Share, price and category expenditure—geographic market effects and private labels

MANAGERIAL AND DECISION ECONOMICS

Manage. Decis. Econ. 20: 175–187 (1999)

Share, Price and CategoryExpenditure—Geographic Market Effects

and Private LabelsWilliam P. Putsis Jr.a,* and Ronald W. Cotterillb,1

a London Business School, Regent’s Park, London, UKb Food Marketing Policy Center, Department of Agricultural and Resource Economics,

The Uni6ersity of Connecticut, Storrs, CT, USA

Focusing on the interaction between national brands and private labels, this paper has twomain empirical contributions: (i) a simultaneous system of demand (share), price andexpenditure equations is estimated, and (ii) differences in the structure of the localgeographic market are incorporated into the analysis. The former represents an importantstep in understanding the complete nature of private label and national brand interaction,while the latter is important for understanding the impact of the local retail environment onmarket behaviour. IRI scanner data from 1991 and 1992 are used to estimate a five-equationsystem across 135 food product categories and 59 geographic markets. The results suggestthat concentration at both the manufacturer and retailer level can significantly affect privatelabel and national brand price. However, while increased retailer concentration is associatedwith higher national brand and private label prices, higher manufacturer concentration isassociated with higher national brand but lower private label prices. Increases in nationalbrand advertising has the effect of raising national brand price and share, but loweringprivate label price and share. This is consistent with previous research and suggests thatadvertising and local market conditions play a significant role in the ability of nationalbrands to price at a premium over private labels. Finally, marketing decision variables, suchas display activity and private label distribution, can have an important impact on totalcategory expenditure. Copyright © 1999 John Wiley & Sons, Ltd.

INTRODUCTION

In the US retail environment, private label or‘store’ brands compete vigorously with nationalbrands. For example, private label brand unitmarket share in supermarkets reached an all-timehigh of 20.8% in the third quarter of 1997 accord-ing to Information Resources Inc. (IRI). Notsatisfied, private label manufacturers havelaunched a frontal assault on successful nationalbrands. For example, private label manufacturerInternational Trading Co. has developed a line ofshelf-stable lunches for retailers based upon

Kraft’s successful ‘Lunchable’ line. Cliffstar Corp.has launched resealable 16-ounce packages offruit juice that compete directly with the offeringfrom Ocean Spray. Even in traditional nationalbrand strongholds, such as in the aerosol cheesespread category, private label manufacturers arenow competing vigorously (Brandweek, 11/24/97).

Although the academic literature has just re-cently begun to address the reasons behind in-creasing private label penetration, the volume ofliterature is growing rapidly. Early work, focusingon the characteristics of consumers who pur-chased private label goods (Myers, 1967; Coe,1971; Bettman, 1974), has given way to researchprimarily concerned with explaining the cross-cat-egory variation in private label market share.Hoch and Banerji (1993), for example, employ

* Correspondence to: London Business School, Sussex Place,Regent’s Park, London NW1 4SA, UK. Tel.: +44 1717066733; fax: +44 171 7241145; e-mail: [email protected] Tel.: +1 860 4862742; fax: +1 860 4864983; e-mail: [email protected]

CCC 0143–6570/99/040175-13$17.50Copyright © 1999 John Wiley & Sons, Ltd.

176 W.P. PUTSIS JR. AND R.W. COTTERILL

warehouse withdrawal and leading national ad-vertisers (LNA) data for 180 product categoriesfor 1987. They find that (i) private labels performbetter in large categories with high margins, (ii)private labels do better when competing againstfewer national manufacturers who spend less onnational advertising and (iii) high quality is moreimportant to private label success than lowerprice. Sethuraman (1992) uses aggregate US scan-ner data (IRI) for 116 categories in a similarattempt to identify the key factors influencingprivate label success across categories, identifying12 marketplace factors as potential determinantsof private label success. These marketplace factorsinclude retail sales volume, average retail price,price differential between the private label and thenational brand, and private label price promotionrelative to that of the leading national brands. Inrelated research, Sethuraman (1995) finds that thenature of the cross-price demand response be-tween private label and national brands variessignificantly by category, with national brandmarket share being an important determinant ofthis relationship.

One aspect that has been mentioned brieflyby various authors (e.g. Sethuraman, 1995;Narasimhan and Wilcox, 1998), yet not addressedin depth, pertains to the importance of the struc-tural characteristics of individual markets and theeffects of retailer concentration. While work inthe industrial organization literature has sug-gested an important link between market concen-tration and price (e.g. Weiss, 1989), this researchhas almost exclusively been conducted at the man-ufacturer level (see Marion, 1979 and Cotterill,1986 for two notable exceptions). In addition,regional differences, differences in market size,and retail store structure can have a profoundimpact on demand response (Hoch et al., 1995).We suggest here that the impact of local marketstructure extends well beyond demand responseand is also likely to influence competitive pricingbehaviour. Further, we expand on the limitedresearch that has addressed demand and priceinteraction for private labels (see Cotterill et al.,1999) by specifying a simultaneous system ofshare, price and expenditure, taking into accountstructural differences across local markets (e.g.differences in local retailer concentration). Thus,there are two main empirical contributions of thisresearch: (i) previous cross-category research onprivate labels is expanded to include a simulta-

neous system of demand, price and expenditureequations, and (ii) differences in the structuralcharacteristics of the local market are incorpo-rated into the analysis. The former represents animportant step in understanding the complete na-ture of private label and national brand inter-action, while the latter is important for under-standing the impact of the local retail envi-ronment on market behaviour.

Conceptually, the nature of manufacturer–retailer competition in any given market willchange both the within channel power and theincentives for stocking and promoting storebrands. For example, in markets where a largechain retailer dominates (versus a more frag-mented competitive retail environment), we wouldexpect that private labels will have a higher share.We specify an empirical model below that allowsus to address these issues empirically. Before weturn to the model, however, we provide anoverview of the data used in the study.

