Price Transmission in Non Competitive Market Channels: A Study of the Boston Fluid Milk Market and the North East Dairy Compact

Transcription

1 Price Transmission in Non Cometitive Market Channels: A Study of the Boston Fluid Milk Market and the North East Dairy Comact Tirtha P. Dhar (Research Associate, Food System Research Grou, Det. of Agricultural and Alied Economics, University of Wisconsin-Madison) Ronald W. Cotterill (Director, Food Marketing Policy Center, Det. of Agricultural and Resource Economics, University of Connecticut) June 4, 00 Contact Author: Ronald W. Cotterill, U-60, Det. of Agricultural and Resource Economics, University of Connecticut, Storrs, CT Ronald.cotterill@uconn.edu Tel: (860) Fax: (860) Suort from the USDA CSREES Secial Research Grant No at the University of Connecticut is gratefully acknowledged. Comments from seminar articiants at Yale University and the University of Wisconsin on earlier drafts are gratefully acknowledged. The usual disclaimer alies.

2 Price Transmission in Noncometitive Market Channels: A Study of the Boston Fluid Milk Market and the North East Dairy Comact Introduction: Research in agricultural economics on rice transmission has concentrated almost exclusively on homogeneous roducts, aggregate national data from secondary sources such as the USDA, and models that assume the market channel is a single industry of cometitive firms (e.g. Gardner 975, Heien 980, Kinnucan and Forker 987). Studies in the Agricultural Economics literature tyically are reduced form time series analysis of farm and retail rices, with no attention to the underlying structural model of the marketing channel. Recently McCorriston, Morgan and Rayner (998) relax the cometitive assumtion and secify a structural model; however, they continue to maintain a single stage industry and homogeneous roduct focus. Moreover, rather than estimate or test their model, McCorriston, Morgan and Rayner assumed its validity and used arameter estimates from other aggregate studies to simulate rice transmission for agricultural roduct industries. Ashenfelter et al., (998) working on the Stales-Office Deot merger case also focus on one stage in a marketing channel. Their artial equilibrium model analyzes two tyes of cost shocks-industry wide and firm secific, but does so only in a residual demand framework that does not squarely distinguish between industry and firm secific cost shocks. Their rimary concern is measuring firm secific effects. In this aer we develo a more general model for the analysis of rice transmission, or what more generally is called the cost ass through rate (CPTR). We secify two stages in the fluid milk marketing channel, rocessing and retailing, and we

3 estimate CPTRs for individual firms in a differentiated roduct oligooly at each stage of the marketing channel. This structural aroach exlicitly measures CPTRs for firm secific as well as industry wide cost shocks. To measure both tyes of cost ass through we advance the theory and emirical analysis by introducing a more disaggregate structural model that identifies strategic cross firm rice shocks and corresonding ass through rates. Given an oligoolistic market structure, a firm secific shock may not only influence that firm s own rice level; it also may cause other firms to react to that rice and change their rices (Cotterill 994, 998, Cotterill, Putsis and Dhar 000). Our structural model secification is more general than the Ashenfelter et al. residual demand aroach because it measures cross rice effects, rice imacts on all firms (brands) in the market as oosed to just one or two margin firms, and it allows us to secify alternative conduct games and test to determine which best relicates observed market conduct. This aer uses Information Resources Inc. (IRI)-Infoscan database for fluid milk roducts for each of the to four suermarket chains in Boston (Sto & o, aw s, Star Market and DeMoulas). The data are for four week eriods from March 996 to July 998. This eriod includes the stabilization of the farm level fluid milk rice due to the advent of the Northeast Dairy Comact (NEDC). Thus we are able to analyze how each chain and fluid milk rocessor changed fluid milk rices when the farm level fluid milk rice, an industry wide cost shift variable, changed. The evidence strongly suggests that rocessors and retailers exercised market ower thereby increasing retail rices and margins by more than one would have observed in a cometitive market channel.

