asymmetric price relationships in the u.s. broiler...

Journal of Agricultural and Applied Economics, 28,2(December 1996):279-289O 1996 SouthernAgricultural Economics Association

Asymmetric Price Relationships in the U.S.Broiler Industry

John C. Bernard and Lois Schertz Willett

ABSTRACT

This study presents a testing methodology to analyze potential price asymmetries amongthe farm, wholesale, and retail levels of the U.S. broiler industry. Lag length, direction ofcausality, and asymmetric relationships are empirically determined. Results suggest thatconcentration and power of the integrators in the industry have allowed the wholesaleprice to become the center, causal price in the market. Asymmetric price transmissions,however, are limited. While downward movements in the wholesale price are passed onmore fully to growers than increases in the wholesale price, only consumers in the NorthCentral region of the U.S. sharea larger portion of wholesalers’ price increases thanpricedecreases.

Key Words: asymmetry, broilers, concentration, Granger causality, price transmission.

Price changes between market levels andamong spatially separated markets should besymmetric. Considerable attention, however,has been devoted to determining if price trans-missions act differently dependent on the di-rection of the initiating price change. Of par-amount concern has been whether or notwholesalers of some agricultural commoditieshave the power to asymmetrically influencethe prices on the farm and retail levels. Paststudies addressing this question have focusedon commodities such as fresh vegetables(Ward), dairy products (Kinnucan and Forker),citrus (Pick, Karrenbrock, and Carman), cattle(Bailey and Brorsen), and pork (Boyd andBrorsen; Punyawadee, Boyd, and Faminow),

JohnC. BernardandLois SchertzWillettaregraduatestudentandassociateprofessor,respectively,in theDe-partmentof Agricultural,Resource, and ManagerialEconomics at CornellUniversity.

The authorswould like to thank Robert Kalter andtwo anonymous journal reviewers for their helpful sug-gestions. This research was supported in part by HatchContract NYC-121407.

and have reached varying conclusions. Ourstudy examined price relationships in the U.S.broiler industry using a complete, refined test-ing methodology built on the efforts notedabove.

The methodology is complete in that em-pirical tests are used at every stage of the anal-ysis. The four-step procedure begins with unitroot testing to determine the correct form forthe data, continues with lag length determin-ation and causality tests for flow durationsand direction, and concludes with asymmetrytests. The methodology is refined in its use ofmultiple criteria and testing strategies wherequestions have arisen concerning their appro-priateness in varying situations.

The broiler industry has not been investi-gated previously for asymmetry, despite struc-tural reasons for concern. Especially apparentis the increased wholesale concentration in theindustry. Further, the broiler industry is highlyvertically integrated, giving the wholesale lev-el strong influence throughout the market. Ad-ditionally, the spatial aspect of broiler produc-

280 Journal of Agricultural and Applied Economics, December 1996

tion means regional variations in pricetransmissions may exist.

The three specific objectives of this re-search were to determine if (a) decreases inthe wholesale price of U.S. broilers are passedon to growers more than increases, (b) con-sumer prices respond more to increases in thebroiler wholesale price than to decreases, and(c) these price transmission processes vary byregion at the retail level.

Broiler Industry Structure

The modern broiler industry is marked bymany structural elements of imperfect com-petition: high concentration, vertical integra-tion, advertising, and product differentiation.While vertical integration has long existed,these other conditions have come into promi-nence during the recent past. In particular,through expansions and buyouts over the de-cade examined in our study (1983 through1992), the industry’s four-firm concentrationratio rose from 32.8 to 40.4 (Broiler Zndustry).This concentrated wholesale level consists oflarge, vertically integrated firms called inte-grators. Integrators own hatcheries, feed mills,and processing plants. They typically purchaseeggs from breeders while hatched chicks aresent out to growers under contract.

Approximately 90% of all broilers aregrown under contract. The contractual ar-rangement between grower and integratormakes a grower’s proximity to integrators es-pecially important. One obvious risk to agrower is the possibility of the local integratorclosing down or relocating. The risk from thepoint of view of this study, however, is thatthe integrator will take advantage of growers,offering them prices below competitive levels.

Provision of inputs is split between growerand integrator. The grower provides housingand labor, and pays for water, fuel, and elec-tricity. In addition to the broiler chicks, theintegrators supply the feed, vaccines, and anynecessary medications. When the broilers havematured enough for market, they are returnedto the integrator where they are processed andshipped, primarily to retail outlets.

The U.S. broiler industry is geographically

concentrated into four regions: South, North-east, West, and North Central. It is the south-ern states, particularly Arkansas and Georgia,which dominate the nation in broiler produc-tion. This geographic concentration suggeststhe potential for price transmissions to varyspatially.

