Do Speculators in Futures Markets Make Cash Markets. More Volatile?

Transcription

1 Do Speculaors in Fuures Markes Make Cash Markes More Volaile? Yingzi Li and T. Randall Forenbery School of Economic Sciences Washingon Sae Universiy PO Box Pullman, WA Seleced paper prepared for presenaion a he Agriculural & Applied Economics Associaion s 213 AAEA & CAES Join Annual Meeing, Washingon, DC, Augus 4-6, 213. Copyrigh 213 by Yingzi Li & T. Randall Forenbery. All righs reserved. Readers may make verbaim copies of his documen for non-commercial purposes by any means, provided ha his copyrigh noice appears on all such copies.

2 Do Speculaors in Fuures Markes Make Cash Markes More Volaile? Yingzi Li and T. Randall Forenbery* * Graduae Suden and Tom Mick Chair, School of Economic Sciences, Washingon Sae Universiy

3 Absrac Do Speculaors in Fuures Markes Make Cash Markes More Volaile? This paper invesigaes he exen o which speculaive rade in fuures markes conribues o volailiy in cash markes. By analyzing coffee, crude oil and whea we find ha fuures and cash prices are coinegraed in levels and exhibi bi-direcional causaliy in variance. Thus, facors causing higher fuures price volailiy will also cause higher cash price volailiy. Resuls sugges increases in speculaive aciviy are associaed wih decreases in fuures price volailiy, hus cash price volailiy. On balance i appears ha policies which limi speculaive rade conribue o de-sabilizing cash prices, raher han reducing volailiy as inended. Key words: speculaion, cash price volailiy, coinegraion, causaliy in variance, crude oil, whea, coffee

4 Do Speculaors in Fuures Markes Make Cash Markes More Volaile? Inroducion Beginning in lae 27, mos commodiy markes experienced an increase in average price levels, accompanied by higher volailiy (Figure 1). Several sudies have aemped o explain his price behavior. Many have focused on he fuures markes of he underlying commodiies. Some of hem blame speculaors for he recen price acion based on he observaion ha speculaive posiions increased significanly before price increases were observed (Masers, 28; Masers, 21; Singleon, 212). Influenced by hese accusaions, he U.S. Commodiy Fuures Trading Commission (CFTC) approved limis on he size of speculaive posiions for 28 core physical commodiies in Ocober 211, inending o miigae speculaive influence in fuures markes. These limis immediaely sirred debae regarding heir necessiy or effeciveness in managing speculaors and he poenial adverse effec on commercial eniies ha use derivaives o hedge price risk. Several oher sudies failed o find causaliy beween speculaion and price movemens and concluded ha speculaors do no desabilize fuures markes (Brunei and Büyükşahin, 29; Büyükşahin and Harris, 211; Sanders and Irwin, 211; Hamilon and Wu, 212; Irwin and Sanders, 212). These appear o have had less influence in driving recen policy iniiaives. Despie he number of previous sudies, he body of he research is no complee. Mos of he earlier sudies focused on speculaive influences on price levels raher han price volailiy. Furher, hey were direced a speculaive influences on 2

5 commodiy fuures prices and did no explicily examine he influence on cash prices. In his paper we examine wheher speculaion in fuures markes conribues o increased volailiy in cash prices. This is an imporan consideraion because cash price volailiy reflecs he price risk faced by boh producers and consumers of he physical commodiy and much of he debae abou speculaive influence is really a debae abou he way oher marke paricipans are impaced by fuures price aciviy. The focus of his work is on crude oil 1, whea, and coffee. These producs differ in erms of commodiy caegory (hey cover energy, and boh impored and expored food suffs, as well as boh hin and deep fuures markes), and he characerisics of heir fuures markes. Crude oil is he larges naural resource commodiy in fuures rading, whea fuures conracs are one of he oldes and mos acively raded agriculural fuures in he U.S., and he coffee fuures marke represens boh a hin marke and one wih no domesic producion. Furher, he poenial of he coffee fuures marke o impac he price risk faced by cash marke paricipans has already been alluded o Forenbery and Zapaa (24). Lieraure Review Masers(28; 21) and Singleon (212) argued speculaors were a major driver in he 28 run-up in commodiy fuures, paricularly energy fuures prices. Their conclusions were based on observing increases in speculaive fuures posiions prior o observing fuures price increases. They essenially observed correlaion wih posiion changes leading price changes, bu did no rigorously es for causaliy. 1 For crude oil, spo marke and spo price are ofen used insead of cash marke and cash price. Bu in his paper we do no inerchange he words and use cash marke and cash price for all he hree commodiies. 3

