Three essays on agriculture economics

Transcription

1 Graduae Theses and Disseraions Iowa Sae Universiy Capsones, Theses and Disseraions 016 Three essays on agriculure economics Chao Li Iowa Sae Universiy Follow his and addiional works a: hps://lib.dr.iasae.edu/ed Par of he Agriculural and Resource Economics Commons, and he Agriculural Economics Commons Recommended Ciaion Li, Chao, "Three essays on agriculure economics" (016). Graduae Theses and Disseraions hps://lib.dr.iasae.edu/ed/15959 This Disseraion is brough o you for free and open access by he Iowa Sae Universiy Capsones, Theses and Disseraions a Iowa Sae Universiy Digial Reposiory. I has been acceped for inclusion in Graduae Theses and Disseraions by an auhorized adminisraor of Iowa Sae Universiy Digial Reposiory. For more informaion, please conac digirep@iasae.edu.

2 Three essays on agriculure economics by Chao Li A disseraion submied o he graduae faculy in parial fulfillmen of he requiremens for he degree of DOCTOR OF PHILOSOPHY Major: Economics Program of Sudy Commiee: Dermo Hayes, Major Professor Sergio Lence Chad Har Keri Jacobs Cindy Yu Iowa Sae Universiy Ames, Iowa 016 Copyrigh Chao Li, 016. All righs reserved.

3 ii TABLE OF CONTENTS Page LIST OF FIGURES... v LIST OF TABLES... vi ACKNOWLEDGEMENTS... vii ABSTRACT... viii CHAPTER 1. PRICE DISCOVERY ON INTERNATIONAL SOYBEAN FUTURES MARKETS: A THRESHOLD CO-INTEGRATION APPROACH... 1 Absrac Inroducion Previous Theoreical Work Theoreical Issues Co-Inegraion and he Vecor Error Correcion Model Threshold Co-Inegraion Auoregressive Disribued-Lag Model Daa Descripion and Timelines Empirical Resuls Long-Run Lead-Lag Relaionship... 14

4 iii 1.4. Shor-Run Causal Effec Discussion and Conclusion Appendix CHAPTER. THE SUPPLY CURVE FOR CELLULOSIC ETHANOL Absrac Inroducion The Model Single Processor Model Collecion Mechanism as a Barrier o Enry From Theoreical Analysis o Pracical Simulaion Conclusion CHAPTER 3. THE EXAMINATION OF MARKET POWER OF U.S. NITROGEN FERTILIZER INDUSTRY: A BAYESIAN BASED APPROACH Absrac Inroducion Relaed Lieraure Mehodology Model Specificaion Esimaion Mehod Empirical Analysis Descripion of Daa... 74

5 iv 3.4. Model Esimaion and Empirical Resuls Single-Equaion Error Correcion Model Conclusion REFERENCES... 88

6 v LIST OF FIGURES Page Figure 1.1: Soybean fuures prices in he US, Brazil and China from 005 o Figure 1.: Timeline of he U.S., Brazilian, and Chinese fuures markes...13 Figure.1: Collecion cos increase as he disance beween wo processors change...57 Figure.: Marginal cos when he collecion radius is fixed...59 Figure 3.1: U.S. consumpion of nirogen, phosphae, and poash Figure 3.: U.S. plan nirogen use by corn, soybeans, coon, and whea, Figure 3.3: Monhly price of urea, naural gas, corn and % of capaciy uilizaion...64 Figure 3.4: Monhly price of ammonia, naural gas, corn and % of capaciy uilizaion...66 Figure 3.5: Time rend of effecs on ammonia price...79 Figure 3.6: Time rend of effecs on urea price...80 Figure 3.7: Impac of corn price on ammonia price...80 Figure 3.8: Impac of naural gas price on ammonia price...81 Figure 3.9: Impac of capaciy uilizaion on ammonia price...8 Figure 3.10: Impac of corn price on urea price...83 Figure 3.11: Impac of naural gas price on urea price...83 Figure 3.1: Impac of capaciy uilizaion on urea price...84

7 vi LIST OF TABLES Page Table 1.1: Bi-variae Johansen co-inegraion ess...15 Table 1.: Weak exogeneiy es for co-inegraed price pairs...17 Table 1.3: Esimaed resuls of linear error correcion model (ECM)...18 Table 1.4: Tes of linear co-inegraion agains hreshold co-inegraion...19 Table 1.5: Tes of no co-inegraion agains hreshold co-inegraion...1 Table 1.6: U.S. overnigh and Chinese dayime reurn...6 Table 1.7: Causaliy es of U.S. and Brazilian seasonal producion effec...7 Table 1.8: Uni roo ess for U.S. and Brazil conracs in each sub-period...30 Table 1.9: Uni roo ess for U.S. and Chinese No. 1 conracs in each sub-period...31 Table 1.10: Uni roo ess for US and Chinese No. conracs in each sub-period...31 Table 1.11: Forecased rading sraegy vs. real reurn...3 Table.1: Comparison beween differen collecion mechanisms...56 Table 3.1: Correlaion coefficiens beween urea, naural gas and corn...65 Table 3.: Correlaion coefficiens beween ammonia, naural gas and corn...66 Table 3.3: ADF and PP uni roo es...75 Table 3.4: Pair-wise Johansen co-inegraion es...76 Table 3.5: MCMC convergence es...78 Table 3.6: Resuls of error correcion model...86

8 vii ACKNOWLEDGEMENTS I hereby sincerely acknowledge my major professor, Dr. Dermo Hayes, for his wise guidance and unlimied supply of ime and paience hroughou he course of his research. His insighs and words of encouragemen always inspired me o complee my research. I would also like o hank my commiee members Dr. Sergio Lence, Dr. Keri Jacbos, Dr. Chad Har and Dr. Cindy Yu, for heir consrucive commens o help me complee his disseraion. Finally, I would like o give my special hanks o my family for heir undersanding and love.