THE DATA

The data used in this study are annual IRI scan-ner data on food products across 59 geographicmarkets and 135 categories from 1991 and 1992.Geographic markets include the 59 major marketsin the US (Albany to Wichita), while categoriesare defined at the subcategory level (examplesinclude grated cheese, frankfurters, cake mixes,distilled water, flour, etc.). Consistent with previ-ous work in the private label area (e.g. Slade,1995; Putsis and Dhar, 1998; Cotterill et al.,1999), composite national brand and private labelvariables were created for the 135 product cate-gories and 59 markets.1

For each market and category, standard IRImeasures are available (an exact variable listingwill be presented in the empirical section). Forexample, we have detailed measures of featureand display activity, price promotions, as well asunit and volume sales for each of the 59 geo-graphic markets and 135 food categories. Wehave also developed measures for the structuralcharacteristics of the market—the manufacturerfour-firm concentration ratio, and the number ofbrands present in the market. These data havebeen combined with independent data from Pro-gressi6e Grocer on each market’s population, in-come distribution and local retail composition,

Copyright © 1999 John Wiley & Sons, Ltd. Manage. Decis. Econ. 20: 175–187 (1999)

177GEOGRAPHIC MARKET EFFECTS

such as the four-firm local retailer concentrationratio and the supermarket-to-convenience storesales ratio. Finally, we have collected LNA datafor all the categories used in the study, providingus with detailed information on national brandadvertising spending by category.2 These com-bined sources provide us with a comprehensive setof variables for each category and each geo-graphic market.

THEORETICAL FOUNDATIONS,EMPIRICAL MODEL AND ESTIMATION

In this section, we build on previous research todevelop an empirical model that simultaneouslyaddresses share, price and total category expendi-ture across multiple categories, incorporating dif-ferences in the structure of the local geographicmarket. To begin, we note that the ability of anyfirm to raise price above marginal cost depends inpart upon own and cross-price demand elasticitiesand the competitive response of rival firms. Inaddition, changes in price and other marketinginstruments (e.g. promotion) may affect the levelof expenditure for all brands in the category(Ailawadi and Neslin, 1998; Putsis and Dhar,1998). This suggests that demand, competitiveprice reactions and total category expenditure areall likely to be jointly endogenous in any econo-metric specification.

We begin by concentrating on the demand-sidespecification. We employ a linear approximatealmost ideal demand (LA/AIDS) model (Deatonand Muellbauer, 1980), which can be stated asfollows:3

MSijk=a10+a11 ln Pij

k+a12 ln Pijl

+a13 ln(EXPENDij/P)+a14dij,

MSijl =a20+a21 ln Pij

k+a22 ln Pijl

+a23 ln(EXPENDij/P)+a24dij, (1)

where ln P is Stone’s linear approximation priceindex, (MSij

k ln Pijk)+ (MSij

l ln Pijl ).

In Equation (1), Pijk (Pij

l ) denotes the pricecharged by the national brand (private label) inthe ith category (i=1, . . . , 135) and the jth geo-graphic market ( j=1, . . . , 59). Similarly, MSij

k

(MSijl ) denotes the dollar market share of the

national brand (private label) in category i andmarket j. In addition, the set of market level

demand shift variables is denoted by dij. Note thatin the LA/AIDS framework, total category expen-diture, denoted by EXPENDij, is deflated byStone’s price index P. Thus (EXPENDij/P) de-notes a deflated (real) measure of per capita ex-penditures with its coefficients (a13 and a23,respectively) providing an estimate of the impactof changes in total category expenditure on de-mand. Contrasting with standard market sharedefinitions used in the IO literature (see, e.g.Weiss, 1989), the LA/AIDS functional form isderived from the consumer’s cost function and,consequently, MSij

k and MSijl are expressed as

share of total dollar expenditure. From the basicformulation in (1), the usual demand restrictionsof symmetry, homogeneity and adding up can beimposed. Further, all (quantity) demand elastic-ities can be recovered from the demand specifica-tion (see Green and Alston, 1990 for additionaldetails).

In the present setting, the use of the LA/AIDSframework is particularly attractive for a numberof reasons. First, the limited work addressing theinteraction between private labels and nationalbrands to date has universally employed a linearfunctional form (e.g. Raju et al., 1995; Kadiyali etal., 1998b; Putsis and Dhar, 1998). However, thelinear form is quite restrictive. For example, whileLee and Staelin (1997) demonstrate that the typeof vertical strategic interaction present depends, inpart, upon the convexity of the demand curve,Genesove and Mullin (1998) note that New Em-pirical IO studies ‘typically impose strong func-tional form [linearity] assumptions on demand,which . . . imply even stronger restrictions on therelationship between price and marginal cost.’Second, previous research has demonstrated thatLA/AIDS demands, combined with the associated(log–log) price reactions, fit well in cross-sectionalapplications (Cotterill et al., 1999). Third, sincethe LA/AIDS model is PIGLOG (Price Indepen-dent Generalized LOGarithmic) in form, it doesnot suffer from linear aggregation bias when ap-plied to first-differenced market level scanner dataof the type used here (Cotterill et al., 1999).Finally, the assumptions necessary to derive pricereaction equations consistent with the LA/AIDSfunctional form are supported by previous re-search (Kadiyali et al., 1998a; Cotterill and Putsis,1998).4

We combine this demand-side model with theassociated price reaction functions. The price



reaction equations consistent with LA/AIDS de-mands are log–log in form (see Cotterill et al.,1999; Cotterill and Putsis, 1999 for additionaldetail on the specification of the price reactionequations). Following Cotterill et al. (1999), wenote that a firm’s ability to raise price 6is-a-6is itsrivals will depend upon both the demand andcompetitive environment. Thus, in addition toprice and own cost, we specify each price reactionequation to be a function of (i) the competitiveenvironment, captured here by structural mea-sures, such as concentration, and denoted by fij ;(ii) factors (denoted by hij), such as nationalbrand advertising and the availability of privatelabel products that impact the ability to raiseprice over marginal cost via brand development(i.e. via building brand loyalty), and (iii) demandshift variables dij. Defining Cij

k and Cijl to repre-

sent supply-side cost shift variables for nationalbrands and private labels, respectively, this im-plies the following specification:

ln Pijk=b10+b11 ln Pij

l +b12 ln Cijk+b12 ln fij

+b14 ln hij+b15dij,

ln Pijl =b20+b21 ln Pij

k+b22 ln Clij+b22 ln fij

+b24 ln hij+b25dij. (2)

