Heterogeneity in Producer s Marketing Strategy

Size: px
Start display at page:

Download "Heterogeneity in Producer s Marketing Strategy"

Transcription

1 Heterogeneity in Producer Marketing Strategy Tong Zhang Reearch Aitant Department of Agricultural Economic Oklahoma State Univerity 42C Ag Hall Phone no Wade B. Broren Regent Profeor Department of Agricultural Economic Oklahoma State Univerity 44 Ag Hall Phone no Abtract: Producer can make their market timing deciion either baed on fundamental or technical analyi to reach pecific financial target. A generalized mixture model i ued to dicriminate producer into more than one egment according to their marketing trategie. The heterogeneou elling repone i the ame within each egment. Key Word: marketing trategy, heterogeneity, technical analyi, fundamental analyi JEL Claification: C, G0. Selected Paper prepared for preentation at the Southern Agricultural Economic Aociation Annual Meeting Orlando, Florida, February 5-8, 2006 Copyright 2006 by Tong Zhang, Wade B. Broren. All right reerved. Reader may make verbatim copie of thi document for non-commercial purpoe by any mean, provided that thi copyright notice appear on all uch copie.

2 Heterogeneity in Producer Marketing Strategy Introduction Previou tudie try to figure out how producer make their elling deciion. Some tudie argue that producer hould ell their product mainly according to fundamental information like torage cot, tranformation cot (Zulauf and Irwin); while Klumpp and Broren point out that Oklahoma wheat producer poitively repond to fundamental analyi (FA) but do not how much relevant to adviory ervice recommendation (TA); further ome producer follow mechanical marketing trategie that involve elling at the ame time every year. Nearly all previou tudie take all producer in one group and have the ame expected objective function except Penning et. al. But they only examine the derivative uage by producer and group market participant by determinant of hedging behavior, like rik attitude, rik perception...further more ince they tudy hedging behavior intead of product elling activitie in the cah market, and thee determinant are coming from a urvey or experiment, it i poible that producer act differently when they make actual financial deciion. We alo argue that pychological information i already reflected in the actual marketing behavior by following different marketing ignal (FA or TA) in each tranaction. Heterogeneity in Producer Selling Activitie When analyzing behavior, epecially the crop elling activitie of different producer, the homogeneity in deciion maker uually can be rejected. For example, ome producer may have ome trategy to make more tranaction in order to hedge their rik, while ome producer only make a few or even only one tranaction. In the real world, producer may have different trategy behavior and following different rule. The previou tudy aume they are the ame and try to

3 find how they may react under ome condition. Since producer may have different marketing trategie to meet different financial target, they may have different behavior function, either deciion to make a tranaction at pecific time or how many percent he hould ell at each tranaction. In thi paper, we examine both the overall market performance and individual behavior. Objective Firt we examine if there i heterogeneity exit with producer tranaction deciion. According to Klummp and Broren, there are fewer tranaction were made in wheat market if future price pread are higher, and technical analyi information, which indicated by market adviorie uggetion (MAS) ha little effect on it, which mean producer mainly conider FA info and expecting make more return by torage. But the R-quare i very mall. In thi paper, we want to figure it out if not all tranaction following thi rule. Second, we examine what the relationhip between the market information with the percentage of crop old at each tranaction by individual producer. Third, thi tudy dicriminate grain producer into different group according to their market timing deciion. Some producer may ell their product mainly baed on fundamental information, ome may mainly bae on technical analyi, and other may not have preferred information type and have mixed marketing trategy. A generalized mixture regreion model i ued to claify producer into egment, o that the elling deciion repone to the different kind of market ignal are the ame within each egment. Thi model alo etimate the influence of the either fundamental analyi (FA) or technical analyi (MAS) ignal on elling tranaction for each egment identified. 2