6 Several oher sudies inroduced more rigor o es for he exisence of excessive speculaion. Alquis and Gervais (211) used Working s T-index (Working, 196) o es for excessive speculaion and found ha he index did increase in 28 when he price of oil increased dramaically. However, his index also reached similar levels in 23 and 25 when oil prices were low. Moreover, he index was low a he end of 21 when he ne long non-commercial posiions in fuures markes was high, suggesing ha speculaive pressures were subdued by hedging demand from commercial firms. Their findings suggesed ha i may be misleading o claim excessive speculaion merely by noicing a high level of speculaive posiions. Similar resuls were presened by Ripple (28) and Büyükşahin and Harris (211). Irwin and Sanders (212) poined ou ha Singleon s measure of index fund posiions in oil fuures was in fac inferred from CFTC daa on agriculural fuures which had lile relaion o index funds acual posiions in oil. Hamilon and Wu (212) demonsraed ha he agriculural index fund posiions used by Singleon prediced he fuures price of oil more accuraely han he fuures price of agriculural commodiies. Moreover, heir model also prediced he U.S. sock marke. They argued, herefore, ha he posiive predicive correlaion found by Singleon (212) on he basis of a very shor sample period was probably driven by he 28 recession. Hamilon and Wu (212) exended he sample period by wo years and found he predicive correlaion breaks down. Some sudies have analyzed speculaive price influences using more deailed non-public daa. Brunei and Büyükşahin (29) employed he CFTC Large Trader 4

7 Reporing Sysem (LTRS) which offers unique, highly disaggregaed posiion-level daa o analyze five fuures markes: crude oil, naural gas, corn, hree-monh Eurodollars and he mini-dow. They considered boh reurns and volailiy and concluded ha speculaive rading aciviy reduced fuures price volailiy. Brunei, Büyükşahin e al. (211) sudied specific caegories of raders and esed wheher posiions aken by speculaors, such as hedge funds and swap dealers, caused changes in oil fuures prices or price volailiy. Their resuls were consisen wih speculaors providing liquidiy o he marke and reacing o marke condiions raher han vice versa. Büyükşahin and Harris (211) also employed CFTC LTRS o es he relaion beween crude oil prices and rading posiions of various ypes of raders in he crude oil fuures marke. Using Granger causaliy ess beween price and posiion daa a daily and muliple day inervals, hey found lile evidence ha he non-commercial (speculaive) posiion changes Granger-cause price changes; insead, hey suggesed ha price changes preceded changes in speculaive posiions. Similarly, Sanders and Irwin (211) found no saisically significan relaionship beween growh in he volume of oil fuures conracs and oil fuures reurns, realized volailiy or implied volailiy. The above sudies focused on he relaionship beween speculaion and fuures prices. As far as we know no sudies have direcly esed he speculaive influence on commodiy cash price volailiy in he curren marke environmen even hough he role of fuures speculaion on cash marke volailiy has been sudied and described in 5

8 earlier work. Furher, i is no reasonable o informally exend recen fuures marke work o cash markes direcly since shor erm flucuaions in commodiy fuures prices may no lead o cash price insabiliy. For example, Alquis, Kilian e al. (211) examined he ou-of-sample accuracy of daily and monhly oil fuures prices and found no compelling evidence ha oil fuures prices help forecas he oil spo price. Figlewski (1981), Chen, Cuny e al. (1995) and Bessembinder and Seguin (1992) found a posiive conemporaneous associaion beween differen cash prices and heir corresponding fuures marke rading aciviies. Neverheless, he findings could no be used as evidence of fuures speculaion causing higher cash price volailiy because correlaion is no causaion (Figlewski, 1981). Kamara (1982) found ha he inroducion of commodiy fuures rading generally reduced or a leas did no increase cash price volailiy. Anoniou and Foser (1992) and Gulen and Mayhew (2) considered ime-varying paerns of price volailiy and came o a similar conclusion. These sudies all considered he impac of a new fuures marke on cash price sabiliy. There are oher sudies focused on he effecs of differen levels of fuures rading aciviy on cash marke volailiy. Darra and Rahman (1995) repored no evidence of causaliy running from S&P 5 fuures rading (boh volume and open ineres) o cash price volailiy. By conras, Charah, Ramchander e al. (1996) argued ha currency fuures rading (rading volume) had a significan posiive (and hence desabilizing) causal impac on he cash price volailiy. Adrangi and Charah (1998) repored ha surges in he paricipaion of large speculaors and small raders 6

9 desabilize exchange rae volailiy. Earlier work focusing on he impac of fuures rading aciviy on cash price volailiy direcly used a variable (rading volume or open ineres) in one marke and a variable (cash price volailiy) in anoher marke wihou firs deermining he acual pricing relaionship beween he wo markes. We argue ha his kind of esing suffers from informaion loss. Insead of esing he relaionship of speculaion in fuures markes on cash price volailiy direcly, we firs es wheher here is a causal relaionship in boh mean and variance beween he fuures and cash prices for one commodiy. In cases where volailiy spillover exiss, we hen es wheher speculaors aciviies in he fuures marke affec fuures price volailiy. If he increase of speculaive posiions in he fuures marke increases (decreases) he fuures price volailiy, hen i will also increase (decrease) he corresponding cash price volailiy when volailiy spillover effecs are found. Daa and Mehodology Daa The analysis presened here focuses on hree markes: coffee, crude oil, and whea. Fuures and cash prices for all hree markes are daily prices provided by he Commodiy Research Bureau (CRB). The coffee fuures price is for he nearby Coffee C conrac raded on he Inerconinenal Exchange (ICE). I is he world benchmark for Arabica coffee. The coffee fuures price series is coninuous and he rollover from conrac o conrac akes place on he firs business day of each delivery monh. 7