Another key development in the broiler in-dustry has been the changing nature of the endproduct. Percentage of sales represented by thetraditional whole, fresh broiler examined heredeclined throughout the sample period. Thebulk of the market now consists of further pro-cessed products, fast food nuggets, and other,harder to quantify forms. Despite this trend,whole broilers still represent a sufficiently sig-nificant portion of the market for examiningthe competitiveness of price transmissionswhile avoiding the complications of calculat-ing margins for extensively processed prod-ucts.

Theoretical Considerations of PriceTransmissions

Early research in perfectly competitive pricetransmission processes suggested that consum-er demand determined retail price, with lowermarket-level prices being determined by sub-traction of processing, transportation, and oth-er costs (Thompson and Lyon). Models basedon this theory considered price spreads to bederived through percentage or absolute pricemarkups, or a combination of the two.

Gardner later constructed a model to in-corporate the concept that supply and market-ing services, as well as demand, were relevantto price spreads. In this framework, markupschange in different ways depending on wheth-er the price change began on the farm supplyor retail demand side of the industry. The sub-stantial demand increases experienced by thebroiler industry during the sample periodcould thus be used in this model to explainany evidence of asymmetry. As Kinnucan andForker suggest, however, growing stocksshould insulate prices from the effects of de-mand shocks—and growing stocks were evi-dent in the broiler industry over the sampleperiod.

Bernard and Willett: Broiler Price Relationships 281

Heien’s margin work was impoflant in an-alyzing the direction of price transmissions incompetitive markets. He developed a dynamicmarkup model where, at each successive mar-ket level from farm to wholesale to retail, theprice was based on the preceding level’s priceplus some markup. Thus, price changes flowedfrom farm to wholesale to retail. Heien’s mod-el depends on the assumptions of constant re-turns to scale and fixed proportions, conditionswhich Wohlgenant’s work has suggested existin the U.S. broiler industry.

Wohlgenant’s sample period, however, pre-dates that of this study and the industry’s in-creased concentration. 1 Ward reported thatsuch concentration may yield informationaladvantages that could lead prices to emanateoutward from the wholesale market level tofarm and retail levels. For instance, a concen-trated wholesale level, as in the broiler indus-try, may be the center of price formation inthe market. Supply and demand factors in sucha model have less effect on margins and flowsthan in the competitive models.

In connecting theory and empiricism, this

research on the broiler industry empirically

examines price transmissions in a manner the-oretically consistent with Ward. Characteris-tics of industry structure were consideredmore important than supply and demand con-cerns. Direction of price flows was tested,rather than assumed on the basis of any of thediscussed models.

Methodology

Most asymmetry studies stem from Wolf-fram’s method for splitting irreversible vari-ables for regression modeling. While Wolf-fram’s method, and Houck’s subsequentrefinements, were useful, they were appropri-ate only within a static framework. Ward cap-tured the dynamics of price transmissions byadding the influences of past prices, while

1Wohlgenant’s sample period encompassed 1955through 1983. For the majority of this time frame, itis likely the broiler industry would be analyzed bestwith a competitive model. The rationate behind thepresent study is that important structural changes haveoccurred since the early 1980s.

Boyd and Brorsen expanded asymmetry test-ing to include testing speed of price adjust-ments alongside overall effects, The four-stepprocedure utilized here closely matches thefurther improvements of Bailey and Brorsen.

Since autocorrelation has been a commonproblem in asymmetry studies (Kinnucan andForker; Pick, Karrenbrock, and Carman; Boydand Brorsen), the initial step was to determinethe correct form for the data by conductingunit root tests. Many economic time series arenonstationary, typically exhibiting a trendaway from an initial value. To remove trend,it must first be determined whether a model isa trend stationary process (TSP) or a differ-ence stationary process (DSP). A TSP modelshould be detrended with the inclusion of atrend variable, while a DSP should be detrend-ed by using first differences (Nelson andKang). A model is identified as a DSP if it hasa unit root.

Unit root testing for the price variables inthe data set was performed using the aug-mented Dickey-Fuller test on a one-period au-toregressive model of the form:

Au, = (30+ (31U,.1+ ~zAu,.,

i- ~3TREND + g,,

where Au( is the first difference of the priceseries at a time i, and a trend and constant termhave been included. MacKinnon critical t-val-ues were used to test the coefficient on Mt.1,with the null hypothesis of a unit root rejectedif the coefficient was determined to be signif-icantly different from zero. From our reviewof previous studies, only Bailey and Brorsenhave tested for unit roots.