9 viii ABSTRACT The main subjecive of his disseraion is o analyze hree issues of curren ineres in agriculural economics. Chaper 1 invesigaes he lead-lag relaionships among soybean prices in U.S., Brazilian, and Chinese fuures markes by using hreshold co-inegraion mehodologies. The empirical resuls indicae he influence of U.S. marke in he long-erm, and also show ha overnigh reurn of U.S. soybean fuures and he dayime reurn of Chinese No. 1 soybean fuures conemporaneously affec each oher in he shor-erm. A weak emporal seasonal causaliy beween U.S. and Brazilian soybean fuures prices exiss. Chaper examines he impac of feedsock supply mechanisms under condiions of spaial monopoly on he supply of cellulose o he plans. The model shows ha, in he absence of compeiion, he processor is indifferen beween processor collecion and supplier delivery, bu ha socieal welfare is higher under supplier delivery. By using a repeaed Nash Equilibrium, his paper shows ha processor collecion is firs bes for boh incumben and enran. By comparing he slope of marginal cos curve for his monopsonisic processor wih he slope of cos curve across oher feedsocks, subsanial quaniies of oher feedsocks may be required o mee he mandae. Chaper 3 invesigaes a change in he marke power of he U.S. nirogen ferilizer indusry by examining he causal linkage beween ferilizer, is main feedsock (naural gas), and oupu (corn) by using a Bayesian-based Kalman filer algorihm. The resuls of he imevarying esimaion show ha he U.S. nirogen ferilizer price follows he value of is marginal produciviy closer han is marginal cos of producion, indicaing a less compeiive marke srucure. The esimaion from he error correcion model suppors hese resuls.

10 1 CHAPTER 1. PRICE DISCOVERY ON INTERNATIONAL SOYBEAN FUTURES MARKETS: A THRESHOLD CO- INTEGRATION APPROACH Absrac This paper invesigaes he lead-lag relaionships among soybean prices in U.S., Brazilian, and Chinese fuures markes. We focus on boh long-run price co-movemens and on shor-run price relaionships. Various co-inegraion mehodologies and causaliy ess are applied o examine he changes in price relaionships over ime. The empirical resuls indicae he following: (a) he soybean fuures marke in he U.S. is sill he mos imporan and influenial marke, and he U.S. price, in he long-erm, leads price changes in Brazil and China; (b) in he shor-erm, he overnigh reurn of U.S. soybean fuures and he dayime reurn of Chinese No. 1 soybean fuures conemporaneously affec each oher, bu here is no significan causaliy beween U.S. overnigh reurn and he dayime reurn of Chinese No. soybean fuures; and, (c) a weak emporal seasonal causaliy beween U.S. and Brazilian soybean fuures price exiss and more ofen han no Brazilian fuures lead U.S. fuures during he Brazilian growing season. 1.1 Inroducion The U.S., Brazil, and Argenina accoun for over 90% of he world s soybean expors. China, which impored 71.4 million ons of soybeans in 014, is by far he larges imporer and ges approximaely 50% of is soybeans from he U.S. and 40% from Brazil.

11 The U.S., Brazil, and China all have acive soybean fuures. China has wo differen markes, one for non-gmo soybeans and he oher for impored GMO soybeans. In he Dalian Commodiy Exchange (DCE), he No. 1 conac is for non-gmo soybeans ha are used for human consumpion, and he No. conrac allows delivery of impored GMO soybean crops, which are used for soy oil and animal feed. Figure 1.1 shows soybean fuures prices in all four markes. There is visual evidence of srong co-movemens among hese prices, and we invesigae wheher his co-movemen is due o a sable long-run price relaionship and examine he price lead-lag relaionship across he four markes. 500 US BR CH1 CH Figure 1.1: Soybean fuures prices in he US, Brazil and China from 005 o 015 This paper is he firs o invesigae he long-run lead-lag relaionship among he U.S., Brazilian, and Chinese markes, and explores he seasonal relaionship beween U.S. and Brazilian fuures markes and he influence he Globex overnigh rading plaform in he U.S.

12 3 has on he dayime reurn of soybean fuures in China. 1 Overnigh rading in he U.S. and dayime rading in China occur conemporaneously, so we apply an auoregressive disribuedlag model o address his problem. The paper is organized as follows: Secion provides a brief lieraure review. Secion 1. describes he mehodology ha characerizes he price lead-lag relaionships using linear and non-linear co-inegraion. Secion 1.3 describes he daa. Secion 1.4 exhibis and explains empirical resuls of co-inegraion and demonsraes wo shor-run causaliy relaionships. Secion 1.5 presens conclusions Previous Theoreical Work Granger (1981) inroduced he mos widely used mehodology o sudy long-run price causaliy co-inegraion. He showed ha wo variables may have a long-run equilibrium relaionship even if hey are non-saionary. Engle and Granger (1987) exended his concep and showed ha co-inegraed variables can be represened by a vecor error correcion model (VECM) and provided es mehodology for his framework. Balke and Fomby (1997) inroduced he hreshold concep o explain possible non-linear long-run equilibrium relaionships. Hansen and Seo (00) and Seo (006) provided wo mehods o es a hreshold and a mehod o esimae he parameers of a hreshold vecor error correcion model (TVECM). Wahab and Lashgari (1993) and Ghosh (1993) invesigaed he forecasing power of he S&P 500 index spo and fuures prices changes using co-inegraion. Their resuls indicaed a 1 Globex is he elecronic rading sysem in he U.S.

13 4 sable long-run equilibrium relaionship beween he index and is fuures price. Chu e al. (1999) invesigaed he price discovery funcion in hree S&P 500 index markes: he spo index, index fuures and S&P Deposiary Receips markes. They found ha he hree price series are a co-inegraed sysem wih one long-run sochasic rend and he fuures marke serves he dominan price discovery funcion when he common sochasic rend is decomposed. Marens e al. (1998) applied a hreshold error correcion model o sudy indexfuures arbirage and found ha he impac of fuures marke on he spo marke is larger when he mispricing error is negaive and ha he impac of he mispricing error increases wih he magniude of ha error. Booh, Brockman, and Tse (1998) invesigaed he relaionship beween U.S. and Canadian whea fuures prices and showed ha boh of hem are inegraed of order one and ha hey are co-inegraed. Fung e al. (013) used daily daa for 16 commodiy fuures conracs raded in China and he corresponding foreign markes o analyze price linkages among markes. They also sudied he impac of Chinese fuures dayime reurns on he U.S. overnigh reurns using a regression of one reurn on he oher; however, heir sudy ignored he effec of he previous reurn on he curren reurn. For soybeans, heir resuls showed ha he price causaliy beween U.S. soybean fuures conracs and boh Chinese No. 1 and No. conracs are saisically significan bu no economically significan. Peri and Baldi (010) employed he hreshold co-inegraion approach o analyze he long-run relaionship beween vegeable oil prices and convenional diesel prices in he EU and suggesed a wo-regime hreshold co-inegraion relaionship for he rapeseed oil and diesel price pair. Naanelov e al. (011) examined price linkages beween crude oil fuures and a series of agriculural