10 The crude oil fuures price is for he second nearby Bren Crude conrac also raded on ICE. I is generally acceped as he world s crude oil benchmark. The second nearby is chosen because here is a fuures conrac for oil delivery every monh. Thus, he nearby is always for delivery in he curren monh. The whea fuures price is for he #2 Sof Red Winer Whea conrac raded a he Chicago Mercanile Exchange (CME). Similar o coffee, we use he nearby conrac o develop a coninuous fuures price series ha rolls over on he firs business day of each delivery monh. To deec long-run relaionships, he iniial sample period runs from January 1, 199 hrough January 23, 212. Since he differen markes vary slighly in rading days, he numbers of observaions for he hree commodiies are no equal. There are 5,514, 5,59 and 5,553 observaions coffee, oil, and whea, respecively. The fuures marke posiion daa are from he CFTC Commimens of Traders repors (COT). These repors provide each Tuesday s open ineres for markes in which 2 or more raders hold posiions equal o or above he reporing levels esablished by he CFTC. Causaliy in Variance Following Granger, Robins e al. (1986) who discuss causaliy in variance, we es wheher fuures price Granger-causes cash price in variance in he following way. Firs, wo informaion ses are considered: In : CPn j, j, and Jn : CPn j, FPn j, j, where CP sands for cash price and FP for fuures price. 8

11 Second, he fuures price Granger-causes cash price in variance if: n n n n n n n n E CP E CP J I E CP E CP J J The reverse relaionship from he cash price o he fuures price is defined similarly. We use he mulivariae generalized auoregressive condiional heeroscedasiciy model (M-GARCH model) o es for he causaliy in variance beween fuures and cash prices: X E X J 1 where X is a 2 1 vecor of he fuures and cash prices and 1 E X J is a 2 1 vecor of condiional means of he wo prices given he informaion se. is he error erm, which will be modeled as a GARCH (1,1) BEKK represenaion (Engle and Kroner, 1995). This leads o: N, H H C C A A G H G ' ' ' ' where h FP and h CP are condiional variances of he fuures and cash prices, respecively. Expanding his expression gives he condiional variance for he fuures price: h c a 2a a a g h 2g g h g h FP, 1 11 FP, FP, 1 CP, 1 21 CP, 1 11 FP, FPCP, 1 21 CP, 1 Therefore, he cash price does no cause he fuures price in variance if and only if 9

12 a21 and g21. Similarly, he condiional variance for he cash price is: h c a 2a a a g h 2g g h g h CP, 2 12 FP, FP, 1 CP, 1 22 CP, 1 12 FP, FPCP, 1 22 CP, 1 The fuures price does no cause he cash price in variance if and only if a12 and g12. In order o esimae he parameers in he GARCH (1,1) model and es for he causaliy in variance, we need o firs model he price means, 1 E X J. If he wo price series (cash and fuures) are saionary, we could use a vecor auoregression model (VAR) for he means. If hey are non-saionary, we could ake differences and hen esimae a VAR on he differences. However, if he commodiy price pairs are no saionary bu are inegraed of he same order, hen we could esimae a coinegraion model (in he error correcion model ECM form). I is desirable o model he price means using a coinegraion model because i will allow us o idenify he long-run relaionship beween he means of he wo price series. If wo variables are coinegraed, hey have a long-run equilibrium relaionship and are moving ogeher. Furher, he direcion of causaliy and he speed of adjusmen o price shocks in eiher marke can be esimaed. In his case, parameer esimaion consiss of wo seps. Firs, we use maximum likelihood esimaion o obain consisen esimaes of he parameers in he mean equaion in he presence of GARCH effecs; and second he GARCH model is esimaed wih he parameers in he mean equaion as given. If he fuures price 1