The second step was to determine thelengths of time between influences and ad-justments among the prices. As bias from anincorrect model will cause parameters to varyfrom their true values, accuracy in lag lengthis crucial for both causality and asymmetrytesting. Some studies (Boyd and Brorsen; Pun-yawadee, Boyd, and Faminow) used past stud-ies for lag determination, while Kinnucan ‘andForker, and Pick, Karrenbrock, and Carmanadded/subtracted lags until significance

282 JoumalofAgricultural and Applied Economics, December 1996

changed. The present study, like that of Baileyand Brorsen, used selection criteria.

Testing of lag lengths was performed intwo stages, following Hsiao’s testing meth-odology for a bivariate autoregressive model.While Hsiao used only Akaike’s final predic-tion error (FPE) criterion, citing its ability tobalance the risk of under- or overestimatingthe correct lag, recent research has used mul-tiple criteria. Since the FPE, as noted byHolmes and Hutton, has a tendency to overfitlags, Schwarz’ Bayesian information criterion(BIC) was also used in testing. BIC offers thebenefit of asymptotically eliminating the pos-sibility of overestimation, although with an in-creased chance of underestimation. In contrast,Bailey and Brorsen used only Akaike’s infor-mation criterion.2

The first stage of Hsiao’s procedure in-volves discovery of the lag length of a uni-variate function, while the second stage con-siders the bivariate case. For a variable, Y,thought to be explainable by itself, and anoth-er variable, X, the bivariate autoregressivemodel can be expressed as follows:

Y, = ~ ~,Y,_( + j (3,+m+lX,_,+ e,,,=, ,=0

where m and n are to be estimated.In both stages, lags were added sequential-

ly. No lag lengths were bypassed or tested forremoval once the lag length increased. Thefirst stage of the analysis consisted of univari-ate testing of each Y to determine the order ofeach of the variables lagged on itself. In thesecond stage, each of the variables was in turnlagged on all other variables with which it hada theoretical relationship. The dependent vari-able in each case was itself lagged to the ex-tent discovered in stage one in each regres-sion.

Discovered lag lengths were then used inthe equations for causality testing. To deter-mine causality, a test proposed by Geweke,Meese, and Dent, based on Sims’ test, was

used.3 To determine if an affected factor (AF)has been acted upon by a causal factor (CF),past values of the causal factor and past, pres-ent, and future values of the affected factorwere used to explain the present value of thecausal factor. This specification was testedagainst the same specification with the excep-tion of the future values of the affected factor.The unrestricted (1) and the restricted (2)equations were:

(1)

and

(2)

LCF

CF, = ~, + ~ (3.CF,_.“=[

IAF

+ ~,=~up PLCF+LAF+.t+ IAFt-~

+ PLcF+2xb4F+2 TREND, + U,

“=,

L4F

+ ~ PLCF+ttt+ IAFt-tt#,.=0

+ ~LCP+LAF+2 TREND, i- V,,

where, in (1), the lead effect is represented bym = –LAF through – 1 on the summation of

the affected factor. The lag length and leadlength on the affected factor were assumed tobe the same (LAF).

Results of causality and lag length testswere used to assess the symmetry of price re-lationships across market levels. The first stepin this process was to transform the variablesinto differences from their initial value. Thiswas done because the hypothesis of revers-ibility depends on the change in the value, notits level. Independent price variables to betested for nonreversibility required furthertransformation. These were segmented by di-viding them into their upward and downwardmovements, such that the new variables werethe cumulative sums over time of the increasesand decreases, respectively.

2We also used this criterion, but the results wereidenticat to Akaike’s FPE, and so were not includedseparately.

~Causality testing was also conducted using Gran-ger’s original method. However, these results wereidentical, and so were not included.


Nonsegmented gas prices were then includ-ed to represent transportation costs to the dif-ferent regions. Using UPCF* and D WCF” torepresent the upward and downward move-ments of the transformed causal price, AF* torepresent the transformed affected price, andGAY for the transformed gasoline price, theunrestricted equation for testing for asymmet-ric price transmissions can be written:

LCF

(3) AF; = ~OTREND, i- ~ ~,,,tUPCF;_,,,,=0

I.CF

+ ,~o P2,11DWCRZ

+ (33GAS; + W,.

No constant term is included since it cancelsout when transformed into its difference fromits first value. Lag lengths, LCF, were againas calculated in the lag length determinationtests. Additionally, in testing price links in-volving the farm price, the transformed costsof the two primary inputs were added to theabove equation.