Next, we note that the demand specification inEquation (1) assumes that the demand for thefood items considered here are weakly separablefrom all other goods. However, relative pricechanges across separable groups may lead to areallocation of expenditures among the groups.Thus, a price change for a national brand in theready-to-eat (RTE) cereal category, for example,may not only elicit an own-price demand responseand a competitive price reaction, but may alsochange the expenditure level for the entire RTEcategory. Further, the net effect of any marketinginstrument (price, promotion, etc.) on expendituremay be very different for private labels versusnational brands. Neither previous research ad-dressing the interaction between private labels andnational brands (e.g. Putsis and Dhar, 1998;Kadiyali et al., 1998a) nor research incorporatingprice reactions in a LA/AIDS framework (Cotter-ill et al., 1999) has addressed the potential endo-geneity of the expenditure effects associated withchanges in price and promotion. Consequently,we do so here by estimating demand (share), priceand expenditure simultaneously in the empirical

analysis below. Choosing a log form to be consis-tent with the price reaction specification, wespecify the expenditure equation as follows:5

ln EXPENDij=g0+g1 ln Pijk+g2 ln Pij

l +g3 ln hij

+g4 ln dij. (3)

The set of equations represented by (1)–(3)constitute the five-equation simultaneous systemto be estimated. In this system, share, price andtotal category expenditure (EXPENDij) will bespecified as jointly endogenous. Although themodel has five equations, one of the demandequations is redundant for estimation purposes.Since the market shares of national brands andprivate labels sum to one, any loss of brandedshare due to changes in any variable, e.g. privatelabel price, must go to private label share. Thisgeneral adding-up property of a demand systemmeans that we can recover the estimated coeffi-cients and standard errors (t-ratios) for thedropped equation. We drop the private label de-mand equation and estimate the remaining fourequations using iterated three-stage least squares(3SLS).

Finally, note that we include variables captur-ing advertising by national brands (LNA data)and a series of trade promotion variables as de-mand shift variables. The promotion variables areshort term percent price reduction, percent ofvolume sold on display, and percent of volumesold with a local newspaper feature advertisement.While one could model these as additional strate-gic variables to create a multi-dimensional game,this would generate six more reaction equationsand prevent estimation of the system due to insuf-ficient exogenous cost shift variables in thoseequations to identify them. Consequently, wespecify these as endogenous strategic factors thatfirms use to determine price levels and/or shiftdemand.6

EMPIRICAL ANALYSIS

In estimating cross-category price equations, it isimportant to note that cross-category analysisprecludes the use of price levels since price in eachcategory is measured in a variety of different andnoncomparable units. Following Kelton andWeiss (1989), we estimated the first differenceform of our model. In the following sections, all



reported estimates use the annual differencerather than the level of the variable for 1991–1992. For example, DBRSHARE (national brandshare) is 1992 BRSHARE minus 1991 BRSHAREand DBRPRICE (national brand price) is the1992 ln BRPRICE minus 1991 ln BRPRICE (theD symbol will be used throughout to denote thechange from 1991 to 1992). Changes in the natu-ral logarithm of price from 1991 to 1992 arepercent changes, which can be analysed acrosscategories. Estimating a first difference model isalso attractive because it controls for first-orderfixed effects due to excluded local market andcategory variables in level regressions.7 Further,to the extent that private label quality is constantfrom 1991 to 1992, estimating a first differencemodel eliminates the need for the inclusion of a

category private label quality measure—an as-sumed constant level of quality drops out of theanalysis when we difference. This is particularlyimportant since quality measurement is sucha difficult task (Hoch and Banerji, 1993;Narasimhan and Wilcox, 1998).

We restricted this analysis to include only thosemarkets and categories for which private labelproducts have been introduced in the market by1991. This left a final sample of a balanced panelof 7823 observations varying both at the categoryand at the local market level. Based upon thediscussion above, the following system of equa-tions was estimated via iterated 3SLS in Limdepv. 6.0 (equation errors v1, . . . , v4 are assumed tobe contemporaneously correlated across equa-tions, but temporally uncorrelated):

DBRSHARE=a0+a1DBRPRICE+a2DPLPRICE+a3DEXPEND+a4DBRFEATURE

+a5DBRDISPLAY+a6DBRPRICEREDN+a7DADVERT+a8DPLDISTN

+a8DAGE+a110DHISPANIC+v1,

DBRPRICE=b10+b11DPLPRICE+b12DBRVPERU+b13DBRSHARE+b14DMCR4

+b15DNBRANDS+b16DSRATIO+b17DGROCCR4+b18DPLDISTN

+b19DADVERT+b210DPLPRICEREDN+v2,

DPLPRICE=b20+b21DBRPRICE+b22DPLVPERU+b23DBRSHARE+b24DMCR4

+b25DNBRANDS+b26DSRATIO+b27DGROCCR4+b28DPLDISTN

+b29DADVERT+b210DBRPRICEREDN+v3,

DEXPEND=g0+g1DBRPRICE+g2DPLPRICE+g3DPLDISTN+g4DADVERT

+g5DBRPRICEREDN+g6DPLPRICEREDN+g7DBRFEATURE

+g8DBRDISPLAY+g9DPLFEATURE+g110DPLDISPLAY+g111DPOP

+g112DINCOME+v4. (4)

Variable definitions, based upon standard IRI measures, are provided in detail below:8

Dependent 6ariablesBRSHARE Dollar share of total category expenditure for national brands in the ith

market, jth category.Dollar share of total category expenditure for private labels in the ithPLSHARE

market, jth category.Natural log of the volume-weighted average (nondeal or ‘regular’) price ofBRPRICE

national brands in the ith market, jth category.Natural log of the volume-weighted average (nondeal or ‘regular’) price ofPLPRICE

private label products in the ith market, jth category.Natural log of (EXPENDij/P), where ln P= (MSij

k ln Pijk)+(MSij

l ln Pijl ).EXPEND



Demand shift 6ariables (dij)Percent of branded products sold with feature advertising in the ith market,BRFEATURE

jth category. Feature advertising is defined to be any newspaperadvertising feature or store flyer regardless of whether or not the item wasaccompanied with a shelf price reduction, store coupon or display.