13 Granger-causes he cash price in variance, we conduc he wo samples -ess o see wheher he increase in speculaors posiions in he fuures marke leads o an increase or a decrease in he fuures price volailiy (and hence cash price volailiy). Coinegraion Analysis Consider a fuures marke and a cash marke for he same underlying commodiy. I is reasonable o expec he wo markes reac similarly o new marke informaion. Coinegraion models (in he ECM form) have been widely used o es wheher his is in fac rue. Based on price saionariy ess (repored below), we esimae bi-variae coinegraion models for each of he commodiy markes o idenify wheher he fuures and cash markes are moving ogeher in he long-run The model used o examine he coinegraion relaions is based on Johansen and Juselius (199). The ECM specificaion is: Under his specificaion, is IN, p 2 ; X is a p 1 vecor of endogenous variables, 1,..., k 1,, are marices of parameers o be esimaed, Xk 1,..., X (1) are fixed wih k corresponding o he lag lengh in he VAR(k) model, D conains deerminisic variables (i.e. dummies, ec.), and is he error correcion erm * X k where * X k conains X and consan, rend or dummy variables ha belong o he long-run equilibrium. 2 The error erms are modeled as GARCH (1,1) raher han independen normal disribuion, so he coinegraion model assumpion is violaed. However, Manalos (21) and Cavaliere, Rahbek e al. (21) showed ha he coinegraion es wih GARCH errors is consisen wih a large daa se. 11

14 The es for coinegraion depends on he rank of he marix. If has full rank (r=p), hen he vecor process X is saionary. If he rank of is, hen he ECM corresponds o a radiional differenced vecor ime series model. If he rank of lies beween he wo exreme cases, hen here are r coinegraing vecors among X. To formally es he rank of, we use boh he likelihood raio es, ofen called he race es or he Johansen es, and he maximum eigenvalue es. The es saisics are: pr T p ln 1 ˆ i, and ˆ r T ln 1 r 1 ir1 In he race es saisic formula, ˆi is he h i larges eigenvalue of marix. The null hypohesis of he es is ha here are r coinegraing relaions (herefore p-r common sochasic rends). The alernaive hypohesis is ha here are a leas r+1 coinegraing relaions (herefore a mos p-r-1 common sochasic rends). We selec he number of coinegraing vecors based on he following crierion: r r C and C when pr1 pr1 pr pr where Cp r is he criical value under he null hypohesis ha here are r coinegraing vecors. In he maximum eigenvalue es saisic formula, ˆr 1is he r 1 h larges eigenvalue of marix. The null and alernaive hypoheses are he same as ha for he race es. The decision rule is given as: 12

15 r r C and C when r1 r1 r r where Cr is he criical value under he null hypohesis ha here are r coinegraing vecors. Two Samples -es To focus on he period when commodiy prices began o exhibi increased volailiy, he sample period for he wo samples -es is from he firs week of 27 o he las week of 211. We use boh fuures and opions posiion informaion. Three differen posiion variables are used as measures of speculaors aciviies. Firs, we use he oal non-commercial open ineres (NonComm) as he measure of he absolue number of speculaive posiions. Using his measure allows us o es Maser s asserion concerning speculaors price influence because his argumen appears o be based on observing increases in his measure before observing increases in prices. Second, we use he percenage of non-commercial oal open ineres relaive o oal marke open ineres (PCofNonComm) as a measure of he marke share of speculaive posiions. Wiherspoon (1993) argued ha when he posiions of agens rading exclusively in he fuures marke (speculaors) exceeded hose rading in he cash marke (hedgers) beyond some boundary level cash prices will become more volaile. PCofNonComm is an appropriae variable o measure he relaive speculaion suggesed by Wiherspoon s heory. If his heory is correc, hen i should be he percenage of speculaive posiions raher han he absolue number of speculaive 13

16 posiions ha influence cash price volailiy. Third, we use Non-commercial ne-long open ineres (NonCommNeL) o measure how many more long posiions compared o shor posiions speculaors hold. Some scholars argue ha he recen increase in speculaive posiions is mainly refleced by increases in long posiions. The argumen is ha his brings significan buying pressure o he marke and leads o an increase in boh price levels and price volailiy. Using NonCommNeL allows us o es wheher he increase of he speculaive rade on he long side makes fuures prices more volaile. We use four saisics o measure fuures price volailiy o check he sensiiviy of he resuls: weekly variance, weekly realized volailiy, absolue weekly reurn and weekly rading range. Since he COTs repor each Tuesday s posiions, we coun each week from Wednesday o he nex Tuesday. Mos of he weeks have five rading days, bu some weeks have days wihou rading and hence conain only four, hree or even wo rading days. The weekly variance is he sample variance of he fuures price for each week: Variance N i1 P i N P, where Variance is he sample weekly variance of he h week, N is he number of rading days of his week, i P is he fuures price of hei h rading day of his week, and P is he average price of his week. The realized price volailiy is calculaed following Meron (198): 14

17 realized volailiy N i1 i r 2, where r ln P ln P is he i i i 1 h i rae of reurn in week. Absolue reurn is also ofen used a measuremen of price volailiy (Halova 212). I is calculaed as: absolue reurn P P, N 1 ln ln where P N is he fuures price of he las day of week and 1 P is he fuures price of he firs day of ha week. Using range as anoher measure of volailiy is discussed in Corrado and Truong (27). I is defined as: range P P, max min where max P and min P are he highes and lowes prices in week, respecively. Our ineres is in esing wheher an increase in speculaive posiions is associaed wih an increase in he fuures price volailiy. Therefore, for each of he hree posiion variables, we ake he difference beween he adjacen Tuesday s posiions o consruc he series of posiion changes. Thus, we ge four groups of volailiy measures which correspond o a decrease in he posiions and also four groups of volailiy measures which correspond o an increase in he posiions for each of he posiion variables. If he increase in he speculaors posiion causes he fuures marke o become 15