4 Cost Pass Through Models for a Market Channel with Differentiated Product Oligoolies: Here we secify horizontal cometition both at the rocessing and retail level as Nash in rices and assume Bertrand rice cometition exists among retailers in a differentiated roduct oligooly. Processors and retailers may lay different ricing games in the vertical channel. We secify three different games: suermarkets with ustream integration (comlete vertical coordination game), a vertical Nash model where each suermarket chooses an exclusive rocessor and rocessors and retailers maximize rofit simultaneously by deciding, in arms length fashion, on the wholesale and retail rice, and finally a vertical Stackelberg game where in the first stage a retailer decides on the rofit maximizing rice and then a rocessor maximizes rofit taking into account the reaction function of the retailer. Our work assumes vertical dyadic relationshis between rocessors and retailers, i.e. each retailer deals with one exclusive rocessor of milk. Other researchers on vertical structure models have the same constraint, e.g. Kadiyali, Vilcassim and Chintagunta (996, 998). For this market the assumtion is not an extreme dearture from actual organization. Sto & o rocessed all of its rivate label and Hood brand milk (under license from Hood) in its own lant. The other chains each received all of their rivate label milk and a leading brand of milk from a single rocessor. In each case this accounted for more than 75% of milk sales. Modeling cometition among rocessors as a vertical game through retailers rather than a direct horizontal game among rocessors at the wholesale level seems sufficient and reasonable. Processors comete with each other, contingent on retailer behavior, in the retail market for the sale of their roducts. 3

5 For simlicity of exosition we resent a two retailers two rocessors model. In the emirical section of this aer we extend the model to four retailers and four rocessors. Let the demand functions of the retailers be the following: q = a 0 + a + a q = b 0 + b + b [a] [b] Processor level demand is derived from the retail level demand secifications given retail conduct and margin. To derive these rocessor level demand functions different conjectures are assumed at the rocessor level concerning retailer reactions. These conjectures can be erceived as assumtions by the rocessors about retailer ricing behavior given a wholesale rice. For the vertical integration (full coordination) game we need no vertical conjecture assumtions. Let the retailer s cost function be the following: TC = w *q TC = w *q [a] [b] where: w and w are the wholesale rices received by the rocessors. So, the retailers rofit functions can be written as: Π R = ( w )*q Π R = ( w )*q [3a] [3b] Following Choi (99), in the Vertical Nash game, a linear mark-u at retail is conjectured by the rocessor on retail rice; so, retail rice can be written as: = w + r [4a] 4

6 = w + r [4b] where: r and r are the linear mark-u at the retail level. In the Stackelberg game, each rocessor develos a conjecture from the first order condition of the retailer. The retailer s first order conditions are: = w = w a b ( a + a ) 0 ( b + b ) 0 [5(a)-(b)] We assume that each manufacturer only knows its own retailer s reaction function, i.e. the manufacturer does not know, and thus ignores the imact of its wholesale rice change on the other retail rice. The resulting Stackelberg conjectures are: w = Conjecture and w = Conjecture. In fact, once we have estimated the model, testing the estimated CPTRs against the conjecture values for vertical Nash () and manufacturer Stackelberg (/) gives us information on which is the most aroriate game. We simlify the rocessor level marginal cost function in the following manner: wmc = m + m wmc = m + m [6a] [6b] where: m is the industry secific marginal cost comonent and m and m are the rocessor secific cost comonents. So, the rocessors rofit functions can be written as:? P = (w m m )*q [7a]? P = (w m m )*q [7b] 5

7 Using the rofit maximizing first order conditions both at the rocessing and retail level we derive the cost ass through rate (CPTR) equations. Table a gives the CPTR for farm to wholesale and from wholesale to retail for the vertical Nash and vertical Stackelberg games. Note that they are only functions of the demand arameters. This is due to the constant marginal cost assumtion. We also give the arameter values for two secial cases: retail monooly and the strong substitutes case. Slade (995), Choi (99), Cotterill, Putsis and Dhar (000) and others have modeled vertical interaction by assuming that retail sales are made by a monoolist that is sulied by more than one manufacturer. Here we assume the converse (multile retailers each sulied by a single manufacturer). Table gives detailed comarative static results for our models. If in fact our retailers are monoolists then row one of Table a indicates that the transmission rates for changes in rocessor s marginal costs to the wholesale rice are / for manufacturer Stackelberg and /3 for vertical Nash. For the CPTR between a firm s own wholesale and retail rices (Row 7, Table (a)), if we have retail monoolies the rates for both the vertical Nash and Stackelberg game reduce to/. Table a also gives the CPTR for what we call the strong substitute case. When, the own and cross rice coefficients are equal in absolute value, for examle, a = a a 5 cent increase in has the same effect as a 5 cent decrease in on the demand for q and the industry total CPTR is equal to for both the vertical Nash and Stackelberg game. Firm secific CPTR in the strong substitute case assume exact numerical values, however are not zero nor are they one. We thus conclude from these two secial cases and insection of the general CPTR formulae that industry wide and firm secific cost shocks do not roduce identical 6