14 5 commodiies fuures using he VECM mehod and showed ha co-movemen of commodiy prices is a emporal seasonal concep and should be reaed accordingly. In he lieraure regarding soybean fuures price discovery beween U.S., Brazilian, and Chinese markes, Han e al. (013) examined he role ha he DCE plays in he global discovery of soybean fuures. They used a srucural vecor auoregressive model (SVAR) and VECM on he reurns of he DCE and he CBOT soybean fuures during rading and nonrading hours. The resuls indicae a bi-direcional causaliy beween he wo markes wih he CBOT leading he DCE. Han e al. did no include he Chinese No. soybean fuures conrac due o a liquidiy problem. Our paper employs more recen daa o capure he role of Chinese No. conracs and focuses on hreshold co-inegraion analysis. Han e al. used a differen approach han he one described below when rolling he price daa when he nearby conrac expires. This may help explain he difference beween he resuls presened in his paper and hose in Han e al. (013). Chrisofolei e al. (01) examined he price linkage beween soybean fuures conracs in China, U.S., Brazil, and Argenina using VECM. The resuls indicaed ha he U.S. price has a dominan role. Liu e al. (015) used a generalized auoregressive condiional heeroskedasiciy (GARCH) model based on generalized error disribuion (GED) and exponenial GARCH-GED models and found ha he spillover from CBOT soybean fuures o DCE No. 1 soybean fuures has weakened hrough ime, indicaing a more influenial Chinese soybean marke. Merener (015) invesigaed how local supply shocks in he globally disribued producion of commodiies are incorporaed ino CME fuures prices and found ha CME soybean fuures prices have become increasingly sensiive o supply shocks ouside of he Unied Saes.

15 6 1. Theoreical Issues 1..1 Co-Inegraion and he Vecor Error Correcion Model In ime-series economerics, a price series ha has a saionary, inverible, ARMA represenaion afer differencing d imes, is said o be inegraed of order d, denoed by p Id ( ). If boh series x and y are ( ) Id processes, heir linear combinaion ε = x βy is also an Id ( ) process. However, if here exiss a vecor, [1, β ], such ha ε Id ( b), where b > 0, hen hese wo series are said o be co-inegraed and he vecor [1, β ] is called he co-inegraing vecor. In appendices A1-A3, wo ypes of uni roo ess indicae ha all fuures prices sudied in his paper can be characerized as I (1) processes. Therefore, we concenrae our sudy on he case when d = b= 1. As a resul, he co-inegraed sysem can be simply characerized as a VECM µ k k i= 1 k x = +Γ x + A x + v (1.1) where x is a n 1 vecor of I (1) processes, µ is a n 1 vecor of consan, Γ and Ak are n n coefficien marices, and v is a n 1 vecor of Gaussian whie noise processes. Johansen (1988; 1991) demonsraed ha he rank of marix Γ represened he number of co-inegraion relaionships in vecor x. Thus, Johansen s co-inegraion es esimaes marix Γ hrough an unresriced VAR and ess possible rejecion of he resricion implied by he reduced rank of Γ. There are wo es saisics, one using he race and he oher using he maximum eigenvalue, and inferences can be differen.

16 7 The null hypohesis for he race es is ha he number of co-inegraing vecors is less han or equal o r. The es saisic is given by λ race n i= r+ 1 i = T ln(1 λ ) where T is he sample size acually used for esimaion and λ i is he esimaed values of he ordered eigenvalues from he esimaed marix. For he maximum eigenvalue es, he es saisic is given by r 1 λmax = T ln(1 λ + ) which ess he null hypohesis ha he number of co-inegraing vecors is exacly r agains he alernaive of r + 1 co-inegraing vecors. When here is a co-inegraion relaionship beween ime series, Granger causaliy can be esed by a Wald es. Specifically, a linear VECM of order r + 1 can be compacly represened as x= AX ' ( β ) + v (1.) 1 wih X 1 ( β ) 1 w 1( β ) x 1 = x x r

17 8 where x is a n-dimensional I (1) ime series, which is co-inegraed wih one n 1 coinegraing vecor β, w( β) = β' x 1 is he error correcion erm (ECT), and ( µ α ) A' = A A A is a n ( nr + ) marix of coefficiens. 1 r Thus, he bi-variae co-inegraed ime series can be wrien as r x1, µ 1 α1 x1, i v1, = + w 1( β ) + Ai + x, µ α i= 1 x, i v (1.3), w ( β) = x βx deermines he ECT. The opimal lengh of lag r is deermined where 1 1, 1, 1 by Akaike Informaion Crierion (AIC) or Schwarz Informaion Crierion (SIC). Inuiively, parameer α i measures he long-run causaliy relaionship and parameer β characerizes he long-run equilibrium beween hese wo series. By esing he null hypohesis of α i = 0 agains he alernaive of αi 0, hree differen resuls may be obained: (a) α 1 = 0 and α = 0 ; (b) α1 0 and α 0 ; and (c) α 0 bu α = 0 i j. The firs case indicaes no coinegraion exiss, he second case indicaes bi-direcional long-run causaliy, and he las case indicaes a unidirecional long-run causaliy relaionship. 1.. Threshold Co-Inegraion The above radiional VECM assumes he adjusmen process o he long-run equilibrium is coninuous and linear. In realiy, he influence of ransacion coss, adjusmen coss, or oher marke fricions makes i likely ha movemen oward he long-run equilibrium may only occur when he deviaion from equilibrium exceeds a criical hreshold level. I is also possible ha he speed a which he sysem reurns o long-run equilibrium differs under regimes. Balke and