18 more volaile, we will expec he average fuures price volailiy in he weeks which experience an increase in he speculaors posiion o be greaer han he average fuures price volailiy in he weeks which experience a decrease in speculaors posiions. Then we can formulae a wo samples -es wih he hypohesis ha he mean in he increased speculaor posiion sample is greaer han he mean in he decreased speculaor posiion sample. The es saisic is: vol sd vol increase in posiions increase in posiions vol vol decrease in posiions decrease in posiions The null hypohesis is he negaion of he research hypohesis, i.e., he mean weekly volailiy when speculaor posiions increase is less han he mean weekly volailiy when speculaor posiions decrease. Resuls Uni Roo Tess As discussed above, we need o firs es wheher he price series are saionary o deermine which model o use for undersanding he relaionship beween mean prices. If hey are non-saionary we need o deermine wheher each pair of prices are inegraed of he same order o decide wheher coinegraion is appropriae for idenifying he relaionships beween fuures and cash price means. Three ypes of uni roo ess are implemened: Dickey-Fuller (DF), Augmened Dickey-Fuller (ADF), and he Phillips-Perron (PP) ess. Moreover, we use hree models for he DF and ADF ess: models wihou inerceps or rends, models wih 16

19 inerceps, and models wih inerceps and rends. We use wo models for he PP es: models wih inerceps and models wih inerceps and rends. The null hypoheses of he uni roo ess are ha uni roos exis. Table 1 gives he resuls of he uni roo ess. We canno rejec he null hypohesis for any of he six price series; however, he firs-order differences of he price series are saionary. This means he series are I(1) and hus coinegraion is an appropriae es for evaluaing price relaionships beween cash and fuures for all hree commodiies. Lag-lengh deerminaion of he ECM for he mean The deerminaion of he opimal lag lengh of he ECMs is idenified based on he Schwarz Bayesian (SB) and Hannan-Quinn (H-Q) informaion crieria. Because he informaion crieria are based on differen penalies, hey do no need o sugges he same lag lengh. The SB crierion ends o penalize more for adding variables ino he model. However, he decision rules for boh crieria are he same: he smaller value, he beer model. We chose he model wih he smalles H-Q or/and he smalles SB. As Table 2 shows, we chose lag lenghs of 6, 7, and 4 for coffee, oil and whea, respecively. Coinegraion Tess Using he above lag lenghs, esimaion of he ECMs and ess for coinegraion were conduced. In all hree bi-variae models, he error-correcion erm * X k includes a consan. For coffee and whea, a dummy variable is used wih 1 indicaing he rollover dae and for ohers o accoun for he rollover effec. 17

20 The coinegraion es resuls from he race as well as he maximum eigenvalue ess are shown in Table 3. The resuls confirm our inuiion ha cash and fuures prices move ogeher in he long run. The race es saisic for coffee is significan a 1% level. All oher saisics are significan a 5% level. Causaliy in variance Table 4 gives he resuls of he parameer esimaes in he GARCH model. The parameers used o es for causaliy in variance from fuures price o cash price, a12 and g 12, and he ones used o es for causaliy in variance from cash price o fuures price, a 21 and g 21, are all significanly differen from. The means here is a bi-direcional causal relaionship in variance beween he fuures and cash prices of coffee, oil, and whea. Thus, if speculaive aciviy is found o resul in increased volailiy in fuures prices, hen here will also be volailiy spillover o he cash marke. This, in urn, suggess ha cash marke paricipans will face increased price risk as a resul of fuures raders speculaive aciviy. Two Samples -es Tables 5.1, 5.2 and 5.3 give he resuls for he wo samples -ess. Panels a, b and c of each able show he resuls wih he posiion variables NonCommercial, PCofNonAll and NonCommercialNeLong, respecively. The firs half of each panel shows he resuls for he null hypohesis Volailiy is greaer when speculaors posiion decreases han when i increases ; while he lower half of each panel shows he resuls for esing he opposie saemen Volailiy is greaer when speculaors 18