8 results. In the strong substitute case, for examle, one obtains effectively cometitive conduct for industry wide cost shocks and noncometitive conduct for firm secific cost shocks. [Tirtha, exlain the corresondence N Good Results]. Returning to the two good formulae in Table a, lets exlore more carefully the nature of noncometitive firm secific cost shocks. If each rocessor is in an effectively cometitive market osition we would exect (dw / ) and (dw / ) to be zero (rows 3 and 6, Table a). Changes in firm secific marginal cost could not or would not be assed on. Thus if manufacturer firm secific CPTR are not zero we have an affirmative test for market ower. Even if we do not observe retail monooly, i.e. the quantity of milk sold at one chain is sensitive to the rices charged by one or more of the other chains, individual retail chains may also ossess a modicum of market ower, retail level firm secific CPTRs (rows 7 and 0, table a) are ositive. This also remains true even if we have strong substitutes and effectively cometitive behavior with regard to industry wide cost shocks. Therein lies an imortant theoretical result. Studies of rice transmission at the industry level can not definitively rule out the exercise of market ower at the firm level. This is esecially true for studies that characterize differentiated roduct industries as homogeneous because firm secific effects are ignored. Using the derivations in Table a one can also derive the cost ass through imact of rocessor level industry and firm secific cost shocks on retail rices. We call these total cost ass through rates. The following total CPTR relationshis hold: In the case of industry wide shocks: 7

9 d = d dw d = d dw dw[] dw[] dw dw [] = [] [] = [] + d dw + d dw dw[] dw[] dw dw [] = [] [] = [] [8 (a)-(b)] Similarly, for channel secific shocks: d d d = dw d = dw dw[] dw[] dw dw = [] = [] + d dw + d dw dw[] dw[] dw dw = [] = [] [9(a)-(b)] d d d = dw d = dw dw[] dw[] dw dw = [] = [] + d dw + d dw dw[] dw[] dw dw = [] = [] [0(a)-(b)] Table b gives the formulae for the total CPTR and their values if we observe retail monoolies or strong substitutes. The same qualitative analysis of industry and firm secific effects holds. In fact we will test for the numerical values of both secial cases when we estimate the model. Table b also gives the CPTR for the integrated or fully coordinated (erfect vertical tacit collusion) game. Note that when one eliminates the double margnialization that occurs in the vertical Nash and Stackelberg games the ass through rates, in the retail monooly case, increase to /. The same is true in the general case, i.e. full coordination reduces double marginalization and increases the CPTR. Conversely evolution of a channel into successive monoolies at several stages with non-coordinated ricing reduces rice transmission. This issue is clearly on the table in many food industries where owerful rocessors sell to owerful retailers. 8

10 Finally note that in Tablea-b that (dw / = dw / ), (d /dw = d /dw ) and and (d / = d / ) in this two-erson game, however this is unique to the two rocessor-two retailer vertical dyadic game. In a game with more than two layers they will not be equal. Variable Definitions and Model Secification: We use IRI scanner data that include quantities sold, average rice er gallon, average ackage size sold, for the four leading retail chains (Sto & o, aw s, Star Market, and DeMoulas) in the Boston market. The data series starts in March 996 and each observation is a four-week eriod with thirteen observations for each calendar year. The fluid milk category covers disaearance of skim/low fat and whole milk within a retail chain. In the resent model, farm level fluid milk rice will be taken as exogenous. We use the Boston Federal Milk Marketing Order Class-I milk ay rice for the farm level milk rice series. 3 Since the Federal Milk Marketing Order sets this rice, based on national manufacturing milk rices and a differential set in the 995 farm law, the assumtion that the farm level fluid rices for Boston are exogenous is not unrealistic. Demand for fluid milk in Boston does not areciably affect the national suly-demand system for manufacturing milk uon which the New England farm level fluid rice is based. To identify the demand side we secify weighted average ercentage rice reduction, a measure of trade romotion activity, for each retailer in each demand equation. To identify the suly side we secify the measure of volume er unit, for each retail chain. Variation in average volume er unit (e.g. shifting from 0.5 to gallon er 9