18 9 Fomby (1997) inroduced he concep of hreshold co-inegraion o analyze his ype of discree adjusmen process. As an exension of model (1.), we propose he following specificaion of a wo-regime hreshold co-inegraion model: ' AX 1 1( β) + v, if w 1( β) γ x = ' AX 1( β) + v, if w 1( β) > γ (1.4) where one regime is close o he equilibrium regardless of he sign of ECT and he oher regime is far from he equilibrium, and γ is he hreshold parameer. Equaion (4) can be rewrien as x = AX ( β) d ( βγ, ) + A X ( β) d ( βγ, ) + v (1.5) ' ' wih d1 ( βγ, ) = 1( w 1 γ), d( βγ, ) = 1( w 1 > γ). The coefficien marices A 1 and A deermine he dynamics in each regime. The advanage of his specificaion capures he idea ha adjusmen speed o long-run equilibrium would be lower when he deviaion is banded. Wih he absolue value of w 1 above or below he criical hreshold value, his TVECM model allows all coefficiens, excep he co-inegraing vecor, β, o swich beween wo regimes. The esimaed coefficiens of w 1 measure differen adjusmen speeds of price moving back owards he long-run equilibrium. The hypohesis es for hreshold co-inegraion involves four differen cases: (a) coinegraion wih no hreshold effec; (b) co-inegraion wih a hreshold effec; (c) no coinegraion wih a hreshold effec; (d) no co-inegraion and no hreshold effec. Thus, given he bi-variae Johansen co-inegraion es resuls, we apply wo kinds of hreshold ess, developed by Hansen and Seo (00) and Seo (006).

19 10 If wo series have shown a co-inegraion relaionship using Johansen s co-inegraion es, we furher deermine wheher or no his co-inegraion is linear using he maximum likelihood mehod developed by Hansen and Seo (00), which involves a join grid search over he co-inegraing vecor β in he region [ β, β ] and a hreshold parameer γ in he region [ γ, γ ]. In our empirical applicaions, we se he number of grid searches for boh L U hreshold parameer and co-inegraing vecor a 300. Since he hreshold effec is only valid when 0 < P( w 1 γ ) < 1, i is imposed by assuming ha π P( w γ) 1 π L U where π 0 > 0 is a rimming parameer and se equal o Hanson and Seo (00) es he null hypohesis of linear co-inegraion (no hreshold effec) agains he alernaive hypohesis of hreshold co-inegraion by developing wo SupLM ess for a given or esimaed β using a parameric boosrap mehod o calculae asympoic p-value. The firs es is used when he rue co-inegraing vecor β 0 is known o be a priori and he es saisic is denoed as 0 SupLM = sup LM ( β0, γ) γ [ γl, γu ] The second es is used when he rue co-inegraing vecor is unknown and he es saisic is denoed as Oherwise, he model reduces o a linear co-inegraion model.

20 11 0 SupLM = sup LM ( βγ, ) γ [ γl, γu ] where β is he null esimae of he co-inegraing vecor. If wo series fail o show a co-inegraion relaionship by Johansen s co-inegraion es, we alernaively apply a supreme es developed by Seo (006). This ess he null hypohesis of no co-inegraion agains he alernaive hypohesis of hreshold co-inegraion using a Band- TVECM, Φ( L) x = µ + α w 1( w γ) + α w 1( w > γ) + v where = 1,..., n, and Φ ( L) is a qh-order polynomial in he lag operaor defined as Φ ( ) = Φ Φ 1 q L I 1L ql. When hreshold parameer γ is fixed, he leas-squares esimaors for he coefficiens are he OLS esimaors. Thus, equaion (1.5) can be specified as x = µγ ( ) + α( γ) w ( β) d ( βγ, ) + α( γ) w ( β) d ( βγ, ) Φ ( γ) x + +Φ ( γ) x + v ( γ) 1 1 q q and he supreme Wald es saisic is defined as SupW = sup W n ( γ ) γ [ γl, γu ] where parameerγ. W n is he Wald saisic from esing he null of no co-inegraion wih a fixed hreshold 1..3 Auoregressive Disribued-Lag Model To sudy he emporal causal effec of soybean fuures prices among differen markes, we employ an auoregressive disribued lag model (ARDL). Specifically, he auoregressive disribued lag model of order p and q, ARDL( pq, ), defined as follows:

21 1 p y = c+ ay + bx + ε k k k k k= 1 k= 0 q where y and x are saionary variables, and ε is whie noise. More sricly, we assume ha ε is saionary and independen of 1 x, x, and y, y 1,, so ha his ARDL model can be esimaed consisenly using ordinary leas squares. The esimaed conemporaneous parameer coefficien b 0 is he impac muliplier ha characerizes he emporal price relaionships. 1.3 Daa Descripion and Timelines Our empirical analysis uses daily nominal prices of soybean fuures conracs raded in he Chicago Mercanile Exchange (CME), he Dalian Commodiy Exchange (DCE) and he Brazilian Mercanile and Fuures Exchange (BM&F). All daa is colleced from a Bloomberg erminal, and he dae range is from 03/01/005 o 06/30/015. A close-o-mauriy mehod is employed o rollover daa across conracs and all daa is proporionally modified o eliminae he price jump across conracs. Owing o differences in naional holidays, daa in all hree markes are no auomaically mached. We have eliminaed mismached daa and he whole sample size is reduced o 377 observaions. We sandardize he price quoaion uni and conver all prices ino he naural log of prices measured in U.S. cens per bushel. The Chinese fuures marke is open from 9:00 a.m. o 11:30 a.m. and from 1:30 p.m. o 3:00 p.m. The rading hours in Brazil are from 9:00 a.m. o 3:15 p.m. The rading floor in he U.S. operaes from 8:30 a.m. o 1:15 p.m., and he Globex overnigh rading runs from 7:00 p.m. o 7:45 a.m. Figure 1. illusraes he imeline of hese rading hours.

22 13 Figure 1.: Timeline of he U.S., Brazilian, and Chinese fuures markes China is 13 hours ahead of he U.S. during he U.S. dayligh saving period and 14 hours ahead during sandard ime. During dayligh saving ime, he Chinese marke and U.S. Globex open a he same ime, and he Chinese marke closes 6.75 hours earlier han he Globex. During sandard ime, he Chinese marke opens one hour laer han U.S. and closes 5.75 hours earlier han he Globex. In oher words, he Globex is always open when he Chinese marke is open. Therefore, we do no need o adjus he daa for his ime change. The difference in dayligh periods beween he U.S. and Brazil does no impac he close-o-close reurn in each rading day in eiher marke. Therefore, we do no need o accoun for he effec of dayligh saving ime in he empirical work. In he following empirical applicaions, we focus on he close-o-close reurns of soybean fuures in each marke, he open-o-close (dayime) reurns of soybeans fuures in he Chinese marke, and he overnigh reurn of soybean fuures in he U.S. marke. Deailed variable noaions are provided in he appendix.