21 posiion increases han when i decreases. The resuls demonsrae ha when we use NonCommercial and PCofNonAll as he posiion variables, for all hree commodiies and for all four volailiy measures, we canno rejec he firs hypohesis ha Volailiy is greaer when speculaors posiions decrease han when hey increase, and almos all he ess 3 for he second hypohesis ha Volailiy is greaer when speculaors posiions increase han when hey decrease rejec he null hypohesis. This suggess ha he saemen Volailiy is greaer when speculaors posiions decrease han when hey increase is rue. Since here is volailiy spillover from he fuures marke o he cash marke, hen we can furher conclude ha increased levels of speculaive aciviy in fuures markes helps o reduce cash price volailiy, or a he very leas does no conribue o increased cash price volailiy. However, when we use NonCommercialNeLong o measure speculaor aciviy, he resuls are inconsisen. For coffee, we canno rejec any of he wo opposie hypoheses using any of he four volailiy measures. This means we canno conclude ha he change in he non-commercial ne long posiions will influence coffee fuures price volailiy. Oil and whea are similar excep ha when absolue weekly reurn is used as he volailiy measure, he hypohesis ha Volailiy is greaer when speculaors posiions decrease han when hey increase is rejeced. However, given ha i is rejeced for he oher measures i is likely ha increases in non-commercial ne long posiions has no effec on oil and whea fuures price volailiies eiher. 3 Only in he es for oil using realized volailiy fails o rejec he hypohesis ha Volailiy is greaer when speculaors posiion increases han when i decreases. 19

22 Because high fuures price volailiy will cause high cash price volailiy, his also means ha he cash markes are likely unaffeced when non-commercial ne long posiions increase. Speculaors in he fuures marke play an imporan role in providing liquidiy o he marke. However, he exen o which heir posiive marke conribuions are diminished, or even urn negaive, as heir marke exposure increases has been acively debaed in recen years. Based on he resuls here, we find ha from 27 o 211, when commodiy prices were experiencing increased volailiy, relaive o earlier ime periods, increases in he oal number of speculaive posiions or he percenage of speculaive posiions relaive o he overall marke size was associaed wih a decrease in weekly price volailiy. Changes in ne long posiions of speculaors appear o have no effec on price volailiy. This indicaes, a leas for he markes considered here, ha speculaive paricipaion in fuures markes has provided liquidiy while no exceeding he boundary idenified by Wiherspoon. Thus, speculaive behavior has no played a desrucive role in commodiy price formaion. On balance, he resuls sugges ha policies focused on limiing speculaive aciviy will likely be more harmful o he marke, as opposed o conribuing o an increase in marke sabiliy. Conclusions This paper examines wheher speculaors aciviies in crude oil, whea and coffee fuures markes make cash prices of hese commodiies more volaile. The 2

23 conclusions are similar across he hree differen commodiies. Firs, he fuures and cash prices of each of he hree commodiies are coinegraed. Second, here exiss bi-direcional volailiy spillover beween he fuures and cash prices for all he hree commodiies bu no evidence is been found o suppor he hypohesis ha increases in speculaive posiions increases fuures price volailiy, hus hey do no impac cash price volailiy. In fac, here is srong evidence suggesing ha increases in speculaive posiions acually conribue o decreased fuures price volailiy. As a resul, he CFTC limis on he size of speculaive posiions for 28 core physical commodiies approved in Ocober 211 is no expeced o conribue o he sabilizaion of commodiy prices, eiher in fuures or cash markes. In order for public policy iniiaes o impac marke volailiy in a posiive way a more complee undersanding of he drivers of recen price volailiy is necessary. 21

24 References Adrangi, B. and A. Charah (1998). "Fuures Commimens and Exchange Rae Volailiy." Journal of Business Finance & Accouning 25(3/4): Alquis, R. and O. Gervais (211). "The Role of Financial Speculaion in Driving The Price of Crude Oil." Working Paper. Bank of Canada. Alquis, R., L. Kilian, e al. (211). "Forecasing he Price of Oil." Board of Governors of he Federal Reserve Sysem Inernaional Finance Discussion Papers. Anoniou, A. and A. J. Foser (1992). "The Effec of Fuures Trading on Spo Price Volailiy: Evidence for Bren Crude Oil Using Garch." Journal of Business Finance & Accouning 19(4): Büyükşahin, B. and J. H. Harris (211). "Do Speculaors Drive Crude Oil Fuures Prices?" Energy Journal 32(2): Bessembinder, H. and P. J. Seguin (1992). "Fuures-Trading Aciviy and Sock Price Volailiy." The Journal of Finance 47(5): Brunei, C. and B. Büyükşahin (29). "Is Speculaion Desabilizing?" Working paper, Commodiy Fuures Trading Commission. Brunei, C., B. Büyükşahin, e al. (211). "Speculaors, Prices, and Marke Volailiy." Working paper. Johns Hopkins Universiy. Cavaliere, G., A. Rahbek, e al. (21). "Coinegraion Rank Tesing Under Condiional Heeroskedasiciy." Economeric Theory 26(6): Charah, A., S. Ramchander, e al. (1996). "The Role of Fuures Trading Aciviy in Exchange Rae Volailiy." Journal of Fuures Markes 16(5): Chen, N.-f., C. J. Cuny, e al. (1995). "Sock Volailiy and he Levels of he Basis and Open Ineres in Fuures Conracs." The Journal of Finance 5(1): Corrado, C. and C. Truong (27). "Forecasing Sock Index Volailiy: Comparing Implied Volailiy and he Inraday High Low Price Range." Journal of Financial Research 3(2): Darra, A. F. and S. Rahman (1995). "Has Fuures Trading Aciviy Caused Sock Price Volailiy?" Journal of Fuures Markes 15(5):