11 unit sold) catures exogenous cost comonents related to ackage size; so, we use it as a suly side variable. Emirical Estimation Procedure: To estimate our models, we secify the fluid milk demand equations for retailers and the aroriate rofit maximizing first order conditions. We secify linear demand functions for the convenience of estimation and tractability. The demand equations are: q q q q D = i + a = i = i = i 4 + b 3 + c + d + a + b + c + d + a + b c + d a + b 4 + c d 4 D D D + ϑ D + ϑ + ϑ + ϑ WRR D WRR WRR WRR D [(a)-(d)] where, q and are quantity and rice variables; and the subscrit -Sto & o, -aw s, -Star Market and D-DeMoulas. We close the model with the following linear marginal/average cost function: mc i = m + m + η VPU [] i i i where, m i (i =,,, D) are the firm secific unobserved (to the econometrician) cost comonent, m is the rice of raw milk, and VPU i (volume er unit) catures the cost comonents related to ackaging. 4 The unobserved cost comonent will be estimated within the system as the intercet for each firm first order condition. The coefficient on m is restricted to be since we assume that all changes in raw milk cost are by definition incororated into marginal costs.? i is a arameter that will be estimated. For our vertical Nash and Stackelberg model, we have two rofit functions that need to be maximized. For the full Coordination game, the two rofit functions become one for the vertically integrated firm. 0

12 At the retail level we have the following rofit function:? R i = ( i w i )*q i [3] and at the rocessor level:? P i = (w i mc i )*q i [4] By maniulating the first order conditions derived from the two rofit functions we obtain the following estimable first order conditions: DM = m+ m = m + m = m+ m = m+ m + η + η DM + η + η VPU VPU DM VPU VPU k a q k b q DM k c q 3 k d 4 + γ + γ q D + γ COMP COMP + γ DM COMP COMP [5] Here, when k =, then the first order conditions will reresent the full coordination game, k = reresents first order conditions from vertical Nash game, and k = 3 reresents the manufacturer Stackelberg game. The four first order conditions with or without the Comact binary discussed below, and with k =, or 3 and the four demand equations are the models that we estimate with nonlinear 3SLS regression using SHAZAM (ver. 8). The imlementation of the North East Dairy Comact near the midoint of our 3 samle eriod suggests that we add a binary variable (COMP) to the first order conditions. It has value after Comact imlementation in July 997 for two reasons. First, according to neoclassical theory for risk averse firms, when one reduces or eliminates inut rice risk the margin for both cometitive (Turnovsky 969) and noncometitive (Azzam 99) firms unambiguously narrows. One no longer has a risk remium built into the margin. Congress used this fact to conclude that rice

13 stabilization via the Comact could raise rices to farmers with a less than commensurate increase in retail rices (Federal Register 997). If this is the case than the binary variable should have a significant negative coefficient. Alternatively, similar to the ethyl case (Hay 999) and as suggested by the focal oint theorem (Schelling 960) the imlementation of the Comact, a distinct non-market event with considerable oortunity to signal rice intentions, may have facilitated collusive ricing by rocessors and retailers. This redicts a ositive inut for the comact binary. Estimation Results: The vertical Nash, vertical Stackelberg and full coordination models were estimated using nonlinear 3SLS. 5 We use the Davidson-Fletcher-Powel algorithm and minimize the error sum of squares in the linear seudo model. The Vuong (989) test gave no significant guidance for choosing one model over another. A artial test for the aroriate model is to comare the estimated wholesale to retail rice CPTRs against the assumed conjectures in each model. The Stackelberg model fits better than the Vertical Nash. The CPTR from wholesale to retail for Stackelberg conduct are 0.5 and our estimates are closer to 0.5 than the Vertical Nash value (.0) (see Table 3 a-b second column). Therefore we will focus on the Stackelberg results in the text. The results for the other two models are quite similar to the Stackelberg and are available from the authors. Table resents the regression results for the Stackelberg game with and without the Comact binary. We use the Vuong test to distinguish between models with and without the Comact binary. With a chi-square test statistic of 0.5 and degree of freedom, the model with the Comact binary erforms better at the % significance level.

