Predicing he Corn Basis in he Texas Triangle Area Vardan Mkrchyan 1, J. Mark Welch 1,2 and Gabriel J. Power 1 Texas A&M Universiy 1 Deparmen of Agriculural Economics and 2 Texas Agri-Life Exension Seleced Paper prepared for presenaion a he Souhern Agriculural Economics Associaion Annual Meeing, Alana, Georgia, January 31-February 3, 2009 Copyrigh 2009 by Vardan Mkrchyan, J. Mark Welch and Gabriel J. Power. 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.
Absrac The basis is a vial concep in he producion, markeing and hedging of many commodiies. Concern over basis levels has inensified in corn markes recenly because of some significan changes in he corn marke place. Corn producers and users would sand o benefi from a new, flexible, and a beer performing mehod o predic he basis. Being able o predic he basis more accuraely makes i easier o marke corn efficienly and o maximize profi. This sudy develops a new and sraighforward economic model of basis forecasing ha ouperforms he simple hree-year average mehod suggesed in much of he lieraure. We use monhly daa of he corn basis in he Texas Triangle Area from February 1997 o July 2008. The resuls and he graphs indicae ha he new model based on economic fundamenals performs beer han basis esimaes using a hree-year moving average. 1
Inroducion A cenral issue for farmers in commodiy markeing is forecasing he basis, which is defined as he difference beween he cash price and he fuures price for a commodiy in a specific delivery locaion and of specific qualiy grade (Tomek, 1997). In he U.S., corn has long been he crop wih he highes oal dollar value. The imporance of corn increased wih he Energy Independence and Securiy Ac of 2007, which mandaes he producion of a leas 36 billion gallons of bio-fuel by he year 2022. I is esimaed ha 15 billion gallons of he 36 billion gallon mandae will come from corn based ehanol. The U.S. currenly has 128 ehanol plans and an addiional 85 under consrucion. Producion is concenraed in he grain surplus Midwesern saes while Souheas and Souhwes saes, including Texas, are grain defici saes (see Figure1). The basis is affeced by wheher a sae is in a corn surplus or defici region. Emerging ehanol producion in he Midwes is expeced o srenghen he basis in ha region meaning imporers like he Texas High Plains will need o bid more o ge he corn supplies hey need. The basis in Texas will be affeced as well. [Figure 1 approximaely here] The focus of his paper is o forecas he corn basis in he Texas Triangle Area, a saisical reporing region defined by he Naional Agriculural Saisics Service and locaed in he Texas High Plains (see Figure 2). I includes elevaors in an area from Plainview o Canyon o Farwell and is comprised of Casro, Deaf Smih, Parmer, Randall, and Swisher counies in he Texas panhandle. The Triangle Area is a leader of Texas corn producion and is a he hear of he Texas cale feeding indusry (TASS, 2008). In addiion, Whie Energy Inc. of Dallas, Texas began operaion of a 100 million gallon per year (mgy) corn ehanol plan in Deaf Smih Couny on January 15, 2008. Whie Energy also operaes a 2
100 mgy corn ehanol plan in adjacen Hale Couny, Texas. An addiional 100 mgy ehanol plan is currenly under consrucion in Deaf Smih Couny. [Figure 2 approximaely here] I is likely ha he paern of corn basis is undergoing changes given he effecs of ehanol policies, increased ransporaion coss, and volailiy in he grain markes more generally. The purpose of his paper is o compare forecass of he basis, given hese dynamic condiions, based on esimaed models of he deerminans of he basis. Two approaches are compared using boh in-sample and ou-of-sample daa: a purely saisical hree-year moving average of he basis, and a model ha uses as explanaory variables publicly available daa on economic fundamenals ha are well suppored by economic heory. By doing so, he paper makes mehodological and policy conribuions o undersanding he relaionship beween grain fuures markes in Chicago and local cash markes. Lieraure Review Even hough predicing he basis and having accurae esimaes for local markes is essenial, Jiang and Hayenga (1997) noe ha here have been few basis behavior sudies and even fewer basis forecasing sudies (no couning simple moving average esimaions of hisorical basis daa). The model used by Jiang and Hayenga includes sorage cos, ransporaion cos, and regional supply and demand variables o explain basis behavior. They use a number of forecasing echniques for he corn and soybean basis, including a simple hree-year-moving average forecas, a srucural economeric model, a modified 3