4 Data and Method Wheat tranaction data are collected from grain elevator, Pondcreek, located in the northern of wetern Oklahoma, from 995 to Tranaction information include the number of buhel old, price per buhel, and date of tranaction, and the individual who made thi trade. Future pread are ued to repreent the expected return to torage and are calculated baed on Kana City future price. Wheat future contract are old in March, May, July, September, and December. The nearby future pread i the future pread that i nearet to the date of the given tranaction, and the ditant pread i the future pread that i econd nearet to the given tranaction date. For example, the nearby pread for a tranaction with a date of March 6 for a given year would be the difference between the July 6 th future price and the September 6 th future price for the given crop year. The ditant pread for the ame tranaction would be the difference between the September 6 th future price and the July 6 th future price for that year. Market adviory recommendation (MAS) are indicated by how many percentage of crop hould ell by producer. In thi tudy we ue generalized linear mixture regreion model (GLIMIX) to imultaneouly claify producer in the ample into egment on the bai of the relationhip between elling deciion and the market ignal, and etimate the influence of the trading ignal on elling action for each egment identified. The claification i baed on whether producer repond to the trading ignal in the imilar manner. Economic Framework and Method Generalized Linear Mixture Regreion Method If a ample of obervation arie from a pecified number of underlying population of 3

5 unknown proportion, GLIMIX method can be ued to decompoe thoe obervation into different group, each ha pecified denity function (Wedel and Kamakura, 998). Since we do not have priori probability of the producer elling activity regarding the uage the indicator of MAS and the future price pread, we need claifie the producer to eparate the activity into different group uch that the effect of independent variable are the ame in each group. In thi tudy, we group crop elling activity into two group uch that the influence of future price and MAS are the imilar in each group, but diimilar acro group. In each GLIMIX procedure, a certain tatitical ditribution i aumed for each group. In order to implify our problem, we aume thee ditribution are normal ditribution which have different expectation but ame variance. The purpoe of the mixture model i to decompoe the producer population into the underlying egment. Firt, aume the producer repone yn arie from a population that i a mixture of S egment in proportion π, π 2,, π, where we do not know in advance the egment from which a particular vector of obervation arie. The probabilitie of π are poitive and um to one. We aume that the ditribution of y, given that y come from egment, f θ ), i one of n n ( y n the ditribution in the exponential family or the multivariate exponential family, where θ i the vector of regreion coefficient for each egment. Conditional on egment, the yn are independent. The ditribution f ( y n θ ) i characterized by parameter θ. The mean of the ditribution in egment (or expectation) are denoted by μ. Since we want to predict the mean of the obervation in each egment by uing the et of explanatory variable ( wah, dit, ma, nearby ), then we pecify a linear model a follow: 4

6 y i X ip β = P k = p () Where X ik are explanatory variable, β k are parameter in egment ; i =,... n. And are X p y n random effect on. The linear predictor i thu the linear combination of the explanatory variable, and the et of beta that are to be etimated. The beta coefficient can be interpreted a the amount of change in producer ue of the MAS compared to the ituation a captured by figure pread. The unconditional probability-denity function of an obervation, can now be expreed in the finite mixture form: f ( y n S φ) = π f = ( y n θ ) (2) Where the parameter vector φ = ( π, θ ) and θ = β. The parameter vector φ i etimated via maximum likelihood uing the expectation-mixture (EM) algorithm (Redner & Walker, 984; Titterington, 990). By maximizing the likelihood, that et of parameter i obtained that mot likely ha given rie to the data at hand. The etimation algorithm i an iterative algorithm that equentially improve upon ome et of tarting value of the parameter, and permit imultaneou etimation of all model parameter. The EM algorithm i baed on a multinomial ditribution for the memberhip; the expectation of the likelihood can be formulated over the miing obervation. Thi involve calculation the poterior memberhip probabilitie according to Baye rule and the current parameter etimate of φ and ubtituting thoe into the likelihood. Once thi i accomplihed, the likelihood can be maximized. Given the new etimate of φ, new poterior can be calculate in the next E (expectation)-tep, followed by a new M-(maximization) tep to find the new φ. The E- and M-tep are thu alternated until convergence occur. Etimate 5