23 Empirical Resuls Long-Run Lead-Lag Relaionship Here, we presen resuls peraining o he long-run causaliy relaionship beween soybean fuures prices in he hree examined markes. Given he saionary es resuls in appendices, all closing prices can be regarded as an I (1) process in every sub-period. Thus, we apply he Johansen co-inegraion es o invesigae wheher here is a long-run linear relaionship beween he closing prices in each marke. As shown in figure 1.1, he co-movemens among he price series are quie srong from he beginning of he daa period, bu he relaionship weakens afer several years, suggesing a srucural break in he relaionship. The radiional approach o es srucural change would be picking an arbirary sample breakpoin, ofen he midpoin of he sample, and using Chow s (1960) F-es. The resul using his approach is very sensiive o he prior choice of break daes, and Hansen (001) suggess ha he Quand-Likelihood Raio (QLR) es is superior for deecing srucural change wih unknown iming. In our analysis, we are ineresed in wheher he U.S. soybean fuures marke is sill a world price leader, and we concenrae on he bi-variae causaliy relaionship beween soybean fuures prices in he U.S. and elsewhere. Therefore, we employ his QLR srucural change es and divide every pair-wise daa sample ino wo sub-periods based on he es resuls. The resuls sugges ha in July 01 a srucural break beween soybean fuures prices in he U.S. and Brazil occurred. Also, in Ocober 008 a srucural break beween he U.S. and Chinese No. 1 conracs occurred, and in Augus 009 a srucural break beween U.S. and Chinese No. conracs occurred as well.

24 15 Table 1.1 repors race saisics ( λ race) and maximum eigenvalue saisics ( λ max ) from he bi-variae Johansen es for each sub-period of he sample, and shows ha a linear coinegraion relaionship beween U.S. and Chinese No. 1 conracs does no exis in any period. Our resul conradics wih Han e al. (013) finding co-inegraion beween U.S. and Chinese No. 1 conracs. A difference in daa sample and modificaion approach may help explain he difference beween he resuls in our paper and hose in Han e al. (013) we employ more recen daa o capure he price relaionship and we proporionally modify fuures prices o eliminae he influence of price jump across conracs when rolling price daa when he nearby conrac expires. The underlying commodiy for he Chinese No. 1 conrac is non-gmo soybeans desined o be used for food. The underlying commodiy for he U.S. and Chinese No. conracs poenially conain GMO soybeans. Inuiively, he price relaionship beween U.S. and Chinese No. soybean fuures should be much closer han ha beween U.S. and Chinese No. 1 fuures conracs. This is shown in panel C of able 1.1 wih a co-inegraing relaionship prior o 009 and for he enire period. The co-inegraion relaionship also exiss beween he U.S. and Brazil. Table 1.1: Bi-variae Johansen co-inegraion ess Panel A: U.S. and Brazil soybean fuures conrac 03/01/005-07/31/01 08/01/01-05/30/015 03/01/005-06/30/015 λ race λ max λ race λ max λ race λ max r = ** 19.41*** ***.9*** r

25 16 Table 1.1 coninued Analysis Co-inegraed No co-inegraed Co-inegraed Panel B: U.S. and Chinese No. 1 soybean fuures conrac 03/01/005-10/31/008 11/01/008-06/30/015 03/01/005-06/30/015 λ race λ max λ race λ max λ race λ max r = r Analysis No co-inegraed No co-inegraed No co-inegraed Panel C: U.S. and Chinese No. soybean fuures conrac 03/01/005-08/31/009 09/01/009-06/30/015 03/01/005-06/30/015 λ race λ max λ race λ max λ race λ max r = 0 0.3** 19.83*** *** 4.03*** r Analysis Co-inegraed No co-inegraed Co-inegraed The VAR specificaion is esimaed by applying up o 1 lags. The opimal lag lengh is deermined by means of Schwarz informaion crierion (SIC). *,**,*** denoes saisical significance a he 10%, 5%, and 1% levels, respecively. In order o find he direcion of long-run lead-lag relaionships, we apply a weak exogeneiy es o he co-inegraing pairs of prices. Table II shows he resuls of hese ess. Parameer α in able 1. characerizes he long-run causaliy relaionship in equaion (3). When soybean fuures price in Brazil is reaed as he dependen variable, α is significanly posiive. This suggess ha U.S. soybean fuures price leads he price in Brazil. For he prices of U.S. and Chinese No. fuures conracs, he es resuls indicae a bi-direcional causaliy

26 17 relaionship, implying he soybean fuures prices in U.S. and Chinese No. conracs are influenced by each oher. Table 1.: Weak exogeneiy es for co-inegraed price pairs Panel A: Lead-lag relaionship beween soybean fuures in U.S. and Brazil H 0: 1=0 03/01/005-07/31/01 03/01/005-06/30/015 α H 0: α =0 H 0: α 1=0 0 α H : = ** *** US Brazil US Brazil Panel B: Lead-lag relaionship beween U.S. and Chinese No. soybean fuures H 0: 1=0 03/01/005-08/31/009 03/01/005-06/30/015 α H 0: α =0 H 0: α 1=0 0 α H : =0 1.7*** 4.13** 9.79*** 8.53*** US ChinaNo US ChinaNo *,**,*** denoes saisical significance a he 10%, 5%, and 1% levels, respecively. Table 1.3 presens he parameer esimaes ha characerize he long-run equilibrium relaionship and he speed of adjusmens o he long-run equilibrium. ECT 1 is he error correcion erm of he VECM model, and is coefficien represens he adjusmen speed o long-run equilibrium. US is he close-o-close reurn of soybean fuures in he U.S. BR is he close-o-close reurn of soybean fuures in Brazilian marke. reurn of Chinese No. soybean fuures. CH is he close-o-close

27 18 Table 1.3: Esimaed resuls of linear error correcion model (ECM) Panel A: U.S. and Brazilian soybean fuures conracs 03/01/005-07/31/01 03/01/005-06/30/015 US BR US BR ECT * ** 1 US *** *** BR * * Consan * ** ECT = US BR ECT = US 1.03 BR Panel B: U.S. and Chinese No. soybean fuures conracs 03/01/005-08/31/009 03/01/005-06/30/015 US CH US CH ECT *** 0.01* *** ** 1 US *** *** 1 CH Consan * * ** ** ECT = US CH ECT = US CH *,**,*** denoes saisical significance a he 10%, 5%, and 1% levels, respecively. The implicaion of esimaion resuls in able 1.3 coincide wih he es resuls in able 1.. The significance of esimaed coefficiens for ECT implies ha U.S. soybean fuures prices lead he price in Brazil and ha he prices of U.S. and Chinese No. fuures influence each