hree-year average model, arificial neural neworks, seasonal ARIMA ime series models, sae-space models, and composie forecass. They repor in heir conclusion ha expor levels have lile o no effec on he local basis. They conclude ha hree-year-average-plus and seasonal ARIMA models are he mos pracical, are much easier o use han oher alernaive models, and slighly ouperform he simple hree-year-average forecas. Sanders and Manfredo (2006) also find, in he case of he soybean complex, ha he gains from using sophisicaed ime series models raher han a simple moving average o forecas he basis are relaively small. In heir sudy Taylor, Dhuyveer, and Kasens (2006) compare pracical mehods of forecasing he basis. They look a curren marke informaion of whea, soybeans, corn and milo (grain sorghum) in Kansas. They use nine differen models o forecas he basis and conclude ha, despie no having any rule o define he bes forecasing mehod, using he one-year average basis o forecas he fuures basis has worked beer han long-erm averages wih some producs. They also sae ha o forecas he whea basis a harves, he five-year average is he bes forecas model. Parcell, Schroeder and Dhuyveer (2000) look a he live cale basis in hree differen saes and use a mulivariae model o predic he basis. The auhors sae ha he explanaory variables explain 85% of he variaion in each sae. They also sae ha corn prices have a significan effec on live cale basis bu he magniude is lower han was suggesed in anoher sudy. Tomek (1997) noes ha here has been considerable research done on modeling basis behavior bu he number of forecasing analyses is small. Tomek adds ha i is ofen very difficul o obain he daa for all he variables influencing basis behavior, herefore 4
forecass of he basis have been made from simple ime series or naïve models. In his analysis, Tomek looks ino wo ypes of basis models. The firs is relaed o invenories carried over from one crop year o he nex. This model uses he cash prices peraining o a period near he end of he curren crop year and fuures quoes for he firs conrac in he new crop year. Tomek saes ha his basis measures how large is he incenive for carrying socks from one year o he nex. The second model is relaed o invenories wihin he same year. This model is relaed o basis changes wihin a year, ha is, changes over a sorage inerval. Tomek concludes ha exising price forecasing models are generally poor predicors of fuures prices bu migh be valuable o individual enerprises as hey develop or obain informaion no available o ohers. He also noes ha he effec of small or dwindling invenories on prices is much larger han he effec of large or pleniful invenories. This finding suggess ha invenories should be included among he explanaory variables for he basis. Garcia and Good (1983) examine he facors influencing he corn basis in Illinois. They argue ha he supply and demand of sorage should be included as explanaory variables for he basis in addiion o he cos of sorage and ransporaion. They wrie ha small socks (invenories) or a srong demand for shipmens (expors) could srenghen he basis. They conclude ha he hree ses of variables ha influence he basis are cos, sock, and flow facors. Garcia and Good use cross-secion daa and ime series daa for heir model. They hypohesize ha high levels of corn and soybean socks creae a high demand for sorage which in iself creaes high price for sorage everyhing else held consan. They also expec ha high levels of corn socks and a high cos for sorage creae 5
a wider basis. Garcia and Good include barge raes, regional dummy variables, monhly dummy variables, and ineres rae o reflec he relaionship beween cos and he basis. They conclude ha he basis paerns are fairly sysemaic. They find ha sorage has a srong posiive impac on Illinois basis during harves ime and slighly diminishes in oher periods. The cos of ransporaion is imporan during he off-harves season bu no during he harves season. Hranaiova and Tomek (2001) discuss he imporance of he iming opion on he basis behavior. They look a he basis as a funcion of ineres rae, convenience yield, sorage cos, ime o mauriy and iming opion. Their OLS regression esimaes show ha a day one of he mauriy monh, he iming opion is saisically imporan and wih convenience yield included, represens abou 92% of he basis. Tomek and Peerson (2001) emphasize he imporance for hedging of undersanding he basis. They discuss