7 of the poterior probability, p, that obervation of day n come from egment can be n calculated for each obervation vector, a hown in equation (3): y n p n = S π f ( y θ ) = n π f ( y θ ) n (3) We will ue equation (3) to claify producer in a particular egment. Heterogeneity in Trade or Not Deciion One producer elling trategy reflect a the tranaction he made at dicrete day with different percentage of the crop he produce in a crop year. Mot producer in Wet Oklahoma make very few tranaction in a ingle year, uually le then 0, ome of them even only make or 2, and we only have four crop year data. We aume thee producer only follow a few marketing trategy, then all thee producer tranaction deciion may come from everal trading rule poibilitie, each of them come from a pecified denity ditribution. Firt we aggregate all the tranaction made by each producer together, and then uing tranaction frequency in each day a dependent variable, and uing future price pread, week from harvet, and MAS a independent variable, uing GLIMIX to examine if there are two trading rule exit for thee producer tranaction deciion. The tatitic model for tranaction deciion i followed: F t = β + β nearbyt + β ditt + β waht + β mat + ε (4) t The ubcript indicate the different tranaction repone group of producer, t indicate the day that one tranaction made; Dependent variable F i the tranaction number (frequency) happened in one day; independent variable, nearby, dit, wah, and ma t 2 indicate nearby and ditant future pread for that day, number of week after crop harvet, lagged MAS repectively. 6

8 Thi mean the obervation have mixture denity ditribution.then there may be more then one poibilitie marketing trategy rule exit and thi problem i a mixture denity one. Each producer marketing deciion may come from one of the different latent ditribution and they can be ditinguihed by what extent he follow thi rule. For example, ome of them may make their tranaction more concern about fundamental analyi, ome may make more of their tranaction by technical analyi, and the other may have mixed trategy, then producer can be declaified into three egment. In thi paper, we try to figure it out under what condition thoe tranaction were taken. In different group, the β will be different. Heterogeneity in Percentage Trading Strategy Beide tranaction frequency of wheat market, how many percent of crop old in each trade by individual i alo examined. The reaon we ue percentage intead of quantity i that every producer try to make a much a poible profit baed on hi own production quantity. Repect to hi financial target, it i how many percent he hould ell matter intead of actual quantity in each trade, epecially when compare producer behavior. In thi tudy, we aume percentage of each producer crop production i equally weighted by each producer when they make their marketing trategie. We ue each producer percentage trade in each day a equation (5): per it = α + α waht + α nearbyt + α ditt + α mat + ε (5) t Where per i the percentage for individual, which producer i of group old in day t. We it take percentage of each tranaction a dependent variable to ee how the effect of future pread and MAS on producer elling deciion. Expected Reult According to economic theory about fundamental and technical analyi, producer who 7

9 following FA will ell if the current expected return are greater than the maximum expected future return to torage; while for thoe who following TA, they may ignore FA but following MAS to hold becaue they expect higher profit in the future. [Place Figure Approximately Here] Figure (a) to (d) are original data of tranaction frequency repect to different market information. Figure (a) and (b) how that the tranaction number in each day eparate into different group with repone to nearby and ditant future pread. Thi mean producer may follow different rule under poitive future pread condition compare to negative future pread condition. Figure (c) how how tranaction frequency repone market adviory uggetion i nearly normal ditributed with mean nearly equal zero, which mean MAS may have little effect on the ell deciion, which i conitent with Klummp and Broren reult. But thi i for the whole market tranaction; we till want to know if there are ome producer do following MAS more try to make aggreive profit target then other producer. [Place Figure 2 Approximately Here] Figure 2 (a) to (d) how the percentage trade with repect to nearby future pread, ditant future pread, MAS and WAH repectively. We can ee alo the repone cluter into different group, but percentage trade are nearly averagely ditributed along y-axi. Then probably there i no relationhip for producer deciding how much to ell regard to market information. Thi tudy will tet thi hypothei and uing GLIMIX method to tet i there are more than one egment exit that the percentage trading may following pecific rule by producer. Reult Statitical Reult for Tranaction Frequency 8