Next, a series of restricted equations wasestimated. The first restricted equation testedthe cumulative effects over all lags, with allup and down variables replaced by the corre-sponding cumulative sums of the two. UsingFULLCF* to represent a combined pricechange, and assuming other appropriate inde-pendent variables, the form of this equationcan be written:

AF; = ~oTREND

LCF

+ ,~o P 1,NJLLCF;L, + h

The restricted and unrestricted equationswere compared statistically to determine if in-creases in the causal price had a different influ-ence than decreases. Tests of significance wereperformed using the F-statistic. The null hy-pothesis was that the coefficients on the upwardand downward price movements were equal, oc

Asymmetry was also investigated on a pe-riod-b y-period basis. For each period, anotherrestricted equation regression was performed,with the upward and downward price move-ments for that lag period replaced by the com-bined price change and the other periods leftsegmented. Using the same notation as in (3),the restricted equations for individual periodsfollowed the form:

A< = ~OTREND + (3,FULLCF;-,

LCF

+ x B3,JDWCH, + %j= 0,]+/

where i was the period being tested, and withany other appropriate nonsegmented indepen-dent variables included.

Each period’s restricted equation was com-pared statistically with the unrestricted equa-tion to determine if price transmissions weresymmetric in that period. This allowed inves-tigation of the speed of adjustment, similar tothe analysis of Bailey and Brorsen and others.The importance of this procedure was to beable to determine if brief asymmetries existbut, as proposed by Pick, Karrenbrock, andCarman, are quickly corrected. The test meth-od was the same as for the cumulative effectcomparison.

Data

Monthly data used cover the period January1983 through December 1992. As the U.S.Department of Agriculture (USDA) changedits reporting methods for the broiler industryin 1983, this was an appropriate starting point.By 1992, the industry’s concentration ratioseemed to have stabilized, marking an appro-priate endpoint for our study.

National wholesale and farm broiler prices,as well as the costs of the two primary feedingredients, the Central Illinois price of No. 2yellow corn and the Decatur price of soybeanmeal (44% solvent), were taken from selectedissues of three USDA publications: PoultryMarket Statistics, Livestock and Poultry: Sit-


uation and Outlook Report, and Livestock andPoultry Update. The wholesale price was a12-city weighted average series calculated bydividing the country into three marketingregions, and then weighing prices in each re-gion by their population.4 The compositeweighted prices consisted of U.S. Grade A andPlant Grade branded and unbranded whole-carcass products and whole birds without gib-lets.

The farm price was calculated as the equiv-alent live weight return to producers. The for-mula is based on the ready-to-cook price witha combination of processing costs and dress-ing percentage subtracted. While such a deri-vation portends possible difficulties with em-pirical analysis, the series closely mirrored aweekly Arkansas farm price collected from asurvey of integrators.

Retail prices for whole, fresh chickenswere collected by the U.S. Bureau of LaborStatistics using four U.S. sampling regions:West, South, North Central, and Northeast.5Additionally, a national price was calculatedas a population weighted average of the re-gional prices.

For transportation costs, a national retailprice of gasoline was collected from NationalPetroleum News: Factbook Issue. This pricewas an average of all types of gasoline avail-able and includes taxes and variations due totypes of service.b We believe this series cap-tures any price relationship differences due tothe varying distances between concentratedbroiler production areas and the spatially dis-tinct retail regions.

Other data variables were investigated, but

were not found to be statistically usable.7 La-bor costs specific enough to represent actualcosts incurred at the various market levels inthe broiler industry, as well as desired mar-keting costs (especially advertising), were notavailable. Data problems thus limited the ex-tent to which potential model componentscould be included. As Ward noted, however,an understanding of pricing relationships canbe determined without complete elaboration ofthe market, and it was believed empirical anal-ysis would not be affected significantly.

Empirical Results

Unit root tests on the data series were evalu-ated first. The retail price in the South rejecteda unit root at the 5% level, while all otherbroiler price variables, except the Northeastretail price, rejected the null hypothesis of aunit root at the 10% level. The price variablesfor corn, soy, and gasoline, however, failed toreject the unit root hypothesis. Based on theseresults and a desire for consistency, equationswere constructed as TSP, using levels and in-cluding a trend term.

Table 1 summarizes the results from the lagdetermination tests. Looking at the FPE re-sults, several relationships indicated the appro-priate lag length was three months or less. Asexpected, the Schwarz criterion suggestedgenerally lower lag lengths more in the rangeof one month, although it reached the identicalconclusion for many relationships.8 Resultsfrom the straightforward maximization of theadjusted R* were included only for comparisonpurposes, but notably often matched the FPEcriterion.