14 All four own rice coefficients in the demand equations are negative and statistically significant at the 5 % level, or higher, in the model with or without the Comact binary. Cross rice coefficients generally are ositive. More are statistically significant without the Comact binary because it catures the effect of the largest rice change in the data set which otherwise hels to identify switching conduct by consumers. Without the Comact binary all chains have at least one significant substitute. For the model with the binary only Sto & o has no significant substitute. Generally aw s and DeMoulas ricing tend to affect other chains volume more than Sto & o and Star Market rices do. In conclusion fluid milk consumers do, to some significant extent, switch their urchases from one chain to another based on rice. For less visible and less frequently urchased roducts, and roducts that are not romoted with eriodic rice secials, one would more likely find less switching between chains, i.e. retail monooly and lower rice transmission. The estimation results for the cost arameters dislayed in Table are robust. Seven of the eight arameters are significant at the 5% or better level and the signs are correct in both models (with and without the Comact binary). In table 3 we resent the estimated ass-through rates and assorted statistical tests for the Stackelberg game with the Comact binary secified in the model. 6 The ass through rates are resented for each chain, for each stage, and for the total channel. First we examine the imact of industry wide cost shock (changes in the farm level milk rice) and then we will discuss firm secific cost shocks. For each of these chains farm level milk rice shocks are almost comletely transmitted to the wholesale and retail level. The cost ass through rates vary from to and all are not significantly different from.0. Returning to Table these results clearly suort the strong 3

15 substitutes secial case. They are based uon the fact that the combined effect of the estimated, ositive cross rice demand coefficients are nearly equal to the absolute value of the own rice coefficients reorted in Table. This near erfect transmission of farm milk rice shocks is based only on the demand system coefficients. More secifically it is not based on regressing retail rices on farm rices in a Gardner/Hein reduced form cometitive channel model. In Figure there is almost no contemoraneous correlation between the retail rice series and the farm rice series, excet for the month that the Comact was imlemented. Regressing the retail rice on farm rices with or without the Comact binary roduces very low transmission rates ranging from 0.6 to Clearly more research is needed on the relation between reduced form and structural rice transmission models. ifting now to the firm secific results one finds more mixed results, but, there is clear suort for strategic interaction between firms and noncometitive ricing. All of the firm secific own ass through rates are significantly different from zero at the 5% level or better. Ashenfelter et al. reorted a 5% firm secific own ass through rate for Stales, an office suerstore chain. As reorted in Table 3a we find higher own ass through rates ranging from 0.55 to 0.65 from rocessor to wholesale and ranging from 0.54 to 0.6 from wholesale to retail. Consider the Sto & o results in Table 3a. The CPTR to the wholesale level from a unique cost shock to its rocessor is This value is not significantly different from the retail monooly value of 0.5 but it is significantly different at the 5% level from the strong substitutes value of five sevenths (0.743). [Tirtha, this needs to be droed here and elsewhere. It assumes we are in a good model. Lets just discuss tests for the strong subs case.] The firm secific CPTR for Sto & o from wholesale to retail 4

16 is (Table 3a). Again this value is not significantly different from the retail monooly value (0.5) but it is different at 5% level from the strong substitutes value (0.667). The total CPTR for a firm secific shock at Sto & o behaves in the same fashion. The total CPTR is quite low at 0.35 and it is not significantly different from the retail monooly value, 0.5. However, it is different at the 5% level from the strong substitutes value (0.574). The firm secific CPTR for Star Market follow exactly the same attern as Sto & o. The CPTR to wholesale, from wholesale to retail and the total CPTR are significantly different at the % level from the strong substitute values but we cannot reject retail monooly. For aw s the test results in Table 3a indicate that the ass through rates for firm secific cost shocks are significantly different at the % level from the values for both secial cases. For DeMoulas (Table 3b) we reject the retail monooly secial case at the % level and find conduct consistent with the strong substitutes secial case. Lets now examine cross firm rice transmission in Table 3. Significant cross shock cost ass through rates range from 0.4 to 0.3 from farm to wholesale from 0.6 to 0.6 and from wholesale to retail. If one is examining cost decreases due to a merger then an added benefit not considered by Ashenfelter et al. in a noncometitive market lace is the rice reductions of other firms when the merging firm has unilateral cost savings. Focusing on Sto & o and Star Market note that unilateral cost shocks to their rocessors and unilateral changes in their wholesale rices have no imact on aw s or DeMoulas rices, nor on each other s rices. These chains seem to be ricing in a vacuum. This clearly is not the case for aw s and DeMoulas. Unilateral cost shocks to their rocessors and unilateral changes in their wholesale rices have ositive 5