28 19 oher. The significan posiive sign of US 1 in each esimaed equaion suggess an increasing price change in oher markes when he soybean fuures price increases in he U.S. marke. Finally, he adjusmen speed o deviaions from he long-run equilibrium is characerized by he magniude of significan coefficien of ECT, and he long-run equilibrium relaionship beween prices is characerized by a co-inegraing vecor in ECT expression. Turning o he analysis of a non-linear long-run causaliy relaionship, we es wheher or no he soybean fuures prices in differen markes are hreshold co-inegraed. For he pairs of linear co-inegraed prices, he presence of a hreshold is esed and esimaed via he applicaion of a SupLM es by Hansen and Seo (00). This ess he null hypohesis of linear (Johansen) co-inegraion agains he alernaive hypohesis of hreshold co-inegraion. Table 1.4 displays he es saisics and heir boosrapped p-values ou of four daa samples. Only he whole period pair of prices beween U.S. and Chinese No. soybean fuures suppors a hreshold co-inegraion a a boosrapped p-value of Table 1.4: Tes of linear co-inegraion agains hreshold co-inegraion Tes Saisic P-value US and Brazil (03/01/005-07/31/01) US and Brazil (03/01/005-06/30/015) US and CH No. (03/01/005-08/31/009) US and CH No. (03/01/005-06/30/015) Compared o he linear esimaion in able 1.3 above, he co-inegraing coefficien for his hreshold co-inegraed price pair decreases from 1.03 o This again shows a srong co-movemen beween Chinese No. soybean fuures price and he U.S. soybean fuures price.

29 0 The esimaed criical hreshold value is 0.6 cens per bushel, which divides he whole daa se ino wo regimes. There are 86.9% observaions ha fall ino he usual regime US 0.97 CH 0.6, while he remaining 13.1% of observaions belong o he unusual regime US 0.97 CH > 0.6. The esimaed TVECM model is fully represened as * * ECT US CH 1, ECT US = ** * * ECT US CH 1, ECT 1 > 0.6 *** ** *** ECT US CH 1, ECT CH = *** ECT US CH 1, ECT 1 > 0.6 where ECT = US 0.97 CH The adjusmen parameers of ECT in he U.S. equaion are and in he usual and unusual regime, respecively. This difference in he saisically significan magniude of ECT coefficien indicaes a faser adjusmen speed oward long-run equilibrium when he absolue value of price deviaion from equilibrium exceeds he criical hreshold. In boh he usual and he unusual regimes, he esimaed coefficiens of US 1 are significanly differen from zero in he Chinese No. equaion, while he esimaed coefficiens of CH 1 are no saisically significan in he U.S. equaion. This suggess ha here is a significan shor-run response of he Chinese No. soybean fuures price o he price change in he U.S. When he U.S. soybean fuures price changes by 1%, he Chinese No. soybean fuures price changes by % and % in he same direcion when deviaion from equilibrium belongs o he usual and unusual regime, respecively. These shor-run adjusmen parameers provide evidence ha prices in he U.S. ypically lead prices for he Chinese No. soybean fuures conrac.

30 1 Turning o he remaining five pairs of fuures prices, which do no exhibi linear coinegraion relaions, we apply a SupWard es by Seo (006) o es wheher or no hey are hreshold co-inegraed and o demonsrae parameer esimaes. Specifically, we implemen hreshold co-inegraion analysis for he pos-break period prices of U.S. and Brazilian soybean fuures, he pos-break period prices of U.S. and Chinese No. soybean fuures and each subperiod of U.S. and Chinese No. 1 soybean fuures. Table 1.5 shows he resuls of he es of no co-inegraion versus hreshold co-inegraion. Two pairs of prices rejec he no co-inegraion null hypohesis a a less-han 10% significan level. This provides evidence of hreshold co-inegraion beween hese prices. The boosrapped p-value is and for he pre-break period and he whole period of prices beween U.S. and Chinese No. 1 conracs, respecively. The es resuls for he oher hree pairs of soybean fuures prices are no significan a convenional levels. Table 1.5: Tes of no co-inegraion agains hreshold co-inegraion Tes Saisic P-value US and Brazil (08/01/01-05/30/015) US and CH No. 1 (03/01/005-10/31/008) US and CH No. 1 (11/01/008-06/30/015) US and CH No. 1 (03/01/005-06/30/015) US and CH No. (09/01/009-06/30/015) The co-inegraing coefficien is esimaed as β = 0.94 for he pre-break period daa of U.S. and Chinese No. 1 soybean fuures, showing a srong responsiveness of he Chinese No. 1 conrac price o he U.S. soybean fuures price. The esimaed hreshold poin is 0.09 cen

31 per bushel, which divides he observaions ino wo regimes. Of he observaions, 84.3% fall ino he usual regime US 0.94 CH1 0.09, and he remaining 15.7% of observaions belong o he unusual regime US 0.94 CH1 > The esimaed TVECM model is fully represened as * ECT US CH1 1, ECT US = ** ECT US CH1 1, ECT 1 > 0.09 *** ECT US CH1 1, ECT CH1 = ** ** *** *** ECT US CH1 1, ECT 1 > 0.09 where ECT = US 0.94 CH1 The long-run adjusmen parameers of ECT in Chinese No. 1 equaions are significanly differen from zero in he unusual regime, while he long-run adjusmen parameers of ECT in U.S. equaions are no saisically significan. This indicaes ha he U.S. soybean fuures price drives he Chinese No.1 soybean fuures price oward he equilibrium level. In paricular, he adjusmen parameers of ECT in he Chinese No. 1 equaion are and in he usual and unusual regime, respecively. This difference in he magniude of ECT coefficien demonsraes a faser adjusmen speed oward long-run equilibrium when he absolue value of deviaion from equilibrium exceeds he criical hreshold. In boh he usual and he unusual regimes, he esimaed coefficiens of US 1 are significanly differen from zero in he Chinese No. 1 equaion. This suggess ha here is a significan shor-run response of he Chinese No. 1 soybean fuures price o he price change in U.S. When he U.S. soybean fuures price changes by 1%, he Chinese No. 1 soybean fuures price changes by 0.650% and % in he same direcion when deviaion from equilibrium belongs o he usual and unusual regime, respecively. These shor-run adjusmen parameers provide evidence ha