Percent of branded products sold with in-store display and/or point-of-saleBRDISPLAYpromotion in the ith market, jth category. Display is defined as anyoff-shelf display, including within and end-of-aisle displays.

BRPRICEREDN Volume-weighted percent average price reduction, national brands, ithmarket, jth category. This is defined as promotional ‘temporary’ pricereductions of at least 5%.

PLFEATURE Percent of private label products sold with feature advertising in the ithmarket, jth category. Feature advertising is defined to be any newspaperadvertising feature or store flyer regardless of whether or not the item wasaccompanied with a shelf price reduction, store coupon or display.

Percent of private label products sold with in store display and/orPLDISPLAYpoint-of-sale promotion in the ith market, jth category. Display is definedas any off-shelf display, including within and end-of-aisle displays.

Volume-weighted percent average price reduction, private label products, ithPLPRICEREDNmarket, jth category. This is defined as promotional ‘temporary’ pricereductions of at least 5%.

Natural log of the total population in the ith local market.POPNatural log of the average household income in the local market.INCOMENatural log of the average age in the local market.AGEPercent of population in the local market of Hispanic decent.HISPANIC

Structural competiti6e measures (fij)MCR4 Four-firm manufacturer concentration (as a percent of total national brand

share) ratio in the ith market, jth category.Natural log of the total number of brands sold in the ith market, jthNBRANDS

category.SRATIO Percentage of grocery sales in the ith market, jth category made by

supermarkets.Percentage of all grocery sales by the top four grocery chains in the ithGROCCR4

market, jth category.

‘Brand de6elopment ’ influences (hij)Natural log of the LNA for manufacturers in the jth category. ThisADVERT

represents total advertising expenditures for all national brands.Private label average distribution (percent of the market’s all commodityPLDISTN

volume (ACV) represented by stores offering a private label in thiscategory).

Cost shift 6ariables (Cijk and Cij

l )Natural log of average volume (weighted) per package sold for nationalBRVPERU

brands.Natural log of average volume (weighted) per package sold for private labelPLVPERU

products.



Before proceeding, three additional points re-garding specification and estimation are in order.First, Connor and Peterson (1992) suggest thatthe degree of product differentiation (and, accord-ingly, preference segmentation) is achieved in partthrough advertising. Accordingly, we incorporatenational brand advertising in the demand equa-tion to capture national brand differentiation gen-erated by advertising spend. In addition,advertising can impact price–cost margins di-rectly. Reasons include competitive effects, suchas the creation of entry barriers (hence f inEquation (2)), and the building of brand loyaltythrough product differentiation (hence h in Equa-tion (2)). Thus, we include advertising expenditurein the price equations as well.

Second, we specify the retail grocery four firmconcentration ratio in the price reaction curves (fin Equation (2)) to capture the increased oligopo-listic interdependence in cities where a few super-market chains account for a high percentage oftotal sales. Prior empirical work on the concentra-tion–price relationship in grocery retailing sug-gests that the general level of the mark-up in alocal area is related to local retailer concentration(Marion, 1979). Since markets characterized byhigher retailer concentration are expected toprovide retailers with a greater ability to priceabove marginal cost (i.e. they are likely to havehigher retail margins), both branded and privatelabel prices are hypothesized to be higher in thesemarkets. We also specify manufacturer concentra-tion (MCR4) expressed as a percentage of totalbranded share in the price reaction equations todifferentiate between manufacturer versus retailerstructural effects. In addition, we proxy for manu-facturer costs in the two price equations by in-cluding a measure of package size to capture thehypothesis that smaller package sizes have highercosts per unit. Such a cost proxy has been usedeffectively in other studies (e.g. Cotterill et al.,1999).

Finally, the theory supporting the demand-shiftvariables included has been examined carefully.For example, market population and income arelikely to influence the total expenditure for thecategory, but not necessarily the distribution ofprivate labels and national brand share within thecategory (note that in the LA/AIDS formulation,EXPEND captures the impact of increases incategory expenditure). Alternatively, the age dis-tribution and ethnic make-up of the local market

is likely to influence the relative shares of privatelabels versus national brands, but not necessarilythe total expenditure on the category (Hoch et al.,1995). Further, since there is a great deal ofevidence to suggest that promotions by weakerbrands (e.g. private labels) are ineffective in steal-ing share from stronger (e.g. national) brands, thethree private label promotion variables are notincluded in the national brand share equation.Lastly, since essentially all of the dynamic interac-tion between private labels and national brandsdepends upon the percentage of stores carryingprivate labels (PLDISTN) in the first place, thevariable PLDISTN is also included as a demand-shift variable in the demand specification.

RESULTS

The results are reported in Tables 1–4 (as before,the D prefix on the variables in Tables 1–4 denotethe 1992 value of the variable minus its 1991value). Since traditional R2 measures are notbounded between 0 and 1 in 3SLS, the multiplesquared coefficient of correlation for simulta-neous systems, Rw

2 , of Carter and Nagar (1977)was used.9 The system Rw

2 was 0.855, which im-plies that 85.5% of the system-wide variation inthe dependent variables is explained by the exoge-nous variables in the system. For a cross-categoryanalysis in first difference form, this suggests that,as a whole, the model fit quite well.

