnancal markets, manly stock and dervatve markets, has recently been extensvely nvestgated. Alon and Assa (2016) explored the dynamc relatonshp between ol prce movements, the US stock market and the dollar exchange rate and found that an ncrease n ol prces s assocated wth the dollar deprecaton and an ncrease n stock prces. Ewng and Malk (2016) found a drect and ndrect transmsson of volatlty between the ol and the stock market for the 1996-2013 perod. The nterdependence between ol and stock markets has also been confrmed by Mens et. al. (2017) and Bour et. al. (2017). The rapd development of markets for dervatve securtes, especally for ol futures, has been a very mportant factor n ol markets, prmarly for two reasons: the transfer of rsk and the contrbuton to prce dscovery. The role of futures tradng n affectng ol prces and ther volatlty has recently attracted sgnfcant research nterest. Conventonally, tradng ol futures s conducted by two groups: Companes or states for the purpose of hedgng rsk exposure (the so called commercal traders) and speculators, such as traders and hedge funds (the so called non-commercal traders). The focus of ths study s on the contrbuton of these traders on prce dscovery process n the ol spot market and ther attrbuted role n explanng ol prce volatlty. More specfcally we show, usng monthly data over the perod 1993-2016, through robust least squares, that both commercals and non-commercals mpact sgnfcantly and postvely ol prce formaton. Other factors lke the Baltc Dry Index, used as a proxy to economc growth and the SP500 have also a postve and sgnfcant mpact on the prce of ol. Consequently we splt our sample nto three perods: 1993-3
2003, 2004-2016, and 2008-2016. It became apparent that durng the 2004-2016, the so called Chna factor perod, the mpact of non-commercals ncreased drastcally whle that of the commercals nearly faded out. The same pattern was observed clearly durng 2008-2016. Tryng to nvestgate further we run quantle regresson under heteroscedastcty and found that even though the mpact of non-commercals on ol prce formaton s stronger and remans consstent as we move to hgher quantles over the 2008-2016 perod, t s the commercals that prmarly nfluence and explan ol prce volatlty n a postve manner. Further, non-commercals at the hgher quantle of 0.9 are found to have a sgnfcant but reducng mpact on volatlty. Ths fndng mpnges drectly upon the nature of ths commodty and llustrates the mpact that unexpected geopoltcal events can have on the prce of ol. Slvero and Szklo (2012) measured the degree of contrbuton of fnancal markets to the prce dscovery process n the ol markets. Usng a contegraton model wth error correcton, they found an ncreasng contrbuton of futures markets to prce dscovery. L et. al (2015) concluded that speculatve and hedgng trades have a domnant role n the ol markets durng the last two decades. Specfcally, speculatve trades domnated durng 2006-2011 perod and hedgng before and after ths perod. L et. al. (2016) found that hedgers and speculators tradng actvty contrbuted to the ol prce bubble of 2008. These results confrmed prevous fndngs by Dng et. al. (2014) who had argued that t was excessve fnancal speculaton n the ol futures market that destablzed crude ol spot prces and contrbuted to the speculatve bubble n the ol futures market. Excessve speculaton n ol futures has also been blamed for the hgh volatlty n ol prces. Kaufmann & Ullman (2009) examned the lnks between spot and futures ol prces. Usng a two-step DOLS error correcton model and a full nformaton maxmum lkelhood estmate for a VECM they found that after September 2004 market fundamentals ntated a long-term ncrease n ol prces that was exacerbated by speculators who recognzed an ncrease n the probablty that ol prces would rse over tme. Shanker (2017) argued that volatlty n crude ol futures market decreases wth adequate speculaton and ncreases wth excess speculaton. On the other hand, Manera et. al. (2016) examned the role of fnancal speculaton n modelng the volatlty n commodty futures prces and concluded that speculaton doesn t destablze crude ol prces. 4