32 3 prices in he U.S. ypically lead prices for he Chinese No. 1 soybean fuures conrac, his resul coincides wih he resuls from he long-run adjusmen parameers of he ECT. For he whole period daa of U.S. and Chinese No.1 soybean fuures price, he coinegraing coefficien is esimaed as β = 0.98, showing a srong responsiveness of he Chinese No. 1 conrac price o he U.S. soybean fuures price. The esimaed hreshold poin is 0.41 cens per bushel, which divides he observaions ino wo regimes. Of he observaions, 9.5% fall ino he usual regime US 0.98 CH1 0.41, and he remaining 7.5% observaions belong o he unusual regime US 0.98 CH1 > The esimaed TVECM model is fully represened as ECT US CH1 1, ECT US = *** ** ECT US CH1 1, ECT 1 > 0.41 *** ** *** ECT US CH1 1, ECT CH1 = ** ** ECT US CH1 1, ECT 1 > 0.41 where ECT = US 0.98 CH1 The long-run adjusmen parameers of he ECT in he Chinese No. 1 equaions are significanly differen from zero in boh he usual and unusual regimes, while he long-run adjusmen parameers of ECT in he U.S. equaions are no saisically significan. This indicaes ha he U.S. soybean fuures price drives he Chinese No. 1 soybean fuures price oward he equilibrium level. In paricular, he adjusmen parameers of he ECT in he Chinese No. 1 equaion are and in he usual and unusual regimes, respecively. This difference in he magniude of he ECT coefficiens suggess a faser adjusmen speed when he absolue value of he deviaion from equilibrium exceeds he criical hreshold. In he usual region, he esimaed coefficien of US 1 is significanly differen from zero in he

33 4 Chinese No. 1 equaion while he esimaed coefficien of CH1 1 is no saisically significan in he U.S. equaion. This suggess ha here is a significan shor-run response of he Chinese No. 1 soybean fuures price o price changes in he U.S. When he U.S. soybean fuures price changes by 1%, he Chinese No. 1 soybean fuures price changes by 0.189% in he same direcion. These shor-run adjusmen parameers provide evidence ha prices in he U.S. ypically lead prices for he Chinese No. 1 soybean fuures conrac, coinciding wih he resuls from he long-run adjusmen parameers of he ECT. In summary, he co-inegraion resuls demonsrae unidirecional long-run price causaliy from U.S. o Brazilian and U.S. o Chinese No. 1 soybean fuures markes, and a bi-direcional long-run causaliy relaionship beween U.S. and Chinese No. soybean fuures markes. However, he esimaion resuls ha indicae Chinese soybean fuures prices leading he price of soybean fuures in U.S. are only significan in he unusual sample regime, indicaing ha soybean fuures prices in Brazil or China are sill led by fuures prices in he U.S Shor-Run Causal Effec We are ineresed in wo differen shor-run causal effecs. One is he conemporaneous effec of U.S. Globex overnigh reurn on he dayime reurns of soybean fuures in China s marke. The oher is he seasonal harves effec on U.S. and Brazilian soybean fuures reurns. Since U.S. soybean fuures conracs can be raded hrough he Globex overnigh plaform, he Chinese soybean fuures dayime reurn may be affeced by his synchronous rading. Our analysis of he shor-run causal effec concenraes on how dayime reurns in China are affeced by he informaion conen of U.S. overnigh prices. For simpliciy, we se p = q = 1 and derive he unsrucured esimaion of he ARDL(1,1) model by ordinary leas

34 5 squares. 3 In paricular, he impac of U.S. overnigh reurns on Chinese dayime reurns is examined by he following regressions: CH = α + β * US + η * US + γ * CH + ε D N N D US = α + β * CH + η * CH + γ * US + ε N D D N 1 1 Table 1.6 repors he esimaion resuls for differen sub-period samples. I shows ha U.S. soybean fuures overnigh reurns and Chinese No. 1 soybean fuures dayime reurns significanly affec each oher in five ou of eleven years, while esimaion resuls of U.S. soybean fuures overnigh reurns and Chinese No. soybean fuures dayime reurns are no saisically significan a convenional levels. These resuls indicae ha informaion abou U.S. Globex overnigh rading influences he price change of Chinese No. 1 soybean fuures conracs raher han No. soybean fuures conracs. This is mainly because Chinese No. 1 conracs are more acive in he marke. Thus, he shor-run marke price of No. 1 conracs would be more sensiive o he price in oher markes due o no arbirage heory. Therefore, he soybean fuures price in Chinese No. 1 conracs are no only hreshold co-inegraed wih hose in he U.S. in he long-run, is shor-run price change is also influenced by he informaion from overnigh price changes in he U.S. marke as well. Excep for 015, all significan coefficiens of βi s are posiive, indicaing ha he price increase in U.S. overnigh rading will simulae he rading of Chinese No. 1 soybean fuures conracs and end o increase dayime reurn. 3 Imposing no srucure on he relaionship of he coefficiens of he lagged explanaors may cause mulicollineariy, leading o high variance of he coefficien esimaes.

35 6 Table 1.6: U.S. overnigh and Chinese dayime reurn Column A: U.S. and Chinese Column B: U.S. and Chinese No. 1 Conrac No. Conrac Year Dependen Variable β i β i D CH 0.117*** N US *** D CH * N US * D CH N US D CH * N US * D CH N US D CH N US D CH * N US * D CH N US

36 7 Table 1.6 coninued D CH N US D CH N US D CH * N US * *,**,*** denoes saisical significance a he 10%, 5%, and 1% levels, respecively. The oher emporal causal effec we are ineresed in is he seasonal harves effec on U.S. and Brazilian soybean fuures. From he long-run analysis above, we conclude ha soybean fuures prices in he U.S. marke lead hose prices in he Brazilian marke. In he shor-run, however, his may no always be he case. The U.S. peak harves period exends from May o Ocober, while peak harves period in Brazil exends from November o April. As a resul, i is likely ha he Brazilian soybean fuures price leads he price in he U.S. in is harves period when a srong seasonal effec exiss. Table 1.7 illusraes he resuls of a causaliy es in each year, where expeced seasonal causaliy exiss in five ou of en years, suggesing a weak seasonal causal effec beween U.S. and Brazilian soybean fuures prices over ime. Table 1.7: Causaliy es of U.S. and Brazilian seasonal producion effec Nov-April Period US vs. Brazil Causaliy Tes H 0: 1=0 H : α =0 α 0

37 8 Table 1.7 coninued Year *** 0.07 Analysis Brazil US Year Analysis - Year ** Analysis US Brazil Year * 8.58** Analysis Brazil US Year Analysis - Year * 3.51* Analysis Brazil US Year ** Analysis US Brazil Year ** 0.9 Analysis Brazil US Year Analysis - Year ** 0.3 Analysis Brazil US - indicaes insignifican causaliy. *,**,*** denoes saisical significance a he 10%, 5%, and 1% levels, respecively.