In discussing the substantive implications to bedrawn from the empirical analysis, we divide theresults into four categories: (i) Demand ReactionVariables (Equation (1)), (ii) IO and Supply-SideCompetiti6e Effects (Equation (2)), and (iii) Ex-penditure Effects (Equation (3)). We discuss eachin turn.

Demand Reaction Variables

A lowering of national brand price or an increasein the temporary price reductions offered by na-tional brands increases national brand share andlowers private label share. Similarly, nationalbrand share falls as private label price decreases,as expected. Promotion variables in the shareequation were all of the expected sign and statisti-cally significant at a=0.01 or better. Advertisingby national brands (as measured by the LNAdata) increases national brand share, as does local



Table 1. Three-Stage Least Square Results,1992–1991: Dependent Variable isDBRSHAREa

Coefficient StandardVariable P [Z\z ]error

DBRPRICE −0.2755 0.0245 0.00000.4183 0.0336DPLPRICE 0.00000.3724 0.0191 0.0000DEXPEND0.3826 0.0374DBRFEATURE 0.00000.3079 0.0243 0.0000DBRDISPLAY0.0699DBRPRICEREDN 0.0188 0.0002

DADVERT 0.2778 0.0592 0.0000DPLDISTN −0.2623 0.0126 0.0000

0.0448 0.0627DAGE 0.4748DHISPANIC −0.0190 0.0833 0.8197

a Mean=−0.201; S.D.=5.237; model size: observations=7783, residual sum of squares=23.29.

Table 3. Three-Stage Least Square Results,1992–1991: Dependent Variable isDPLPRICEa

P [Z\z ]Variable StandardCoefficienterror

DBRPRICE 0.1331 0.0214 0.0000DBRSHARE −0.1095E−02 0.2393E−03 0.0000DBRPRICEREDN −0.2884E−03 0.2175E−03 0.1848DPLDISTN −0.3107E−03 0.1508E−03 0.0394DADVERT −0.5840E−02 0.3353E−03 0.0000

0.00960.1383E−02−0.3584E−02DMCR4DNBRANDS −0.1123E−02 0.3742E−03 0.0027DSRATIO −0.6142E−03 0.4209E−03 0.1445DGROCCR4 0.3493E−03 0.2273E−04 0.0000DPLVPERU −0.8316 0.0153 0.0000

a Mean=−0.015; S.D.=0.252; Model size: observations=7783, residual sum of squares=93.38.

feature advertising for national brands. Con-versely, national brand advertising lowers privatelabel share, consistent with previous findings ofHoch and Banerji (1993) that private labels have aharder time competing when national brand ad-vertising is higher. The direction and significancelevels of the expenditure effects indicate that na-tional brands are viewed as luxuries and privatelabels as necessities—as expenditures on a cate-gory increase, more goes to national brands thanto private labels. As predicted, higher privatelabel distribution was associated with a lowernational brand share (and a higher private labelshare) and a lower private label price. This isconsistent with both a demand-side (price must belowered to sell more) and a supply-side explana-tion (scale economies in production and distribu-tion). Finally, neither age nor levels of percent

Hispanic in the local market had a discernableeffect on brand share.

IO and Supply-Side Competitive Effects

The results for the share and concentration vari-ables were perhaps the most interesting. A highermanufacturer concentration increases nationalbrand prices, but lowers private label prices. Oneexplanation for this is that while more concen-trated markets facilitate manufacturer coordina-tion (Weiss, 1989) for national brands, a lowerprice may be the only way private labels cancompete in highly concentrated markets. This isconsistent with the finding that private labelprices are lower in categories where there are alarge number of national brands. On the other

Table 4. Three-Stage Least Square Results,1992–1991: Dependent Variable isDEXPENDa

Variable Coefficient Standard P [Z\z ]error

DBRPRICE −0.7310 0.0373 0.0000DPLPRICE −0.1853 0.0296 0.0000DBRPRICEREDN 0.0117 0.9527E−03 0.0000DPLPRICEREDN 0.0105 0.3840E−02 0.0063DBRFEATURE 0.00000.01880.1509DBRDISPLAY 0.1470 0.0120 0.0000DPLFEATURE 0.4645E−03 0.5444E−03 0.9320DPLDISPLAY 0.3963E−02 0.6451E−03 0.0000DPLDISTN 0.0971 0.6093E−02 0.0000DADVERT 0.0885 0.0147 0.0000DPOP 0.3392E−06 0.2973E−06 0.2540DINCOME 0.00000.5737E−04 0.1136E−04

a Mean=0.004; S.D.=0.181; model size: observations=7783, residual sum of squares=185.21.

Table 2. Three-Stage Least Square Results,1992–1991: Dependent Variable isDBRPRICEa

Coefficient P [Z\z ]Variable Standarderror

0.0129 0.0025DPLPRICE 0.0389DBRSHARE 0.7811E−02 0.1903E−02 0.0000

0.10130.1263E−03−0.2070E−03DPLPRICEREDN0.1121E−030.2263E−03 0.0436DPLDISTN0.0025 0.0000DADVERT 0.0186

0.0372DMCR4 0.2297E−02 0.1103E−020.2965E−03 0.0000DNBRANDS 0.2516E−02

0.53420.3024E−030.1880E−03DSRATIODGROCCR4 0.1612E−03 0.00000.2700E−02

0.00000.0161DBRVPERU −0.9131

a Mean=−0.021; S.D.=0.198; model size: observations=7783, residual sum of squares=48.64.



hand, in the price reaction equations, we find thatthe four-firm retailer concentration has a signifi-cant and positive impact on both branded andprivate labels prices—higher local retailer con-centration clearly affords retailers the ability toraise market price for private labels, perhapsthrough building brand loyalty for individualstore brands. This is consistent with prior workon the relationship between concentration andprice in grocery retailing (Marion, 1979; Cotterill,1986), but these results cover a much larger num-ber of categories and markets. It appears asthough retailers are able to obtain—and use—local monopoly power when it is available. Thus,not only does higher manufacturer concentrationenable manufacturers and retailers to raise pricefor national brands (Weiss, 1989), but under the‘right’ retail environment, local retailers may beable to raise prices for all products in the cate-gory. Finally, a higher private label (nationalbrand) share results in a higher private label(national brand) price. This once again suggeststhat retailers have the potential to build brandloyalty for individual store brands—the buildingof brand loyalty appears not to be unique tonational brands.