25 Engle, R. F. and K. F. Kroner (1995). "Mulivariae Simulaneous Generalized Arch." Economeric Theory 11(1): Figlewski, S. (1981). "Fuures Trading and Volailiy in he GNMA Marke." The Journal of Finance 36(2): Forenbery, R. and H. Zapaa (24). "Developed Speculaion and Underdeveloped Markes The Role of Fuures Trading on Expor Prices in Less Developed Counries." European Review of Agriculural Economics 31(4). Granger, C. W. J., R. P. Robins, e al. (1986). "Wholesale and Reail Prices : Bivariae Time-Series Modeling wih Forecasable Error Variances." Model reliabiliy. - Cambridge, Mass. [u.a.] : MIT Press, ISBN : Gulen, H. and S. Mayhew (2). "Sock Index Fuures Trading and Volailiy in Inernaional Equiy Markes." Journal of Fuures Markes 2(7): Hamilon, J. D. and J. C. Wu (212). "Effecs of Index-Fund Invesing on Commodiy Fuures Prices." Working paper. Universiy of California a San Diego. Irwin, S. H. and D. R. Sanders (212). "Tesing he Masers Hypohesis in commodiy fuures markes." Energy Economics 34(1): Johansen, S. and K. Juselius (199). "Maximum Likelihood Esimaion and Inference on Coinegraion--Wih Applicaions o he Demand for Money." Oxford Bullein of Economics & Saisics 52(2): Kamara, A. (1982). "Issues in Fuures Markes: A Survey." Journal of Fuures Markes 2(3): Manalos, P. (21). "ECM-Coinegraion Tes wih GARCH(1,1) Errors." hp://inersa.sajournals.ne/. Masers, M. W. (28). "Tesimony before he U.S. Senae Commiee on Homeland Securiy and Governmenal Affairs." Masers, M. W. (21). "Tesimony before he Commodiy Fuures Trading Commission." Meron, R. C. (198). "On Esimaing he Expeced Reurn on he Marke: An Exploraory Invesigaion." Naional Bureau of Economic Research Working Paper Series No Ripple, R. D. (28). "Fuures Trading: Wha Is Excessive." The oil and gas journal 23

26 16: 24. Sanders, D. R. and S. H. Irwin (211). "The Impac of Index Funds in Commodiy Fuures Markes: A Sysems Approach." Journal of Alernaive Invesmens 14(1): Singleon, K. J. (212). "Invesor Flows and he 28 Boom Bus in Oil Prices." Working paper. Sanford Universiy. Wiherspoon, J. T. (1993). "How Price Discovery by Fuures Impacs he Cash Marke." Journal of Fuures Markes 13(5): Working, H. (196). "Speculaion on Hedging Markes." Sanford Universiy Food Research Insiue Sudies 1. 24

27 199/1/2 1991/1/2 1992/1/2 1993/1/2 1994/1/2 1995/1/2 1996/1/2 1997/1/2 1998/1/2 1999/1/2 2/1/2 21/1/2 22/1/2 23/1/2 24/1/2 25/1/2 26/1/2 27/1/2 28/1/2 29/1/2 21/1/2 211/1/2 1/199 12/199 11/1991 1/1992 9/1993 8/1994 7/1995 6/1996 5/1997 4/1998 3/1999 2/2 1/21 12/21 11/22 1/23 9/24 8/25 7/26 6/27 5/28 4/29 3/21 2/211 Figure 1 Annual food price index and crude oil price from 199 Jan o 211 Dec Food price index Daa source: Food and Agriculure Organizaion of he Unied Naions (22-24=1) Crude oil Daa source: Commodiy Research Bureau 25

28 Table 1 Uni Roo Tes Coffee Oil Whea Cash Price Δ Cash price Fuures price Δ Fuures price Cash Price Δ Cash price Fuures price Δ Fuures price Cash Price Δ Cash price Fuures price Δ Fuures price DF Tes ADF Tes PP Tes Wihou inercep or rend Wih inercep Wih inercep and rend Wihou inercep or rend Wih inercep Wih inercep and rend Wih inercep Wih inercep and rend Noe: The Dickey-Fuller es, Augmened Dickey-Fuller es and Phillips-Perron es share he same 5% criical value for he model wihou inercep or rend, model wih inercep and model wih inercep and rend. These are , , and respecively. The hree ess consisenly show ha he cash and fuures prices of he hree commodiies are no saionary. However, he firs-order differences of all he price series show srong saionariy. Therefore, all he price series are I(1) series. 26