The followng secton presents the data characterstcs n the ol market and the thrd secton dscusses the methodology used. The fourth secton presents and dscusses our emprcal fndngs. The fnal secton concludes ths study. 2. Data Characterstcs The data employed n ths paper s derved from the OECD database and Clarksons, a shppng database and consultng company based n the UK. Our dataset ncludes 276 monthly observatons over 1993-2016 for the prce of crude ol, the net noncommercal poston, the net- commercal poston and the effectve rate of nterest The Baltc Dry Index s taken from Clarksons. Table 1 provdes the descrpton and summary statstcs of the full dataset. TABLE 1: Data descrpton Panel A: Defnton of varables LOIL Natural Logarthm of the prce of crude ol. NONCOM (Long non- commercal poston short non- commercal poston)/ (Long + short + 2*spread) COMM FR LBDI LSP (Long commercal- short commercal)/ (Long + short) Federal Funds Effectve Rate of Interest Natural Logarthm of the Dry Bulk Index as a proxy to economc growth. Natural Logarthm of the SP500. Panel B: Descrptve statstcs LOIL NONCOM COMM LBDI LSP FR Mean 3.696146 0.075018-5.312171 7.472291 6.953415 0.028017 Medan 3.655359 0.081116-3.566103 7.290966 7.045546 0.029600 Maxmum 4.949185 0.593781 18.20051 9.237372 7.653205 0.065400 Mnmum 2.422144-0.462716-32.96261 6.251904 6.035003 0.000700 Std. Dev. 0.678052 0.158108 8.895635 0.646262 0.406117 0.022863 5
Fgure 1: The Prce of Ol, Non-Commercal Traders and Commercals over 1993-2016 5.0 LOIL 4.5 4.0 3.5 3.0 2.5 2.0 94 96 98 00 02 04 06 08 10 12 14 16.6 Net non commercal traders.4.2.0 -.2 -.4 94 96 98 00 02 04 06 08 10 12 14 16 20 COMM 10 0-10 -20-30 -40 94 96 98 00 02 04 06 08 10 12 14 16 6
3. Methodology In the presence of outlers, the senstvty of ordnary least squares estmators mght result n coeffcent estmates whch do not accurately reflect the underlyng statstcal relatonshp. Robust least squares refers to a number of regresson methods desgned to be robust, or less senstve, to outlers. S-estmaton (Rousseeuw and Yoha (1984), s a computatonally ntensve procedure that focuses on outlers n the regressor varables. Let ˆ mn ˆ ( e, e,.., e ) be the S-estmator that satsfes 1 2 n mn n 1 n y x n 2 ( we (1) nk j 1 1 ) where ˆ ˆ 1 u K=0.199, w w ( u ) and the ntal estmate s 2 u ˆ medan e medan( e 0.6745 ) The soluton s obtaned by dfferentatng (1) w.r.t β, so that n 1 y k j0 x j xj ( ) 0, j 0,1... k (2) ˆ Ψ s the dervatve of ρ. u ( u ) ( u ) u 1 ( ) c or ( u ) ( u ) 0, u c 2 2 u c In our case, the chosen set of regressors s x= (NONCOM, COMM, D(FR), LBDI, LSP) wth a dstnct outlers presence n COMM. We run ntally a regresson over the whole sample perod and then we dvded our sample nto two sub perods.one between 1993-2003 where the market even though volatle (Fgure 1) dd not experence any strong uprsng trend and one between 7
2004-20016 where the market was domnated by the so called Chna factor effect and also the sharp fall due to the 2008 crss. To further nvestgate the mpact of NONCOM as well as COMM on the ol prce formaton and ts volatlty, we resorted to quantle regresson, n order to descrbe the relatonshp between our dependent varable and the regressors at dfferent ponts of the dstrbuton of the prce of ol.the quantle regresson estmator for a quantle q mnmzes the objectve functon N : y x Q( ) q y x (1 q) y x (3) q q : y x Ths non dfferentable functon s mnmzed through the smplex method whch yelds a soluton after a fnte number of teratons. The man advantages of the quantle regresson s ts robustness to non-normal errors and outlers whle t also allows us to consder the mpact of our covarates on the entre dstrbuton of the dependent varable and not only on ts condtonal mean. q 4. Emprcal Results and Dscusson 4.1 Robust Least Squares Table 2 Regresson on the determnants of ol prce formaton over the perod 1993-2016. The table presents the results of the Robust least squares regresson analyss for D(LOIL),the frst dfference of the prce of ol, for statonarty. The symbols ***, ** and * denote statstcal sgnfcance at the 1%, 5%, and 10% levels, respectvely. The numbers reported n parentheses are Huber Sandwch S.Es. No denotes the number of observatons. Dep Var. D(LOIL) 1993-2016 Dep Var. D(LOIL) 2008-2016 Const -0.680*** -0.021 (0.128) (0.388) NONCOM 0.226*** 1.444*** (0.057) (0.522) COMM 0.003*** 0.007* (0.001) 0.004 D(FR) 0.647* 1.024* (0.364) (0.638) LBDI 0.026*** 0.053*** (0.009) (0.016) LSP 0.073*** -0.062 (0.017) (0.048) R 2 0.15 R 2 0.17 No 276 No. 96 8