When viewed in light of these results, the esti-mated price reactions and results for the advertis-ing variables in the price reaction equations paintan interesting picture of the competitive interac-tion between private labels and national brands.Both price reactions are positive, but private la-bels react more strongly to national brand pricechanges (estimated reaction coefficient of 0.133)than do national brands react to private labels(estimated reaction coefficient of 0.039). Nationalbrand advertising increases the ability of nationalbrands to raise price, while a higher level ofnational brand advertising results in lower privatelabel prices. These results are consistent with thefindings of Hoch and Banerji (1993), who foundthat private labels have a difficult time competingin an environment characterized by high nationalbrand advertising. Here, it appears that whennational brand advertising is high, retailers lowerthe price of private label products in order tocompete. Given the demand side results, it is notclear how effective such a strategy is likely to be.This implies that national brand advertising canbe an effective way of differentiating nationalbrands, enabling them to attain a higher pricepremium over private labels.

A higher degree of brand proliferation, as mea-sured by the number of products available(NBRANDS), had the effect of increasing theprice of national brands, but it had the effect ofdecreasing the price of private labels. For nationalbrands, there are at least two available explana-tions for the observed positive relationship be-tween NBRANDS and BRPRICE. First, the entrydeterrence argument set forth by Schmalensee(1978) suggests that a higher number of nationalbrands makes it more difficult for private labels tocompete effectively. Conversely, Putsis (1997) ar-gues that a larger number of brands can alsosuggest the loss of scale economies in production,thereby increasing costs. For private label prod-ucts, in markets characterized by higher nationalbrand advertising, higher manufacturer concentra-tion, a higher degree of proliferation, and lowerretailer concentration, private labels appear tocompete primarily on price.

Expenditure Effects

Since the literature on asymmetric competition(e.g. Blattberg and Wisniewski, 1989; Allenby andRossi, 1991) suggests that private labels have adifficult time stealing share from national brands,it has often been asserted that private label pro-motion is futile. The results for the expenditureequation, however, suggest that most forms ofpromotion can indeed expand the category (theexception is private label feature advertising).Thus, since category expansion can result frompromotion by both weaker and stronger firms,private label promotion does not need to stealshares from national brands in order to be prof-itable. Recent research, conducted on a limitednumber of categories and using micro-level data,has reported similar findings (Ailawadi andNeslin, 1998; Chandon and Wansink, 1999).However, our analysis was conducted across 135categories and 59 geographic geographic markets,suggesting that the ability of promotions to ex-pand the size of the category may be widespreadand not just limited to certain categories. Finally,it is important to note that increased private labeldistribution and advertising were both estimatedto exert a positive and significant impact on cate-gory expenditure. Thus, private label distribution,in-store display and national brand advertising allseem to play important roles in determining cate-gory size.



DISCUSSION, CONCLUSIONS ANDFUTURE RESEARCH

Any theory that attempts to explain the growth ofprivate labels in the US in recent years must bemulti-faceted. The growing body of work address-ing cross-section variations in private label growth(e.g. Hoch and Banerji, 1993; Raju et al., 1995;Narasimhan and Wilcox, 1998) has enriched ourunderstanding of why private labels flourish insome categories and wither in others. Indeed,much of our analysis has been based upon whathas been learned from this research. We haveattempted to address at least three additionalfactors not previously studied. First, we haveattempted to assess the impact of local marketfactors on private label share and pricing. Second,we have attempted to understand the role thatmarket structure variables affect the interactionbetween private labels and national brands at boththe level of the manufacturer and the retailer.Third, we have attempted to explore the role thatprivate labels and promotion plays in expandingtotal category sales.

There are a number of interesting findings. Forexample, previous work on the effects of manu-facturer concentration has been expanded to in-clude the impact of local retailer concentration.We find that industry concentration at both themanufacturer and retailer level can significantlyaffect private label and national brand price.However, while increased retailer concentration isassociated with higher national brand and privatelabel prices, higher manufacturer concentration isassociated with higher national brand, but lowerprivate label price. Further, we estimate that in-creases in national brand advertising has the effectof raising national brand price and share, butresults in lower private label price and share. Thisis consistent with previous research (e.g. Hochand Banerji, 1993), which suggests that advertis-ing and local market conditions play a significantrole in the ability of national brands to price at apremium over private labels. In markets whereprivate labels have a difficult time competing,perhaps due to high national brand manufacturerconcentration, low local retailer concentrationand/or high national brand advertising spend,they lower price in order to compete. Further,certain forms of promotion and other marketingmix decision variables play a role in determiningoverall category expenditure, something relevant

to mangers operating in a category managementenvironment. For example, private label distribu-tion plays a significant role in expanding categorysales.

This paper began by noting that private labelshare hit an all-time high of 20.8% in the thirdquarter of 1997 and suggested that private labelshares had grown in a number of product cate-gories over time. In light of the empirical results,it is interesting to speculate as to the cause of thisgrowth. For example, the categories that showedthe smallest gain (indeed a loss of share in certaincategories) in private label share over the 1988–1992 period are the categories with the largestgrowth in national brand advertising expenditureover the same period. National brand advertisinghas a twofold impact on private label growth (atleast according to our empirical results). First,increased national brand advertising increases na-tional brand share directly (Table 1), but it alsoincreases total category expenditure (Table 2),which in turn increases national brand share rela-tive to private labels. Thus, it appears as thoughnational brand advertising may be able to createsignificant barriers to private label share growthover time. On the other hand, the single largestdeterminant of private label share appears to beprivate label penetration. Over the 1988–1992period, average private label distribution grew byapproximately 1.5%, with categories exhibitingthe greatest penetration growth seeing the greatestshare growth over this 5 year period. Accordingto the empirical results presented earlier (Table 1),such an increase in private label distributionshould have had a significant and positive impacton private label share. Thus, a casual observationof the changes that have taken place across timesuggests that national brands may be able toincrease share through extensive advertising,whereas private labels increase share by increasingdistribution. Future research should address thesedynamics over time more carefully, with particu-lar care given to the impact of changes in privatelabel quality over time.