differen markeing sraegies for farmers o maximize profis and argue ha geing a good forecas of he basis is a difficul bu imporan ask. Mos previous sudies conclude ha an averaging mehod o forecas he basis is he mos pracical. This paper compares an alernaive mehod based on a few relevan variables from readily available daa sources o he radiional moving average approach. If he new model is seen as providing beer esimaes of he cash o fuures price relaionship, i will be useful o producers and users of corn in he Texas panhandle in formulaing price expecaions. I may also provide a foundaion for corn producers in oher areas who seek a beer way of forecasing he basis in heir region. 6
Mehodology Based on economic heory, he previous lieraure, and he goal of keeping he model succinc, we choose seven variables ha we anicipae o be significan in predicing he Texas corn basis. These variables and heir prediced signs are: 1. Local cash price (+); 2. Fuures price, December mauriy (-); 3. Esimaed markeing year ending socks (-); 4. Transporaion coss (+); 5. The basis in a previous ime period (+); 6. Texas Off-Farm Invenories (-); and a 7. Harves Dummy (-). The choice of average cash and average fuures prices is based on he definiion of he basis (basis = cash price minus fuures price). The relevan fuures conrac for corn markeing in his region is he December conrac on he Chicago Board of Trade. The ending socks variable is included following he Kaldor-Working heory of sorage because corn is a sorable commodiy and esimaed levels of ending socks are imporan measures of supply and demand fundamenals. A ransporaion cos variable is included since Texas is a corn defici sae and corn is impored ino he sae from corn-abundan saes. This is inended o capure he effec of oil price increases from 2005 o 2008. A lagged basis variable is added o sabilize he daa and o accoun for serial correlaion. A Texas Off- Farm invenories variable is added o capure he affec of local invenories on local basis. A harves-ime dummy variable is added o capure he influence of harves on he Triangle 7
Area basis. All regressions are run in SAS and predicions are calculaed in Excel. The model ha we propose is given by: Basis = β0 + β1basis 1 + β Transporaion 5 + β Avg.Cash + β TexasOffFarm 6 2 + β 3 Avg.Dec.Fuures + β HarvesDummy + ε 7 + β 4 EndingSocks + for = 1,, 138 where: Basis 1 is he lagged basis one period (monhly); Avg.Cash is he average cash prices in ime in he Texas riangle region. ; Avg.Dec.Fu ures is he average December Fuures Price of corn a ime a he Chicago Board of Trade; EndingSoc ks is he projeced ending sock of corn repored by USDA; Transpora ion is he ransporaion index wih a base year of 1985; TexasOff Farm is he invenory daa for he Texas Off-farm corn repored quarerly; HarvesDummy is a dummy variable for monh of Ocober. The baseline model chosen is he hree-year moving average suggesed by he lieraure o be he simples and mos pracical way of calculaing he basis: Basis = β + β MA3 + ε 0 1 where Daa for = 1,,103 MA3 is he hree-year moving average of he basis. The daa for he basis model are readily available. The average cash corn price daa in he Triangle Region is from he Texas AgriLife Exension websie a Texas A&M Universiy s 8
Deparmen of Agriculural Economics. Fuures prices are from he Commodiy Research Bureau Daa Xrac. The average monhly price is a simple average of daily closing prices in he neares December conrac. Corn ending socks are from he USDA Naional Agriculural Saisics Service (NASS). Monhly updaes of projeced ending socks are colleced from he USDA s World Agriculural Supply and Demand Esimaes. Transporaion daa is a monhly producer price index for railroad ransporaion coss. I is obained from he Bureau of Labor Saisics in US Deparmen of Labor. Texas Off-Farm invenory levels are from he USDA websie. The ime period for all he daa is from February 1997 o July 2008. Table 1 conains he descripive saisics for he variables chosen for his sudy. [Table 1 approximaely here] Tesable Hypoheses In our model he join null hypohesis is ha: (i) he following se of economic fundamenal variables is significan in explaining he basis, and ha (ii) he variable coefficiens have he signs prediced by economic heory. I is expeced ha he basis will be: Increasing in average cash price in he Triangle Area from he ideniy Basis=Cash-Fuures; Decreasing in he average December fuures price, also from he ideniy; Decreasing in he monhly updae of projeced ending U.S. socks (invenories), since higher ending invenories are associaed wih igh sorage condiions ha may force cash sales hus weakening he basis; 9