From Table 2 above t appears that over the whole perod under examnaton 1993-2016 both groups commercals and non-commercals together wth economc growth and the fnancal markets sgnfcantly nfluence and n a postve manner ol prce formaton. The 2008 crss though reveals a changng role n the two tradng groups, commercals and non-commercals. We observe from the same Table 2 that the mpact of non-commercals has drastcally rsen over 2008-2016, whle the commercals have nearly become nsgnfcant. A plausble explanaton for ths mght be found n the quanttatve easng of the FED. The non-commercals may act based on fundamentals lke the commercals but qute mportant for them s also the rate of nterest. So we have a drastc reducton n the mportance of commercals due to the low economc actvty and at the same tme a rse n the non-commercals both proportonally and n terms of sgnfcance due to the low nterest rate. Table 3 Regresson on the determnants of ol prce formaton over the perod 1993-2003 and 2004-2016. The table presents the results of the Robust least squares regresson analyss for D(LOIL),the frst dfference of the prce of ol, for statonarty. The symbols ***, ** and * denote statstcal sgnfcance at the 1%, 5%, and 10% levels, respectvely. The numbers reported n parentheses are Huber Sandwch S.Es. No denotes the number of observatons. Dep Var. D(LOIL) 1993-2003 Const -0.909*** 0.004 (0.289) (0.322) NONCOM 0.208*** 0.652*** (0.071) (0.271) COMM 0.003 0.002* (0.002) (0.003) D(FR) 0.254 0.801* (0.486) (0.473) LBDI 0.049* 0.042*** (0.030) (0.013) LSP 0.083*** -0.057 (0.020) (0.042) R 2 0.23 R 2 0.15 Dep Var. D(LOIL) 2004-2016 No 129 No 144 9
Table 3 above confrms the rsng mportance of the non-commercals snce the begnnng of the Chna factor effect n 2004. The varous events untl the fnancal crss of 2008 dd not affect the contnuaton of the rsng mportance of noncommercals n the ol market. The frst drastc reducton n the prce of ol occurs wth the drop n the nternatonal trade and the fear over the Chnese economy slowdown, around 2013.Snce then the prce has never really recovered. We at the same tme experenced gradual escalaton of geopoltcal sources of tenson whch threatened market stablty and ths also dd not allow the prce of ol to resume. We then proceed to run a quantle regresson over the perod 2008-2016 n order to nvestgate further the changng roles of the commercals and non-commercal traders on the ol prce formaton and ts volatlty. 4.2 Quantle Regresson Table 4 Regresson on the determnants of ol prce formaton over the perod 2008-2016. The table presents the results of the quantle regresson analyss (tau=0.5) for D(LOIL),the frst dfference of the prce of ol, for statonarty and the varance of the error. The symbols ***, ** and * denote statstcal sgnfcance at the 1%, 5%, and 10% levels, respectvely. The numbers reported n parentheses are Huber Sandwch S.Es. No denotes the number of observatons. Dep Var. D(LOIL) Dep Var. Var(Error) Const 0.203-0.060*** (0.727) (0.003) NONCOM 1.417** 0.010 (0.708) (0.009) COMM 0.006 0.002*** (0.006) (0.0002) D(FR) 0.754 0.630 (0.787) (7.132) LBDI 0.047* 0.001 (0.003) (0.001) LSP -0.089 0.007*** (0.083) (0.0009) Pseudo R 2 0.14 Pseudo R 2 0.07 No 96 No 96 10
Table 5 Regresson on the determnants of ol prce formaton over the perod 2008-2016. The table presents the results of the quantle regresson analyss (tau=0.75) for D(LOIL),the frst dfference of the prce of ol, for statonarty and the varance of the error. The symbols ***, ** and * denote statstcal sgnfcance at the 1%, 5%, and 10% levels, respectvely. The numbers reported n parentheses are Huber Sandwch S.Es. No denotes the number of observatons. Dep Var. D(LOIL) Dep Var. Var(Error) Const -0.467-0.271*** (0.394) (0.003) NONCOM 1.376*** 0.005* (0.612) (0.003) COMM 0.009** 0.002*** (0.005) (0.0002) D(FR) 1.366** 0.310 (0.586) (0.864) LBDI 0.047** 0.002 (0.002) (0.001) LSP 0.017 0.039*** (0.056) (0.001) Pseudo R 2 0.14 Pseudo R 2 0.11 No 96 No 96 11
Table 6 Regresson on the determnants of ol prce formaton over the perod 2008-2016. The table presents the results of the quantle regresson analyss (tau=0.90) for D(LOIL),the frst dfference of the prce of ol, for statonarty and the varance of the error. The symbols ***, ** and * denote statstcal sgnfcance at the 1%, 5%, and 10% levels, respectvely. The numbers reported n parentheses are Huber Sandwch S.Es. No denotes the number of observatons. Dep Var. D(LOIL) Dep Var. Var(Error) Const -0.004-0.158 (0.379) (0.050) NONCOM 1.246*** -0.226*** (0.546) (0.072) COMM 0.009** 0.001** (0.004) (0.0007) D(FR) 1.446** -0.002 (0.638) (14.608) LBDI 0.028-0.022** (0.023) (0.009) LSP -0.021 0.057*** (0.051) (0.005) Pseudo R 2 0.09 Pseudo R 2 0.29 No 96 No 96 Tables 4-6 above reveal that the non-commercal traders even though ther sgnfcance and mportance remans strong and consstent as we move across to hgher quantles of the dstrbuton of ol prce they exert at the same tme a stablzng effect on the market, at hgh ol prces, 0.9 quantle they negatvely mpact on volatlty. The commercals on the other hand mantan a low but sgnfcant mpact and account postvely for the ol prce volatlty. It appears that nterest rate polcy by the FED has not managed to shft out of the ol market the non-commercal traders despte the recent hke whle ts volatle nature s substantally supported by the nterest of the commercal traders. 12