Previous research has taught us a great dealabout the nature of private label competition, yetthere is much more to be done. There are essen-tially three relevant levels of variation in privatelabel penetration that need to be understood.First, research on cross-category variation is wellunder way. Second, the present research is meantto be a first step in understanding the effects of



local market structural influences. Third, as justdiscussed, research investigating the influences onprivate label growth o6er time remains. This is oneof the most important issues facing researchersinterested in understanding the growth of privatelabels in the US in recent years.

Our research, while advancing related work, isnot without its limitations. For example, we notethat the use of deflated (real) expenditure as thedependent measure in the expenditure equation(see system (4)) is composite in form (i.e. it con-tains elements found on the right-hand-side of theequation). To see this, note that deflated expendi-ture contains private label and national brandprice in the denominator (since they are defini-tional components of P), while private label andnational brand price are also right-hand-side en-dogenous variables in the expenditure equation.While this is not uncommon in a variety of set-tings in the marketing literature (e.g. market shareand ROI analysis), and while expenditure func-tions have been analysed frequently in economics(see Deaton and Muellbauer, 1980), it doespresent some difficult econometric issues. Al-though these concerns are alleviated somewhat bythe exact functional form specification employedhere, in order to assess the impact of the use of acomposite dependent variable, we performedsome simple diagnostics (see Farris et al., 1992 fora detailed discussion). These diagnostics suggestedthat the impact of the use of a composite formdependent variable on the empirical results maynot be particularly problematic in our analysis.Nonetheless, it would be useful for future researchto build the expenditure equation from underlyingutility theory, hopefully deriving a functionalform that does not suffer from such a limitation.Finally, we should note that given our empiricalfocus, we did not consider several strategic rea-sons that may also determine interaction in themarketplace between national brand manufactur-ers and retailers. For instance, retailers maychoose to introduce a private label product in acategory to use as a competitive weapon with theother national brand manufacturers in that cate-gory (Narasimhan and Wilcox, 1998). In a similarvein, the criteria for promoting brands may havelittle to do with increasing the category revenuebut instead to use the entire category as a trafficbuilder (i.e. a loss leader) to increase total retailvolume. We encourage future research in these,and other related directions.

Acknowledgements

Support from the USDA CSREES Special Research GrantNo. 95-34178-1315 at the University of Connecticut and theCentre for Marketing at the London Business School is grate-fully acknowledged. The authors would like to thank BruceHardie, Subrata Sen and especially Margaret Slade and ananonymous reviewer for helpful comments on earlier drafts.

NOTES

1. For example, aggregate private label and nationalbrand variables were created for share, regular priceand temporary price reduction. Private label (na-tional brand) share is the sum of all private label(national brand) dollar shares in the ith market, jthcategory. Private label (national brand) price is thevolume-weighted average price of all private labels(national brands) in the ith market, jth category.The two price reduction variables are volume-weighted average percent price reduction for all pri-vate label and branded products, respectively. Notethat we include variables representing both regularprice and percent price reduction during temporaryprice promotions since there is a great deal of evi-dence to suggest that the demand response to tempo-rary price reductions is different than the response topermanent price reductions of equal magnitude(Blattberg and Neslin, 1990). Finally, note that thechoice of variables was influenced by data availabil-ity. For example, no coupon information was avail-able, while average age, income and percentHispanic were the only demographic variablesavailable.

2. We note that essentially all private label advertisingover this time period was through the use of featureadvertising. Thus, the advertising data used in thisstudy, which includes leading national brand adver-tising (through the LNA data) and feature advertis-ing (capturing private label advertising), provides forcomprehensive coverage of advertising for both pri-vate labels and national brands.

3. The LA/AIDS framework is well-established and,consequently, we will not present it in great detailhere. For a discussions of the model as applied toscanner data, see Chalfont (1987), Green and Alston(1990) and Cotterill et al. (1999) for discussions ofelasticity estimation.

4. Specifying LA/AIDS demands and deriving a log–log price reaction equation that can be estimatedusing retail price data uses the assumption thatmanufacturers act as Stackelberg leaders within thechannel and that retailers follow a proportionalmark-up rule when deciding upon retail prices. Cot-terill and Putsis (1998), in a detailed analysis ofvertical conduct for national brands and privatelabels across multiple categories, test these assump-tions and find strong empirical support. Further, ourtreatment of vertical conduct in this regard is consis-tent with other studies addressing retail price con-duct for private label products (Slade 1995; Kadiyaliet al., 1998a; Cotterill et al., 1999).



5. In the empirical analysis that follows, we will usedeflated (real) expenditures (i.e. actual expendituresdeflated by Stone’s price index) as the dependentmeasure in specifying Equation (3) for estimation.This presents some econometric concerns, which arediscussed in the Discussion, Conclusions and FutureResearch section.

6. We address the endogeneity of the trade promotionvariables through the use of instrumental variables.The principle is similar to the approach taken byBerry et al. (1995). Specifically, each promotionalvehicle for market i, category j, is expressed as afunction of the promotional activity in each of theother 58 markets, using the fitted value as the instru-ment. Note that, in order for this approach toeliminate the endogeneity bias, the equation errorsfor each promotion instrument have to be indepen-dent. This implies that display and feature decisions,for example, are made on a market by market (orchain by chain) basis.