29 Table 2 Lag lengh deerminaion Model k T SB H-Q VAR(1) VAR(9) VAR(8) Coffee VAR(7) VAR(6) VAR(5) VAR(4) VAR(3) VAR(2) VAR(1) VAR(1) VAR(9) VAR(8) Oil VAR(7) VAR(6) VAR(5) VAR(4) VAR(3) VAR(2) VAR(1) VAR(1) VAR(9) VAR(8) Whea VAR(7) VAR(6) VAR(5) VAR(4) VAR(3) VAR(2) VAR(1)

30 Table 3 Trace and Eigen Value Tess of Coinegraion Coffee Oil Whea p-r r Eig. Value Trace max * ** ** ** ** ** Noe: Given by Johansen and Juselius (199), he 5% criical value for esing he null hypohesis of r= and r=1 are and in he race es; and and 9.94 in he maximum eigenvalue es. **: significan a 5% level. * : significan a 1% level. 28

31 Table 4 Causaliy in Variance Coffee Oil Whea Parameer Esimae P-value Esimae P-value Esimae P-value c c c a a a a b b b b

32 Table 5 Resuls for he wo samples ess (Daa: ) 5.1 Coffee a. Posiion daa use NonCommercial all T.S. Pvalue Conclusion H : Var Var Fail o rejec H : Range Range H : AbsReurn AbsReurn H : RealizedVol RealizedVol H : Var Var H : Range Range H : AbsReurn AbsReurn H : RealizedVol RealizedVol Fail o rejec Fail o rejec Fail o rejec Rejec Rejec Rejec Rejec H : Var Var b. Posiion daa use PCofNonAll T.S. Pvalue Conclusion Fail o rejec H : Range Range H : AbsReurn AbsReurn H : RealizedVol RealizedVol H : Var Var H : Range Range H : AbsReurn AbsReurn H : RealizedVol RealizedVol Fail o rejec Fail o rejec Fail o rejec Rejec Rejec Rejec Rejec 3

33 c. Posiion daa use NonCommercialNeLong T.S. Pvalue Conclusion H : Var Var Fail o rejec H : Range Range H : AbsReurn AbsReurn H : RealizedVol RealizedVol H : Var Var H : Range Range H : AbsReurn AbsReurn H : RealizedVol RealizedVol Fail o rejec Fail o rejec.51 Fail o rejec Fail o rejec Fail o rejec Fail o rejec.499 Fail o rejec 31

34 5.2 Crude Oil a. Posiion daa use NonCommercial all T.S. Pvalue Conclusion H : Var Var Fail o rejec H : Range Range Fail o rejec H : AbsReurn AbsReurn Fail o rejec H : RealizedVol RealizedVol Fail o rejec H : Var Var Rejec H : Range Range Rejec H : AbsReurn AbsReurn Fail o rejec H : RealizedVol RealizedVol Rejec b. Posiion daa use PCofNonAll T.S. Pvalue Conclusion H : Var Var Fail o rejec H : Range Range Fail o rejec H : AbsReurn AbsReurn Fail o rejec H : RealizedVol RealizedVol Fail o rejec H : Var Var Rejec H : Range Range Rejec H : AbsReurn AbsReurn Fail o rejec H : RealizedVol RealizedVol Rejec 32

35 c. Posiion daa use NonCommercialNeLong T.S. Pvalue Conclusion H : Var Var Fail o rejec H : Range Range Fail o rejec H : AbsReurn AbsReurn Rejec H : RealizedVol RealizedVol Fail o rejec H : Var Var Fail o rejec H : Range Range Fail o rejec H : AbsReurn AbsReurn Fail o rejec H : RealizedVol RealizedVol Fail o rejec 33

36 5.3 Whea a. Posiion daa use NonCommercial all T.S. Pvalue Conclusion H : Var Var Fail o rejec H : Range Range H : AbsReurn AbsReurn H : RealizedVol RealizedVol H : Var Var H : Range Range H : AbsReurn AbsReurn H : RealizedVol RealizedVol Fail o rejec Fail o rejec Fail o rejec Rejec Rejec Rejec Rejec H : Var Var b. Posiion daa use PCofNonAll T.S. Pvalue Conclusion Fail o rejec H : Range Range H : AbsReurn AbsReurn H : RealizedVol RealizedVol H : Var Var H : Range Range H : AbsReurn AbsReurn H : RealizedVol RealizedVol Fail o rejec Fail o rejec Fail o rejec Rejec Rejec Rejec Rejec 34

37 c. Posiion daa use NonCommercialNeLong T.S. Pvalue Conclusion H : Var Var Fail o rejec H : Range Range H : AbsReurn AbsReurn H : RealizedVol RealizedVol H : Var Var H : Range Range H : AbsReurn AbsReurn H : RealizedVol RealizedVol Fail o rejec Rejec Fail o rejec Fail o rejec Fail o rejec Fail o rejec Fail o rejec Noe: 1% significance level is used for ess in ables 5.1, 5.2 and