Increasing in ransporaion cos because higher fuel coss imply i is more expensive o bring corn ou of grain surplus regions (i.e. near he par delivery for Chicago Board of Trade fuures) o grain defici regions such as he Triangle Area; Increasing in lagged basis, because he basis is (weakly) serially correlaed; and Decreasing in he Texas off-farm invenories, because higher regional invenories should depress local cash prices and weaken he basis. Dummy variable is included for seasonaliy (harves). Precisely, he seasonaliy dummy variable akes he value 1 if i is Ocober and akes he value 0 oherwise. Diagnosic ess are performed on he daa o evaluae he presence of heeroskedasiciy and serial correlaion, wih he necessary adjusmens being made in he posiive case. Resuls and Inerpreaion This secion presens he resuls obained from running correced Ordinary Leas Squares (OLS) regressions on he wo principal specificaions as well as specificaions ha exclude one or more insignifican independen variables. Economic Fundamenals Model The resuls for our proposed economic fundamenals model are summarized in Table 2. All of he resuls are repored a he 95% confidence level. The coefficien for he Lagged Basis variable is 0.4752 and is significan. The implicaion is ha, all else held consan, if he basis in he previous monh is one cen/bushel higher, hen he basis in he curren monh increases by abou half a cen. This finding confirms he expecaion ha he basis is weakly serially correlaed. In oher words, if he basis for previous monh is geing 10
sronger (more posiive) he basis for he nex monh will keep srenghening everyhing else held consan. [Table 2 approximaely here] The average cash price variable is also significan wih a coefficien of 0.1033. If he local cash price in he Triangle Area region goes up by one cen per bushel, he basis will increase by one enh of one cen, all else held consan. This resul is consisen wih he basis formula expressed as cash minus fuures. The average December fuures price variable has a negaive and significan coefficien of -0.1446. Again, he sign for his variable is consisen wih he basis definiion as cash price minus fuures price. If December fuures prices go up by one cen hen he basis in he Texas Triangle region will weaken by 0.1446 cens per bushel, all else held consan. The Projeced Ending Socks variable is saisically significan and negaive as expeced bu he coefficien is very small. The coefficien associaed wih one million bushels of ending socks is -0.00002964, implying ha i akes a change of one billion bushels in ending socks o change he basis by 3 cens, ceeris paribus. Curren esimaed U.S. ending socks for 2008-2009 are 1.154 billion bushels. I would ake a change in projeced ending socks of abou hiry percen o change he basis one cen. This resul is consisen wih he heory because higher projec year-end invenories sugges declining demand or increasing supplies and lower cash prices. The ransporaion index variable has a posiive and significan esimaed coefficien of 0.00203. This resul is consisen wih he fac ha Texas is a corn defici sae and corn is being impored o Texas from oher corn abundan saes. If he ransporaion 11
index goes up by one percenage poin, he basis srenghens by 0.2 cens per bushel, all else consan. As i coss more o bring corn from oher saes o Texas, buyers can afford o pay more o local producers raher han ranspor i from ou of sae, srenghening he basis. Some variables are no saisically significan and are excluded from he final regression specificaion. These are he Texas Off-Farm Invenory levels variable and he harves dummy variable. Exclusion of hese wo variables does no subsanially affec he RMSE, alhough boh R 2 and goodness-of-fi decrease. The parameer associaed wih he Texas Off-Farm invenories variable is negaive bu no significan. The sign indicaes ha he basis weakens as local grain invenories increase. Increasing invenories could be a sign of weakening demand which could weaken he basis. Increasing invenories migh also reflec large grain producion in he area or difficuly arranging ransporaion o move grain ou of invenory. Elevaors wih full bins would no offer price incenives o encourage producers o bring in more grain. They are more likely raher o weaken basis bids o discourage shor erm grain deliveries. The Texas Off-Farm invenory variable may no be significan because he daa are measured quarerly which is a lower frequency han he monhly daa colleced for he oher variables or because local sorage capaciy relaive o oal local demand is small. The harves dummy variable has a coefficien of -0.00746 and is no significan. I is dropped from he final regression model. The negaive sign of he parameer is consisen wih he heoreical predicion ha a harves, he local increase in corn supply depresses he cash price and weakens he basis. 12