10 To illutrate the uefulne of the generalized linear mixture-modeling framework we etimated equation (4) acro the whole ample. Table how the OLS and GLIMIX regreion reult of expected return to torage (future price pread) and market adviory ervice recommendation on tranaction frequency of the wheat market. The one egment reult from OLS regreion reulted in a relative low 2 R of 0.02, indicating that ignoring heterogeneity reult in a model that can explain only.2% of the variance of producer repone to the cenario. [Place Table Approximately Here] Account for heterogeneity poible exit, GLIMIX model i ued to dicompoe the data et uing equation () through (3). To ae the eparation of the egment, an entropy tatitic can be ued to invetigate the degree of eparation in the etimated poterior probabilitie a defined in equation (7): E = N S n= = p n N ln S ln p n (7) Where p the poterior probability that crop producer i n come from latent group. For n example, the entropy value of 0.8 indicate that the mixture component are well eparated, that i, the poterior probabilitie are cloe to or 0. The mixture regreion how there are two egment exit. Note, that thee egment are defined by the mixture model baed on tatitical difference in the etimated regreion coefficient for each egment. That i, the egment reveal different behavior with repect to the likelihood of information of future pread and MAS ue. The reult for the two-egment model are compared with OLS in Table. The GLIMIX reult how that the coefficient of thee two group are not ignificantly different except nearby future pread. Tranaction ha 5.9% and 48.8% poibility made 9

11 following rule in egment and 2 eparately. Coefficient for nearby future pread are negative and thoe of ditant future pread are poitive, but the abolute value of nearby future pread coefficient are larger then thoe of ditant future pread. Thi mean that the expected hort run torage return ha more effect on producer to make ell or not deciion. The higher nearby future pread, the le chance tranaction happen. Thee reult are conitent with a marketing trategy that ue fundamental analyi. The coefficient of WAH for thee two egment have ame negative value. Thi reult how that producer are more likely to make more elling tranaction right after harvet. Both egment how negative relationhip with market adviory recommendation, which indicate that producer do not care market adviory adviory or even trade oppoite to thoe recommendation, which i conitent with Klumpp and Broren reult. Econometric Reult for Percentage Trading in Each Trade Made by Individual Now we examine the percentage trade in each tranaction of in wheat market. Table 2 how the relationhip between percentage of each tranaction and market information. [Place Table 2 Approximately Here] From the above table, the reult how that the percentage ell by one producer in each tranaction will increae according to time and nearby future pread, negative to ditant future pread. And the regreion alo find out that the percentage elling by producer doe not have relationhip with MAS ignificantly. Conider WAH data range i from 0 to 49, while future pread i from - to, then the combine effect of ditant future pread and WAH are higher than that of nearby future pread. Thi mean producer mainly conider long run profit then hort run torage profit from torage. Thi maybe becaue that for thoe who have low torage cot in that year (uch a they build torage place themelve before), will prefer torage to make more profit before next harvet time. But the 2 R i only , which mean only 2.43% of data are 0

12 explained by thi model. From Figure 2, we can ee that the linear relationhip between percentage trade and future pread, WAH and MAS are not clear. We can ay that mot producer do not have trongly rule like following FA or TA. The reult how that even producer believe they can make more profit by trying to ell different percentage of crop according to market information, from tatitic point of view for the whole market, how many percent a producer ell in each trade are randomly chooe at different ituation. Thi reearch alo compare each producer trading trategy and we did not find ignificantly difference acro different producer. Concluion Thi paper tudie whether wheat producer marketing trategie are different under different condition and from the whole point of view, if producer have different elling rule in Wetern Oklahoma. The reult how that producer care little about how market adviory ugget them to do, which mean they do not following technical analyi to ell their product. The reult aociate with tranaction frequency indicate that producer are reluctant to ell if the future pread i poitive and hope make more return by torage. But when future price pread are negative, producer may more likely to ell their product regard little of the market information, no matter fundamental or technical analyi. In addition, thi paper alo how that producer do not have different trading rule ignificantly repect to percentage trade in each tranaction. Even producer eem do have trading philoophy when they decide ell or not at the current ituation, eem they do not know how much they hould ell and jut make their deciion randomly.