17 and significant imacts on Sto & o, Star and each other. These asymmetric rice interdeendencies indicate that rocessor initiated increases in wholesale rices only to aw s and DeMoulas will also increase retail rices at Sto & o and Star Market, thereby widening these latter two chain s rofit margins. Turning now to analysis of the industry s resonse to the North East Dairy Comact rogram, Figure dislays the farm fluid milk rice aid by rocessors and the four chains retail milk rices in the Boston market from March 996 to July 998. The vertical line denotes the advent of the Comact ricing rogram which required that New England rocessors ay a minimum rice of $6.94/CWT or $.46/gal for milk rocessed for fluid consumtion. Since the Federal Class-I rice for Boston remained below this level for the rest of the samle eriod the farm rice remains constant at $.46/gal. Although the Comact is often described as a rice enhancement rogram, it also is a rice stabilization rogram. In fact it is described as such in the final rule issued by the federal government (Federal Register 997). As given in Table 4 the farm level milk rice averaged $.40/gal rior to the NEDC and it was extremely volatile with a standard deviation of $0.0/gal. The Comact raised the rice $0.06/gal above its exected rice level to $.46/gallon; however it reduced rice variance to zero. Returning now to Table let s examine the imact of the Comact binary variable on retail milk rice, above and beyond the imact of changes in the farm level milk rice, m, to see if there was a structural shift in the model. Two oosite imacts are: ) the elimination of a risk remium and a consequent reduction in the margin or ) a shift to more collusive ricing with a widening of the margin at the July 997 focal oint. The evidence for each of the chains clearly indicates that margins, ceteris aribus, did not narrow when rice risk to the key inut was eliminated. To the 6

18 contrary a shift to a new less cometitive ricing regime raised rices nine cents a gallon at Sto & o and aw s and eleven cents a gallon at Star Market and DeMoulas. These rice increases significant at the 0% or 5% level. It is instructive to comare these redicted rice and margin changes to the actual observed changes. Table 4 gives the actual farm to retail rice sread (channel gross rofit margin) and its variance for each chain during the before and after NEDC eriods. For each chain the actual margin widened and its standard deviation droed dramatically. For examle the Sto & o margin in the re Comact eriod averaged $.05/gallon and its standard deviation was $0./gallon. After Comact imlementation its average margin increased 7 cents to $./gallon and the standard deviation for its margin droed 83% from cents to cents er gallon. When viewed in this fashion farmers increased the ost Comact average rice 6 cents er gallon over the re Comact average rice when they increased the rice 8 cents from June to July in 997. Retailers and rocessors closed ranks around an identical 8 cents er gallon increase that locked in their very wide margins. Those wide margins on the eve of Comact imlementation were due to their insensitivity to rice moves at the farm level and the fact that farm rices were in a trough. Consequently the channel firms tit for tat 8 cents move increased their average margin in the ost Comact eriod over the re-comact eriod average margin. Actual margins increased 7 cents at Sto & o, cents aw s, 7 cents at Star Market, and 4 cents at DeMoulas. We conclude, based uon redicted and actual margin changes that between one and two thirds of the rice increase at retail that generally has been attributed to North East Dairy Comact gains for farmers went to rocessors and/or retailers. 7