Moving-Average Model Our comparison model is a hree-year moving average of he basis. The resuls for his model are presened in Table 3. The coefficien for he hree-year moving average is 0.4. This model has less explanaory power han does he economic fundamenals model. The R 2 is lower (0.1062) and he Roo Mean Squared Error is higher (0.082). The resuls show ha he economic fundamenals model has greaer explanaory power for he Triangle Area corn basis. The R 2 of 0.6738 is much greaer han he moving-average model R 2 of 0.1062, and he economic model RMSE of 0.0524 is much smaller han he moving-average model RMSE of 0.082. [Table 3 approximaely here] The improved accuracy of he economic fundamenals model also provides economically significan gains. Consider he problem of a grain merchan who owns an invenory of 100,000 bushels of corn sored for fuure sale and who esimaes he basis o implemen his markeing sraegy. If he chooses he economic fundamenals model insead of he baseline model he will save $0.02963/bu or $2963 for he sale of his invenory. Even hough our model is more complex han a sraighforward hree-year moving average model, he resuls clearly sugges ha he added difficuly is worhwhile. The superioriy of he model is illusraed by Figures 3 and 4. [Figures 3 and 4 approximaely here] This resul is obained by muliplying he quaniy of corn (100,000 bu) by he difference beween he RMSE for he economic model and he RMSE for he moving-average model, ha is, (0.082-0.0524)x100.000=$2963 13
Conclusion Undersanding he behavior of he basis is essenial in grain markeing. I is he means by which he price discovery funcion of he fuures exchange is expressed o producers and users of commodiies in specific locaions. Recen changes in he fundamenals of corn demand due o ehanol producion may have alered he cash-fuures relaionship in many areas. Specifically, he consrucion of an ehanol plan in he Texas panhandle may change hese marke dynamics. This paper shows ha a radiional hree-year moving average model of he basis does no rack changes in he basis as well as a relaively simple economic model is able o do. We creaed a model ha uses a few significan variables from easily obained daa ses o explain he basis in he Triangle Area beer han a hree-year moving average. Addiional research is needed o improve he basis predicions o make hem more responsive o changes in marke fundamenals and he oher facors ha drive he basis levels. I is a challenge o balance poenial gains from using more sophisicaed mehods agains he cos of collecing exra daa and esimaing more complicaed models. Alhough his paper considers a wide range of economically meaningful variables, here remain some explanaory variables ha could be furher sudied o evaluae heir conribuion o he basis forecasing. One example of a poenially useful explanaory variable is he level of expor aciviy from he pors of Texas. Our economic fundamenals model includes limied daa on he impac of new corn ehanol producion capaciy in he Texas panhandle. New esimaes of he basis afer plans under consrucion have come on line and been in operaion longer will provide insigh ino wheher here has been a fundamenal shif in he basis due o ehanol manufacure in 14
he area. All of hese effors are designed o give regional farmers and corn users more accurae predicions and guidance for fuure markeing decisions. References Chicago Board of Trade. 2000. Undersanding Basis. Chicago, IL: Chicago Board of Trade. Available a < www.cbo.com/cbo/docs/46577.pdf>. Dhuyveer C. K. and T. L. Kasens. 1998. Forecasing Crop Basis: Pracical Alernaives. In: Proceedings of he NCR-134 Conference on Applied Commodiy Price Analysis, Forecasing, and Marke Risk Managemen. Available a <hp://www.farmdoc.uiuc.edu/nccc134>. Garcia, P. and L. D. Good. 1983. An Analysis of he Facors Influencing he Illinois Corn Basis, 1971-1981. In: Proceedings of he NCR-134 Conference on Applied Commodiy Price Analysis, Forecasing, and Marke Risk Managemen. Available a <hp://www.farmdoc.uiuc.edu/nccc134>. Hranaiova, J. and W.G. Tomek. 