5. Concludng Remarks. We have explored ol prce formaton from the pont of vew of the changng roles for the commercal and non-commercal traders. We dscovered a rsng mportance n the mpact of non-commercal traders and a dmnshng mpact for the commercal traders. Ths development mght be combned wth a transfer of wealth from the fnancal sector nto the ol market nternatonally. The vehcle for ths transfer of wealth are hedge funds, prvate large speculators even states whch undertake an ncreasng role n the happenngs of the ol market. These fndngs bear mportant mplcatons for nvestors who pck ol stocks on the bass of rsk, return and effcent captal asset allocaton n the ol ndustry. Moreover, ol market regulators and captal-market partcpants can employ our results n order to rearrange or rethnk polces n the nternatonal ol market and not restrcted to, n the drecton of reducng geopoltcal conflcts, thereby protectng all provders of captal and fosterng growth n the nternatonal economy. Whle ths research has spotted and explored sgnfcant regulartes n the changng role of agents n the ol market, more work s needed n order to shed lght to organzatonal structures and shfts wthn the nternatonal ndustry whch are entangled wth geopoltcal strateges and developments. Moreover, our quanttatve approach can be complemented wth qualtatve research: ndepth ntervews wth both commercal and non-commercal traders can help us dscover the tact aspects of ther changng roles n an ever changng envronment. 13
References Alou, R., Assa M.S.B. 2016. Relatonshp between ol, stock prces and exchange rates: A vne copula GARCH method. North Amercan Journal of Economcs and Fnance 37, 458-471. Bour E., Chen, Q., Len, D., Lv, X. 2017. Causalty between ol prces and the stock market n Chna: The relevance of the reformed ol product prcng mechansm. Internatonal Revew of Economcs and Fnance 48, 34-48. Dng, H., Km, H., Park, S., 2014. Do net postons n the futures market cause spot prces of crude ol? Economc Modellng 41, 177-190. Ewng, B., Malk, F., 2016. Volatlty spllovers between ol prces and the stock market under structural breaks. Global Fnance Journal 29, 12-23. Fan, Y., Xu, J.H., 2011. What has drven ol prces snce 2000? A structural change perspectve. Energy Economcs 33, 1082-1094. Hamlton, J. 2009a. Causes and consequences of the ol shock of 2007-08. Brookngs Papers of Economcs Actvty, 2009 (1), 215-261. Hamlton, J. 2009b. Understandng crude ol prces. Energy Journal 30, no. 2, 179-206. Kaufmann, R., Ullman, B., 2009. Ol Prces, speculaton, and fundamentals: Interpretng causal relatons among spot and futures prces. Energy Economcs 31, 550-558. L, H., Km, M. J., Park, S.Y. 2016. Non-lnear relatonshp between crude ol prce and net futures postons: A dynamc condtonal dstrbuton approach. Internatonal Revew of Fnancal Analyss 44, 217-225. L, H., Km, H., Park, S.Y. 2015. The role of fnancal speculaton n the energy future markets: A new tme-varyng coeffcent approach. Economc Modellng 51, 112-122. Lu, L., Wang, Y., Wu, C., Wu, W. 2016. Dsentanglng the determnants of real ol prces. Energy Economcs 56, 363-373. Manera, M., Ncoln, M., Vgnat, I., 2016. Modellng futures prce volatlty n energy markets: Is there a role for fnancal speculaton? Energy Economcs 53, 220-229. Mens, W., Hammoudeh, S., Sahsad, S., Shahbaz, M. 2017. Modelng systemc rsk and dependence structure between ol and stock markets usng a varatonal mode decomposton-based copula method. Journal of Bankng and Fnance 75, 258-279. Shanker, L., 2017. New ndces of adequate and excess speculaton and ther relatonshp wth volatlty n the crude ol futures market. Journal of Commodty Markets. (Artcle n press). Slvero, R., Szklo, A. 2012. The effect of the fnancal sector on the evoluton of ol prces: Analyss of the contrbuton of futures market to the prce dscovery process n the WTI spot market. Energy Economcs 34, 1799-1808. 14
15