4 The 12 cities were Boston, Chicago, Cincinnati,Cleveland, Denver, Detroit, Los Angeles, New York,Philadelphia, Pittsburgh, St. Louis, and San Francisco.

5These pricedatawere supplied by R. BuzzeU of

the U.S. Bureau of Labor Statistics, whose help wegratefully acknowledge,

6 While broiler transport vehicles run mainly ondiesel fuel, such a distinct series was unavailable.Graphical analyses of leaded, unleaded, and premiumunleaded prices over the past decade showed strongcorrelation, and it was believed that diesel fuel wouldbe similarly related. Therefore, the overall average wasaccepted as a suitable proxy.

7 Other variables considered included the concen-tration ratio, broiler exports, broilers slaughtered, andthe.retail prices of beef and pork, the competing meats.Unfortunately, these variables proved too highly cor-related with trend and with one another to include inthe models (all correlations greater than 0.85). In ad-dition, there were insufficient data to generate a growerinput variable for buildings and equipment.

8To help gauge the appropriateness of the laglengths, geparate tests were performed on a weektydata set. Results suggested lags ranging primarily fromthree to seven weeks. Hence, we conclude the monthlylags were accurate.


Table 1. Lag Determination Test Results

Criteria

Adjusted

R2Dependent Lagged FPE Schwarz

Variable Variable Lag Lag Lag Value

corn corn 3 2 3

Soybean Soybean 1 1 1

Farm Farm 3 1 3

Gas Gas 4 2 4

Wholesale Wholesale 3 1 3

Retail (Nat) Retail (Nat) 2 2 2

Retail (W) Retail (W) 1 1 1

Retail (NE) Retail (NE) 1 1 1

Retail (S) Retail (S) 2 2 2

Retail (NC) Retail (NC) 1 1 1

Farm Corn o 0 0

Farm Soybean o 0 0

Farm Wholesale 1 0 1Farm Retail (Nat) 1 1 1

Farm Retail (W) 3 1 3

Farm Retail (NE) 3 1 3

Farm Retail (S) 1 1 1

Farm Retail (NC) 3 1 3

Wholesale Farm 1 1 1

Wholesale Gas o 0 0

Wholesale Retail (Nat) 1 1 3

Wholesale Retail (W) 2 2 3

Wholesale Retail (NE) 3 1 3

Wholesale Retail (S) 4 4 4

Wholesale Retail (NC) 3 1 3

Retail (Nat) Farm 3 3 3

Retail (W) Farm 4 1 1

Retail (NE) Farm 2 2 2

Retail (S) Farm 3 3 3

Retail (NC) Farm 2 2 2

Retail (Nat) Wholesale 3 3 3

Retail (W) Wholesale 1 0 0

Retail (NE) Wholesale 2 2 2

Retail (S) Wholesale 3 3 3

Retail (NC) Wholesale 2 0 2

Notes:Nat = National, W = West, NE = Northeast, S = South, and NC = North Central.

0.949

0.878

0.686

0.936

0.652

0.899

0.800

0.924

0.834

0.783

0.676

0.683

0.979

0,829

0.747

0.730

0.801

0.768

0.977

0.647

0.794

0.721

0.700

0.766

0.745

0,963

0.847

0.946

0.926

0.866

0.956

0.834

0.943

0.917

0.853

The lag tests between prices and inputssuch as corn, soybeans, and gas indicated lowR2 values and zero lags. While these tests sug-gested the input variables were unnecessary,they were left in the asymmetry model due tothe theoretical rationale for including inputprices. Lag results may be due simply to theaggregated, averaged nature of the data.

Lag results were used to determine causal-

ity and asymmetry.g Causality results, dis-played in table 2, showed wholesale pricescause farm prices and retail prices in allregions, and farm prices cause retail prices inall regions, while the retail prices in each re-

9The resultsdiscussedwere derivedusing the lagresultsfrom the FPE criterion.End resultsfrom theSchwarzcriteriondid not vary.


Table 2. Causality Test Results

Hypothesized

Effect Cause F-Statistic

Farm Corn 0.427

Farm Soybean 0.885

Farm Wholesale 6.268*

Farm Retail (Nat) 0.449

Farm Retail (W) 0.754

Farm Retail (NE) 0.372

Farm Retail (S) 1.240

Farm Retail (NC) 0.126

Wholesale Farm 1.145

Wholesale Gas 0.570

Wholesale Retail (Nat) 0.343

Wholesale Retail (W) 0.446

Wholesale Retail (NE) 0.264

Wholesale Retail (S) 1.016

Wholesale Retail (NC) 0,685

Retail (Nat) Farm 9.804**

Retail (W) Farm 14.590**

Retail (NE) Farm 19.457**

Retail (S) Farm 13,616**

Retail (NC) Farm 21.090**

Retail (Nat) Wholesale 12.913**

Retail (W) Wholesale 16.577**

Retail (NE) Wholesale 22,995**

Retail (S) Wholesale 18.157**

Retail (NC) Wholesale 24.516**

Notes: Single and double asterisks (*) denote rejection ofthe hypothesis of no causality cause to effect at the 0.05and 0.01 levels, respectively. Nat = National, W = West,NE = Northeast, S = South, and NC = North Central.