2001. Role of Delivery Opions in Basis Convergence. Journal of Fuures Markes 22(8): 783-809. Jiang, B., and M. Hayenga. 1997. Corn and Soybean Basis Behavior and Forecasing: Fundamenal and Alernaive Approaches. In: Proceedings of he NCR-134 Conference of Applied Commodiy Price Analysis, Forecasing, and Marke Risk Managemen. Available a <hp://www.farmdoc.uiuc.edu/nccc134>. Miner J., K. Dhuyveer, E. E. Davis, and S. Bevers. 2002 Undersanding and using Feeder and Slaugher Cale Basis. Managing for Today s Cale Marke and Beyond. Deparmen of Agriculural Economics, Kansas Sae Universiy. March. Available a <hp://www.agecon.ksu.edu/livesock/maser%20web%20page%20folder/exension%20bulle nins.research/managefortodayscalemk/feederslghrcalebasis2002.pdf>. Oulaw, L. Joe, D. P. Anderson, S. L. Klose, J. W. Richardson, B. K. Herbs, M. L. Waller, M. J. Raulson, S. L. Sneary, R. C. Gill. 2003. An Economic Examinaion of Poenial Ehanol Producion in Texas. Deparmen of Agriculural Economics, Texas A&M Universiy, College Saion, Texas, February. 15
Parcell L. Joe, T. C. Schroeder, and K. C. Dhuyveer. 2000 Facors Affecing Live Cale Basis. Journal of Agriculural and Applied Economics 32(December): 531-541. Sanders, R. D., and M. R. Manfredo. 2006. Forecasing Basis Levels in he Soybean Complex: A Comparison of Time Series Mehods. Journal of Agriculural and Applied Economics, 38 (December): 513-523. Seamon, V. F., C. H. Kahl, and C. E. Curis Jr. 1997. A Regional Comparison of U.S. Coon Basis Paerns. Deparmen of Agriculural and Applied Economics, Clemson Universiy, Working Paper WP123197, December. Taylor, R.M., K. C. Dhuyveer, and T. L. Kasens. 2006. Forecasing Crop Basis Using Hisorical Averages Supplemened wih Curren Marke Informaion. Journal of Agriculural and Resource Economics 31(December): 549-567. Tomek, W. G. and H.H. Peerson. 2001. Risk Managemen in Agriculural Markes: A Review. Journal of Fuures Markes 21(10): 953-985. Tomek, W. G.. 1997. Commodiy Fuures Prices as Forecass. Review of Agriculural Economics, 19(1): 23-44. 16
Figure 1: Corn Consumpion Surplus/Defici in he Unied Saes. 17
Figure 2: Texas Triangle Region Source: hp://agecoex.amu.edu/files/images/maps/triangle.jpg 18
0.35 0.3 0.25 Basis Economic Fundamenals Model Three Year Moving Average Model Dollars per Bushel 0.2 0.15 0.1 0.05 0-0.05-0.1-0.15 Feb-97 Oc-97 Jun-98 Feb-99 Oc-99 Jun-00 Feb-01 Oc-01 Jun-02 Feb-03 Monhs Figure 3: Acual Basis, Basis Predicion from he Economic Fundamenals Model and Basis Predicion from he Three Year Moving-Average Model, using he Complee Sample from Feb. 1997 o Jul. 2008 Oc-03 Jun-04 Feb-05 Oc-05 Jun-06 Feb-07 Oc-07 Jun-08 19
Dollars per Bushel 0.3 0.2 0.1 0 Basis Economic Fundamenals Model Three Year Moving Average Model -0.1 Aug-07 Sep-07 Oc-07 Nov-07 Dec-07 Jan-08 Feb-08 Monhs Figure 4: Acual Basis, Basis Predicion from he Economic Fundamenals Model and Basis Predicion from he Three Year Moving-Average Model, from Aug. 2007 o Jul. 2008 Mar-08 Apr-08 May-08 Jun-08 Jul-08 20
Table 1: Descripive Saisics of he Variables Variables Unis Mean Sandard Dev. Kurosis Skewness Min Basis dollars/bu 0.113 0.09 0.020-0.353-0.140 0.330 Basis Lagged dollars/bu 0.115 0.088 0.025-0.320-0.140 0.330 Average Cash dollars/bu 2.752 0.924 6.570 2.447 1.913 7.110 Average Dec. Fuures Max dollars/bu 2.728 0.941 8.191 2.727 1.890 7.304 Texas Off-Farm in 1000bu 57523.435 32090.190-1.113 0.069 6032 115256 Ending Socks in million bu 1489.130 495.565-1.042 0.107 673 2540 Transporaion index 128.402 18.551-0.048 1.101 111.5 180.3 Sample size: T=138 Table 2: Economic Fundamenals Model Parameer Esimaes, Sandard Errors and - Saisics Variables Parameer Esimaes Sandard Error -saisic p-value Inercep -0.048 0.03716-1.29 0.1987 Basis, lagged 0.47525** 0.07489 6.35 <0.0001 Average Cash Price 0.10327** 0.03343 3.09 0.0024 Avg. Dec. Fuures Price -0.14456** 0.03221-4.49 <0.0001 Ending Socks -0.00002964* 0.0000134-2.21 0.0287 Transporaion 0.00203** 0.00049761 4.08 <0.0001 Dropped (Insignifican) Variables Texas Off-Farm Invenories -0.21041 0.15286-1.38 0.1710 Harves Dummy -0.00746 0.01844-0.4 0.6867 *significan a he 5% level, ** significan a he 1% level Model 1 Roo MSE 0.05239 R 2 = 0.6738 Table 3: Three-Year Moving Average Parameer Esimaes, Sandard Errors and -Saisics Variables Parameer Esimaes Sandard Error -Value Confidence Level Pr> Inercep 0.08879** 0.01508 5.89 <0.0001 Three-Year Moving Average 0.4002** 0.1155 3.46 0.0008 ** significan a he 1% level Model 2 Roo MSE 0.08202 R 2 = 0.1062 21