19 Conclusions and Suggestions for Future Research: One can advance our understanding of rice transmission and strategic ricing by marketing firms through structural modeling. It is ossible to decomose, and test cost ass-through rates within the context of exlicit strategic games. This is the first research effort to systematically exlain the theoretical differences between transmission of industry wide and firm secific cost shocks in a structural model. The result is a much more detailed analysis of market channel conduct. Industry wide and firm secific cost shocks are not identical, nor does the latter necessarily aggregate to the former. In a differentiated roduct industry one can observe 00% transmission of an industry wide cost shock. This is a necessary but not sufficient condition for effective cometition. If one has nonzero transmission of firm secific cost shocks, then one does not have effective cometition. Cross CPTRs, in emirical industrial organization hel to exlain firm behavior and strategic interaction among firms. Industry cost ass through rates can, and in this study do, suggest cometitive transmission of industry wide cost shocks while firm secific cost ass through rates can, and in this study do, suggest noncometitive ricing conditions for individual firms. We find that the ass-through rates due to changes industry wide costs, i.e. the farm level milk rice are near 00%. Pass through rates at wholesale and retail for firm secific cost shocks fall generally around 50 ercent. For the total channel, firm secific ass through rates are near 5%, the retail monooly level. Sto & o and Star Market seem to enjoy retail monooly status more than aw s and DeMoulas. The effectively cometitive result at the industry level differs from the corresonding contemoraneous reduced form rice transmission result. Insection of Figure and regressing retail rices on contemoraneous farm rices gives reduced 8

20 form ass through rates below 0.5 for an industry cost shock, i.e. the change in raw milk rice. This dichotomy suggests that secific structural models are too constrained to cature observed rice conduct. Perhas relaxing the linear demand and cost assumtions in the structural model will resolve this anomaly, but we doubt that it would fully exlain the dichotomy. The missecification also may reside in the contemoraneous reduced form model secified here. One may need to secify other cost variables. Also one may need very long run dynamic model. Secifying one or two month lags as we did is not sufficient. Yet if this sort of market smoothing inut rice model holds for all firms with no defectors it suggests, iso facto, the exercise of market ower. Inut rice volatility is shifted back to farmers unless stabilization rograms, such as the Dairy Comact or a return to federal stabilization rograms such as the rogram that has relaced the Dairy Comact, damen the increased farm rice fluctuations. In conclusion, one can no longer analyze farm olicy imacts on consumers in a model that assumes erfect cometition and homogenous roduct ricing of rocessed food roducts. One must also carefully distinguish between industry and firm secific cost shocks. The same is true for analysis of merger secific cost efficiencies. References Ashenfelter, O., D. Ashmore, J. B. Baker and S.M. Mckernan. Identifying the Firm-Secific Pass-Through Rate. Federal Trade Commission Bureau of Economics Working Paer No. 7, January 998. Azzam, A. and J. Schroeter. Marketing Margins, Market Power, and Price Uncertainty. Amer. J. of Agr. Econ. 73 (November 99): Baker, J.B., and T.F. Bresnahan. The Gains from Merger or Collusion in Product-Differentiated Industries. Oligooly, Cometition and Welfare. Geroski, P. A., L. Phlis, and A. Ulh, eds O. Oxford and New York: Blackwell in cooeration with the Journal of Industrial Economics, Estimating the Residual Demand Curve Facing a Single Firm. International Journal of Industrial Organization 6 no. 3 (988): Berry, S., J. Levinsohn, and A. Pakes. Automobile Prices in Market Equilibrium. Econometrica. 63(July 995): Choi, S.C. Price Cometition in a Channel Structure with a Common Retailer. Marketing Science. 0 (Fall 99):