gion do not cause either wholesale price orfarm price. The oddest of these results, thefarm-to-retail link in the absence of a farm-to-wholesale link, is difficult to interpret theoret-ically. Unless retailers are being cautious ofthe potential market power of the wholesalersand check farm prices to make sure they arenot being exploited, little justification can befound, This result may be a residual effect ofthe causality exerted on both by the wholesaleprice. The wholesale-to-farm and wholesale-to-retail links are consistent with increasedconcentration of the integrators, and conformto Ward’s beliefs. Still, the market power ofthe integrators implied by these results may ormay not be harmful, depending on the resultsof the asymmetry tests.

Autocorrelation proved a serious problem

in conducting the asymmetry tests, with Dur-bin-Watson statistics from initial runs rangingfrom 0.5 to 0.9. Normally, this would suggestthe variables should be first difference, butthis would have conflicted with the results ofthe unit root tests.[o Investigating further, thosevariables found to have a unit root were dif-ference and the regressions rerun; but auto-correlation persisted. Removal of the unit rootvariables did improve the Durbin-Watson sta-tistic, suggesting there may be difficultiesmodeling TSP and DSP variables in the sameequation. In the end, autocorrelation was cor-rected by running the regressions as autore-gressive processes of order one.

The results of the asymmetry tests for thecumulative effect, current month, and onemonth lagged effects can be seen in table 3.The key result was the downward price asym-metry from wholesale to farm; there was a sig-nificant difference in the effect of increasesand decreases at the O.01 level, both cumula-tively and for the lagged month. It appears thatintegrators are not passing on price changes atthe wholesale level to producers symmetrical-ly. This implies wholesalers may be usingmarket power in updating contract prices.

The results from the wholesale to retail lev-els were not consistent over the differentregions, suggesting upward price asymmetryfrom wholesale to retail cumulative in theNorth Central region, cumulatively and at onemonth lagged on the national level, and at onemonth lagged in the South region, all at the0.05 level. Curiously, despite the differentconclusion from the cumulative effects, thevalues of the relationships in the South andthe North Central regions were very close. Inthe Northeast and West, upward cumulativeprice effects were slightly higher than down-ward effects, but neither was significant. TheWest in particular showed near matching totalscumulatively and at each lag. These resultsadd evidence to the theory of geographicalmarket power by the integrators in the South,but suggest their power may be balanced out

10Unit root tests were run again, this time without

a trend term, but with the same end results.


Table 3. Asymmetry Test Results

Cumulative Effect CurrentMonth Previous MonthPrices In- De- In- De- In- De-

Effect Cause crease crease F crease crease F crease crease F

Farm

Retail (Nat)

Retail (W)

Retail (NE)

Retail (S)

Retail (NC)

Retail (Nat)

Retail (W)

Retail (NE)

Retail (S)

Retail (NC)

Wholesale

Farm

Farm

Farm

Farm

Farm

Wholesale

Wholesale

Wholesale

Wholesale

Wholesale

0.656

1.492

1.181

1.430

1.527

1.594

0.977

0.771

0.886

1,006

1.055

0.744

1.105

0.937

0.923

1.140

1.169

0.837

0.735

0.725

0.888

0.890

6.33**

4.15**

1.43

2.55

4.44**

6.77**

2.82*

0.18

1.18

2.28

3.38*

0.640

0.662

0.840

0.595

0.677

0.631

0.457

0.498

0.323

0.392

0.441

0.652

0.571

0.597

0.436

0.520

0.761

0.324

0.433

0.309

0.313

0.419

0.010.27

0.55

0.31

0.51

0.21

0.97

0.08

0.01

0.21

0.01

0,016

0,676

0.340

0,725

0.860

0.833

0.524

0.280

0.565

0.676

0.635

0.092

0.324

0.339

0.401

0.272

0.301

0.200

0.302

0.258

0.206

0.223

12.0**

3.6

0

1.65,9*

2.65+6*

o

2.66.7*

3.0

Notes: Single and double asterisks (*) denote rejection of the hypothesis that the effect from a price increase equalsthe effect from a price decrease at the 0.05 and 0.01 levels, respectively. Nat = National, W = West, NE = Northeast,S = South, and NC = North Central.

by the power of retailers in more distant mar-kets.