38 9 1.5 Discussion and Conclusion This paper offers a comprehensive sudy on price causaliy beween soybean fuures prices in differen markes from 005 o 015. Boh long-run and shor-run price relaions are examined by various ime-series mehods. The long-run empirical resuls in secion 4 indicae ha he U.S. soybean fuures marke is he mos influenial marke, and soybean fuures prices in oher markes like Brazil and China are led by he price change in he U.S. However, some sub-period co-inegraion ess show ha here is no direcional causaliy beween U.S. and Chinese No. soybean fuures prices, and ha here is a unidirecional causaliy beween U.S. and Chinese No. 1 soybean fuures prices. Inuiively, he rapid growh of boh Chinese soybean spo and fuures markes make hem more influenial o fuures prices in he world. Thus, he lead-lag relaionship beween U.S. and Chinese soybean fuures has been changed in recen years. If we form a liquid rading sraegy based on whole period lead-lag relaionship beween U.S. and Chinese soybean fuures prices, 4 we find in appendix A4 ha he reurn for rading sraegy beas he real reurn in only four ou of eleven years, showing ha he direcional causaliy beween U.S. and Chinese soybean fuures markes is no always robus. This paper also invesigaes wo ypes of emporal price causaliies. One is he effec of overnigh price changes of U.S. soybean fuures hrough he Globex elecronic rading sysem on he dayime reurn of Chinese soybean fuures, he oher is he seasonal harves effec beween U.S. and Brazilian soybean fuures prices. The resuls indicae ha he Globex 4 The sraegy is o purchase and sale of he Chinese soybean fuures conrac a ime when he U.S. soybean fuures reurn was posiive a ime -1. Oherwise, keep he asse and earn risk-free ineres reurn.

39 30 overnigh price change in soybean fuures affecs, o some exen, he dayime price of Chinese No. 1 soybean fuures, bu here is no significan evidence indicaing ha overnigh rading affecs Chinese No. soybean fuures. Moreover, half of empirical ess abou seasonal harves effec mach wih our inuiive expecaion, indicaing a week seasonal causaliy beween U.S. and Brazilian soybean fuures prices according o heir harves periods. In general, we can conclude from his paper ha he U.S. sill plays an imporan role in he worldwide soybean marke, and he price changes in U.S. soybean fuures will affec he fuures price in oher markes like China and Brazil. However, wih he developmen of soybean markes in Brazil and China, his long-run unidirecional price causaliy from he U.S. o Brazil or from he U.S. o China has been weakened, and he opposie direcion of price causaliy has begun o emerge. 1.6 Appendix Table 1.8: Uni roo ess for U.S. and Brazil conracs in each sub-period 03/01/005-07/31/01 08/01/01-06/30/015 03/01/005-06/30/015 ADF Tes PP Tes ADF Tes PP Tes ADF Tes PP Tes US US *** *** *** *** *** *** BR BR *** *** *** *** *** *** *,**,*** denoes saisical significance a he 10%, 5%, and 1% levels, respecively.

40 31 Table 1.9: Uni roo ess for U.S. and Chinese No. 1 conracs in each sub-period 03/01/005-10/31/008 11/01/008-06/30/015 03/01/005-06/30/015 ADF Tes PP Tes ADF Tes PP Tes ADF Tes PP Tes US US *** *** *** *** *** *** CH CH *** *** *** *** *** *** N US *** *** *** *** *** *** CH1 D *** *** *** *** *** *** *,**,*** denoes saisical significance a he 10%, 5%, and 1% levels, respecively. Table 1.10: Uni roo ess for US and Chinese No. conracs in each sub-period 03/01/005-08/31/009 09/01/009-06/30/015 03/01/005-06/30/015 ADF Tes PP Tes ADF Tes PP Tes ADF Tes PP Tes US US *** *** *** *** *** *** CH CH *** *** *** *** *** *** N US *** *** *** *** *** *** CH D *** -3.8 *** *** *** *** *** *,**,*** denoes saisical significance a he 10%, 5%, and 1% levels, respecively.

41 3 Table 1.11: Forecased rading sraegy vs. real reurn Year % of correc forecas direcion Sraegy reurn Real reurn % 94.47% % % 90.58% % % 158.% % % 10.14% 83.3% % 13.64% % % 164.6% % % 95.80% % % 94.5% % % 66.91% 96.07% % 48.35% 76.74% % % 9.09%

42 33 CHAPTER. THE SUPPLY CURVE FOR CELLULOSIC ETHANOL Absrac This paper examines he impac of feedsock supply mechanisms under condiions of spaial monopoly on he supply of cellulose o he plans and by implicaion on he supply of cellulosic ehanol. we show he minimizaion problem for cellulosic processors under hree differen collecion mechanisms and provide opimal pricing rule and he opimal collecion radius needed o mee feedsock supply requiremens. These show ha in he absence of compeiion he processor is indifferen beween processor collecion and supplier delivery, bu ha socieal welfare is higher under supplier delivery. We hen use a repeaed Nash Equilibrium game o show ha processor collecion is firs bes for boh incumben and processor and is an effecive deerren agains an enran locaing a plan wihin he draw area of he incumben. We suppor he heoreical resuls wih a numerical simulaion showing he opimal premium and draw area under each mechanism. Third, we use he resul of he simulaion show he rae a which sover collecion coss increase for a monopsonisic sover processor consrained o he original draw area. The slope of he marginal cos curve for his monopsonisic processor is hen compared wih he slope of he cos curve across oher feedsocks. These resuls sugges ha subsanial quaniies of hese oher feedsocks may be required o mee he mandae.