21 Cotterill, R.W. Scanner Data: New Oortunities for Demand and Cometitive Strategy Analysis. Agr. and Res. Econ. Rev. 3(994): Continuing Concentration in the US: Strategic Challenges at an Unstable Status Quo. Financial Times Retailing Monograh 999 (also available as Research Reort No. 49 at htt:// ).. Estimation of Cost Pass Through to Michigan Consumers in the ADM Price Fixing Case. Research Reort No. 39. Food Marketing Policy Center, 998 (also available as Research Reort No. 39 at htt:// Neoclassical Exlanations of Vertical Organization and Performance of Food Industries. Agribusiness, forthcoming. Cotterill, R.W., W. P. Putsis; Jr. and R. Dhar. Assessing the Cometitive Interaction between Private Labels and National Brands, Journal of Business 73(000): Dhar, T.P., R.W. Cotterill., A Structural Aroach to Price Transmission in Non-Cometitive Market Channels: A Study of the Fluid Milk Market. Paer Presented at the USDA-ERS Conference on The American Consumer in the Changing Food System ; Washington D.C, May 000. Federal Registrar, North East Dairy Comact Commission. Part VI, 7 CFR, Chater XIII, 6(04), May 30, 997,.967. Gardner, B.L. The Farm-Retail Price Sread in a Cometitive Food Industry. Amer J. of Agr. Econ. 57(August 975): Gasmi, F., Q.H. Vuong. An Econometric Analysis of Some Duoolistic Games in Prices and Advertising. Econometric methods and models for industrial organizations. Advances in Econometrics-vol. 9. George F. Rhodes, Jr., ed., 5-54., Greenwich, Conn. and London: JAI Press, 99. Harris, R. G., and L. A. Sullivan. Passing on the Monooly Overcharge: A Comrehensive Policy Analysis. University of Pennsylvania Law Review 8 no. (979): Hay, G.A. Facilitating Practices: The Ethyl Case (984). The Antitrust Revolution: Economics, Cometition and Policy 3 rd Edition. Edited by John E. Kowka, Jr. and L.J. White, eds., 8-0. New York: Oxford University Press, 999. Heien, D.M. Marku Pricing in a Dynamic Model of the Food Industry. Amer. J of Agr. Econ. 6 (February 980): 0-8. Jeuland, A. and S. ugan. Channel of Distribution Profits When Channel Members Form Conjectures: Note. Marketing Science. 7 (Sring 988): 0 0. Judge, G.G., H.R. Carter, W.E. Griffiths, and T.C. Lee. Introduction to the Theory and Practice of Econometric, 3 rd ed. New York: John Wiley & Sons., 988. Kadiyali, V., Vilcassim, N.J., and Chintagunta, P. Emirical Analysis of Cometitive Product Line Pricing Decisions: Lead, Follow, or Move Together. Journal of Business. 69( October 996): Manufacturer Retailer Interactions and Imlications for Channel Power: An Emirical Investigation of Pricing of Analgesics in a Local market. Unublished Manuscrit. Ithaca, N.Y.: Cornell University, Johnson Graduate School of Management, 998. Kinnucan, H.W. and Forker O.D. symmetry in Farm-Retail Price Transmission for Major Dairy Products. Amer. J. of Agr. Econ. 69 (May 987): McCorriston, S., C.W. Morgan, and A.J. Rayner. Processing Technology, Market Power and Price Transmission. J. of Agr. Econ. 49 (May 998): Nevo, A. Demand for Ready to Eat Cereal and its Imlications for Price Cometition, Merger Analysis and Valuation of New Brands. Ph.D. Dissertation. Deartment of Economics, Harvard University. May 997. Rogers, R.T. Structural Change in US Food Manufacturing, 958 to 997. Agribusiness., forthcoming. Schelling, T.C. The Strategy of Conflict. Cambridge, M.A.: Harvard University Press., 960. Slade, M.E., Product Rivalry with Multile Strategic Weaons: An Analysis of Price and Advertising Cometition. Journal of Economics and Management Strategy. 4 (Fall 995): Turnovosky, S.J. The Behavior of a Cometitive Firm with Uncertainty in Factor Market. New Zealand Econ. Paers 3 (969):5-58. Vuong, Q.H. Likelihood Ratio Tests for Model Selection and Non-nested Hyotheses. Econometrica. 57(March 989):

22 Ashenfelter et al. ( 7) also make this oint This fact requires us to comute data values for other variables by using a weighted average of the calendar month data. 3 In earlier work we used the cooerative ay rice which is the Agrimark full service rice charged to handlers and includes handling charges for quality control and balancing lus any negotiated remiums. However less than half of the fluid milk sold in the Boston IRI market is sulied by Agrimark. For most of the milk, rocessors erform on farm quality control and balancing services, thus the Class-I fluid milk rice seems to be the most aroriate farm level milk rice. See Dhar and Cotterill (000) for models that used the cooerative ay rice. Results are similar. 4 We also secified a short run dynamic cost model. In it or month lags and a weighted average of months lagged rices of Class-I milk were secified in addition the current rice of milk. The results were not significantly different from the results we resent here with only the current rice of milk in the model. 5 The descritive statistics for all the variables are available from the author. 6 Results without the binary and for the other two games are nearly identical and are available from the authors uon request.