Both the South and North Central regionsshowed significant upward asymmetry fromfarm to retail market levels, at the 0.01 levelcumulatively, and at 0.05 for one monthlagged in the South. Additionally, the nationalretail price also showed upward price asym-metr y cumulatively at the 0.01 level. TheNortheast region had a wide spread betweenupward and downward cumulative effects,which surprisingly failed to be significant.While other results were not significant, inmost cases the F-statistic was higher than ithad been in the wholesale to retail cases. It isunderstandable that the farm to retail resultstend to be more significant than the wholesaleto retail tests. As the farm level was shown tosuffer downward asymmetry, and some retailprices upward, their values should be more op-posed than wholesale and retail. In otherwords, both appear to emanate outward, op-positely, from the wholesale price.

For asymmetry tests conducted on two andthree months’ previous effects, there was anoccasional appearance of negative coefficients,results contrary to our a priori hypotheses.Possible explanations for these include incor-rect lag specification, poor model specifica-tion, multicollinearity, or insignificant effects.

The problem was most pronounced at later lagperiods, suggesting some lag lengths may havebeen overestimated.

Also not reported in table 3 was the sig-nificance of the other inputs. Coefficients ongas prices were positive for all the retailregions, but were only significant at the 0.10level in the North Central and the West. Whenincluded nationally, gas price coefficients werenegative with even lower t-statistics. The in-significance of gas prices suggests this vari-able has less relative importance in total costsin large portions of the country than had beenhypothesized.

Corn and soybean prices typically had op-posite signs from one another, and further testsindicated the coefficients were not robust. It islikely that even though these expenses are in-curred by the integrators, they are still farminput costs that enter the system and are ac-counted for prior to the farm and wholesaleprice spread. Hence, corn and soybean priceswere not included in the final test results re-ported in table 3.

As another convenient way to visualize theresults, elasticities of price transmission werecalculated from the asymmetry regressions.Elasticity measures for increases and decreas-es in the causal prices in both the short andlong run are displayed in table 4. For both the


Table4. Elasticities of Price Transmissions

Elasticity When Causal Price Is:

Prices Increasing Decreasing

Effect Cause Short Run Long Run Short Run Long Run

Farm Wholesale 1.072 1.099 1.092 1.247

Retail (Nat) Farru 0.254 0.572 0.219 0.424

Retail (W) Farm 0.308 0.432 0.219 0.343

Retail (NE) Farm 0.203 0.488 0.149 0.315

Retail (S) Farm 0.281 0.633 0.215 0.472

Retail (NC) Farm 0.255 0.644 0.308 0.472

Retail (Nat) Wholesale 0.293 0.627 0.208 0.538

Retail (W) Wholesale 0.305 0.473 0.266 0.451

Retail (NE) Wholesale 0.185 0.507 0.177 0.415

Retail (S) Wholesale 0.273 0.699 0.217 0.617

Retail (NC) Wholesale 0.299 0.715 0.284 0.603

Notes: Nat = National, W = West, NE = Northeast, S = South, and NC = North Central.

short and long run, the decreasing wholesaleprice elasticities with respect to farm price ex-ceeded those for the increasing wholesaleprice. Similarly, with one exception, the in-creasing farm and wholesale price elasticitieswith respect to the retail prices surpassed therespective decreasing price elasticities, In allcases, long-run elasticities appeared substan-tially greater than the short run, supporting thelag length test results that full price adjust-ments require some months.

Conclusions and Implications

The concentration and power of the integratorshave allowed the wholesale price to becomethe center, causal price in the market. The det-rimental effects of this situation, however, ap-pear limited. For broiler growers, there arereasons for concern with downward move-ments in wholesale price passed on more fullyto them than increases in the wholesale price.Consumers, on the other hand, are not suffer-ing from price asymmetries. Only in the NorthCentral region of the nation are wholesalersable to share a larger portion of their priceincreases with consumers than they do theirprice decreases. Consumer broiler prices in theNortheast even appear more affected by fac-tors not determined by this research than bythe power of the integrators.

With the changing nature of the end prod-

uct in the broiler industry, consumers in factmay be more concerned with the competitive-ness of price transmissions for further pro-cessed products. While the whole broiler dataset used here contained both branded and un-branded prices, product differentiation is eas-ier and more persuasive for these products.How much, then, the implications of this studycarry over to such products is an open area forfuture research. The collection of accurate andconsistent price data, especially in light of themultitude of product forms, will be the great-est hindrance to such an effort.