13 Reference Anderon, K.B., and H.P. Mapp. Rik Management Program in Extenion. Journal of Agricultural and Reource Economic 2(996):3-38. Benirchka, M., and J.K. Binkley. Optimal Storage and Marketing Timing over Space and Time. American Journal of Agricultural Economic 2(990): Fama, E.F., and K.R., French. Commodity Future Price: Some Evidence on Forecat Power, Premium, and the Theory of Storage. The Journal of Agricultural Buine 60(987): Irwin, S.H., D.L. Good, J.Martine-Filho, and L.Hagedorn. The Pricing Performance of Market Adviory Service in Corn and Soybean Over AgMAS Project Reearch Report Internet ite: (March 2005). Klumpp, J.M., and B.W. Broren. The impact of Marketing Strategy Information on the Producer Selling Deciion. NCR-34 Conference on Applied Commodity Price Analyi, Forecating, and Market Rik Management 2005 Conference, April 8-9, Penning, J.M., O. Iengildina, S. Irwin, and D. Good. The Impact of Market Adviory Service Recommendation on Producer Marketing Deciion. Journal of Agricultural and Reource Economic. 29(2004): Penning, J.M., and P. Garcia. Hedging Behavior in Small and Medium-ized Enterprie: the Role of Unoberved Heterogeneity. Journal of Banking and Finance. 28(2004): Tomek, W. Commodity Future Price a Forecat. Review of Agricultural Economic 9(997): Working, H. The Theory of Price of Storage. The American Economic Review 39(949):

14 Zulauf, C.R., and S.H.Irwin. Market Efficiency and Marketing to Enhance Income of Crop Producer. Review of Agricultural Economic 20(998): Wedel, M., W.A. Kamakura. Market Segmentation: Conceptual and Methodological Foundation. In: International Serie in Quantitative Marketing. Kluwer Academic Publiher, Boton

15 Appendix: Table and Figure. Table. OLS v. Mixture Regreion Reult for Tranaction Frequency of Whole Market a Regreion Coefficient Etimate (Standard error in bracket) OLS model ( S = ) LIMIX model ( S = 2 ) = = 2 Ditant Future Spread (0.2266) Nearby Future Spread (0.240 ) Lagged market advie ** (0.0354) Week after Harvet (0.0029) Intercept (0.0728) Proportion of producer in egment (π ) R 2 = a Two aterik indicate ignificance at the 95% level Table 2. OLS Regreion of Percentage Selling in Each Tranaction b Regreion Coefficient Etimate (Standard error in bracket) Etimate t-value Pr > t Ditant Future Spread **** ( ) Nearby Future Spread ( ) Week from Harvet ( ) Intercept ( ) R 2 = < <.000 b Four aterik indicate ignificance at the 85% level. 4

16 Trancation Frenquency v. Ditant Future Spread Frenq uency Ditant Future Spread (a) Frequency v. Nearby Future Spread (b) Frequency v. Ditant Future Spread Tranaction Frenquency Tranaction Frenquency v. MAS MAS Tranaction Frequency v. Week from Harvet (WAH) Frequency WAH (c) Frequency v. MAS (d) Frequency v. WAH Figure. Tranaction Frequency v. Future Spread, WAH and MAS 5

17 Percentage vs. Nearby Future Spread Percentage v. Ditant Future Spread Percentage trade Nearby Futur Spread Percentage Ditant Future Spread (a). Percentage v. Nearby Future Spread (b). Percentage v. Ditant Future Spread Percentage v. MAS Percentage v. WAH Percentage MAS Percentage WAH (c). Percentage v. MAS (d). Percentage v. WAH Figure 2. Percentage of One Tranaction v. Future Spread, WAH and MAS 6