Based on the results, it appears evident thatadditional study should also be conducted onother industries exhibiting increasing concen-tration or other components of imperfect com-petition. The analytical framework identifiedhere is an appropriate method to evaluate theseprice transmission relationships.

References

Akaike, H. “A New Look at the Statistical Identi-fication Model. ” IEEE Trans. Auto. Control19(1974):716–23.

Bailey, D., and B. W. Brorsen. “Price Asymmetryin Spatial Fed Cattle Markets. ” West, J. Agr.

Econ. 14(1989):246-52.

Boyd, M, S., and B. W. Brorsen. “Price Asymmetryin the U.S. Pork Marketing Channel. ” N. Cent.

J. Agr. Econ. 10(1988): 103–09.Broiler Industry [monthly magazine]. Mt. Morris

Bernard and WiUett: Broiler Price Relationships 289

I IL: Watt Publishing Co. Selected issues, 1983–1 92.

Gardner, B .L. “The Farm-Retail Price Spread in a

Competitive Food Industry.” Amer. J. Agr.

Econ. 57(1975):399–409.

Geweke, J., R. Meese, and W. Dent. “ComparingAlternative Tests of Causality in Temporal Sys-

1 terns.” J. Econometrics 21(1983): 161–94.1 Granger, C.W.J. “Investigating Causal Relations by1 Econometric Models and Cross-Spectral Meth-

ods. ” Econometrics 37(1969):424–38.Heien, D.M. “Markup Pricing in a Dynamic Model

of the Food Industry. ” Amer. J. Agr. Econ.

62(1980):10-18.Holmes, J. M., and F!A. Hutton. “An Application of

a New Test of Prima Facie Causality to Live-

stock Markets, ” Can, J. Agr. Econ. 39(1991):

481-91.Houck, J.P, “An Approach to Specifying and Esti-

mating Nonreversible Functions. ” Amer. J. Agr.Ecor’Z. 59(1977):570-72.

Hsiao, C. “Autoregressive Modeling of Canadian

Money and Income Data.” J. Amer. Statis. As-

SOC. 74( 1979):553–60.Kinnucan, H.W., and O.D. Forker. “Asymmetry in

Farm-Retail Price Transmission for Major DairyProducts.” Amer. J. Agr. Econ. 69(1987 ):285–

92.National Petroleum News: Factbook Issue {annual

,I bonus issue of National Petroleum News

monthly magazine]. New York: Hunter Publish-

ing Ltd. Partnership. Selected annual issues,

1984–93.

Nelson, C. R., and H. Kang. “Pitfalls in the Use ofTime as an Explanatory Variable in Regres-

sion.” J. Bus. and Econ. Statis. 2(1984):73–82.

Pick, D. H., J. Karrenbrock, and H.F! Carman.“Price Asymmetry and Marketing Margin Be-havior: Arizona Citrus.” Agribusiness 6(1990):

75–84.

Punyawadee, V., M.S. Boyd, and M.D. Faminow.

“Testing for Asymmetric Pricing in the AlbertaPork Market. ” Can. J. Agr. Econ. 39(1991):

493–501.

Schwarz, G. “Estimating the Dimension of a Mod-el.” Ann. Statis. 6(1978):461-64.

Sims, C.A. “Money, Income, and Causality. ”

Amer, Econ. Rev. 62( 1972):540–52.Thompson, G. D., and C.D. Lyon. “Marketing Or-

der Impacts on Farm-Retail Price Spreads. ”Amer. J. Agr. Econ. 7 1(1989):647–60.

U.S. Department of Agriculture, Agricultural Mar-keting Service. Poultry Market Statistics.

USDA/AMS, Washington DC, Various annualissues, 1983–92.

U.S. Department of Agriculture, Economic Re-search Service. Livestock and Poultry: Situation

and Outlook Report. USDA/ERS, WashingtonDC. Selected issues, 1983–92.

—. Livestock and Poultry Update. USDAIERS, Washington DC. Selected issues, 1983-

92.

Ward, R. W. “Asymmetry in Retail, Wholesale, andShipping Point Pricing for Fresh Vegetables. ”Amer. J. Agr. Econ. 62(1982):205–12.

Wohlgenant, M.K. “Demand for Farm Output in a

Complete System of Demand Functions. ”

Amer. J. Agr. Econ. 71(1989):241-52,Wolffram, R. “Positivistic Measures of Aggregate

Supply Elasticities: Some New Approaches—Some Critical Notes. ” Amer. J. Agr. Econ.53(1971):356-59,

asymmetric price relationships in the u.s. broiler...

Documents