7. Hausman and Taylor (1981) argue that excludedlocal market variables in panel data of this type canbias estimation results for level regressions. Theyshow that this can be avoided by specifying a set ofcity binary variables. These drop out of the modelwhen one takes the first difference. This is also truefor specifying a set of category binary variables inlevel regressions to control for excluded variables inindividual categories.

8. In order to be consistent with previous researchemploying scanner data, standard IRI measureswere used wherever appropriate. IRI relies on con-sumer panel surveys, in-store visits and individualstore-level scanner data to compile the measuresused here. For additional detail, see the IRI Market-ing Factbook, which is published annually. All vol-ume measures/weights used are all commodityvolume (ACV) measures as defined by IRI.

9. Rw2 has a usual R2 interpretation. Specifically, it

measures the percent of system-wide variation in theendogenous variables explained by all independentvariables in the system. It is bounded by 0 and 1.However, collinearity may inflate the estimates ofRw

2 and, accordingly, this statistic should be inter-preted with caution (see Berndt, 1991, p. 468).

REFERENCES

K. Ailawadi and S.A. Neslin (1998). The effect ofpromotion on consumption: buying more and con-suming it faster. Journal of Marketing Research, 35,390–398.

G.M. Allenby and P.E. Rossi (1991). Quality percep-tions and asymmetric switching between brands.Marketing Science, 10, 185–204.

E.R. Berndt (1991). The Practice of Econometrics, NewYork: Addison-Wesley.

S. Berry, J. Levinsohn and A. Pakes (1995). Automo-bile prices in market equilibrium. Econometrica, 63,841–890.

J.R. Bettman (1974). Relationship of information pro-cessing attitude structures to private brand purchas-ing behavior. Journal of Applied Psychology, 59,79–83.

R.C. Blattberg and K.J. Wisniewski (1989). Price in-duced patterns of competition. Marketing Science, 8,291–309.

R.C. Blattberg and S.A. Neslin (1990). Sales Promo-tion: Concepts, Methods and Strategies, EnglewoodCliffs, NJ: Prentice-Hall.

R.A.L. Carter and A.L. Nagar (1977). Coefficients ofcorrelation for simultaneous equation systems. Jour-nal of Econometrics, 6, 39–50.

J. Chalfont (1987). A globally flexible, almost idealdemand system. Journal of Business and EconomicStatistics, 5, 233–242.

P. Chandon and B. Wansink (1999). When and whydoes consumer stockpiling accelerate consumptionvolume? London Business School, Unpublishedmanuscript.

B.D. Coe (1971). Private versus national preferenceamong lower and middle income consumers. Journalof Retailing, 4, 61–72.

J.M. Connor and E.B. Peterson (1992). Market struc-ture determinants of national brand–private labelprice differences of manufactured food products. TheJournal of Industrial Economics, 40, 157–172.

R.W. Cotterill (1986). Market power in the retail foodindustry: evidence from Vermont. Re6iew of Econom-ics and Statistics, 68, 379–386.

R.W. Cotterill and W.P. Putsis Jr. (1998). Testing thetheory: vertical strategic interaction and demandfunctional form, University of Connecticut, FoodMarketing Policy Center, Research Report No. 40.

R.W. Cotterill and W.P. Putsis Jr. (1999). Market shareand price setting behavior for national brands andprivate labels. Re6iew of Industrial Organization,forthcoming.

R.W. Cotterill, W.P. Putsis Jr. and R. Dhar (1999).Assessing the competitive interaction between privatelabels and national brands. Journal of Business,forthcoming.

A. Deaton and J. Muellbauer (1980). An almost idealdemand system. American Economic Re6iew, 70,312–326.

P.W. Farris, M.E. Parry and K.L. Ailawadi (1992).Structural analysis of models with composite depen-dent variables. Marketing Science, 11, 76–94.

D. Genesove and W. Mullin (1998). Testing staticoligopoly models: conduct and cost in the sugarindustry, 1890–1914. The Rand Journal of Econom-ics, 29, 355–377.

R. Green and J.M. Alston (1990). Elasticities in AIDSmodels. American Journal of Agricultural Economics,May, 442–445.

J.A. Hausman and W.E. Taylor (1981). Panel data andunobservable individual effects. Econometrica, 49,1377–1398.

S.J. Hoch and S. Banerji (1993). When do private labelssucceed? Sloan Management Re6iew, Summer, 57–67.

S.J. Hoch, B.-D. Kim, A.L. Montgomery and P.E.Rossi (1995). Determinants of store-level price elas-ticity. Journal of Marketing Research, 22, 17–29.



V. Kadiyali, N.J. Vilcassim and P. Chintagunta(1998a). Product line extensions and competitivemarket interactions: an empirical analysis. Journalof Econometrics, 89(1/2), 339–363.

V. Kadiyali, P. Chintagunta and N. Vilcassim(1998b). Manufacturer–retailer channel interac-tions and implications for channel power: an empir-ical investigation of pricing of analgesics in a localmarket. Cornell University, unpublishedmanuscript.

C. Kelton and L. Weiss (1989). Change in concentra-tion, change in cost, change in demand, and changein price. In Concentration and Price (edited by L.Weiss), Cambridge: MIT Press.

E. Lee and R. Staelin (1997). Vertical strategic inter-action: implications for channel pricing strategy.Marketing Science, 16, 185–207.

B. Marion (1979). The Food Retailing Industry, NewYork: Praeger.

J.G. Myers (1967). Determinants of private brand at-titude. Journal of Marketing Research, 4, 73–81.

C. Narasimhan and R.T. Wilcox (1998). Private la-bels and the channel relationship: a cross-categoryanalysis. Journal of Business, 7, 573–600.

W.P. Putsis Jr. (1997). An empirical study of the ef-fect of brand proliferation on private label–national brand pricing behavior. Re6iew of Indus-trial Organization, 12, 355–371.

W.P. Putsis Jr. and R. Dhar (1998). The many facesof competition. Marketing Letters, 9, 269–284.

J.S. Raju, R. Sethuraman and S.K. Dhar (1995). In-troduction and performance of store brands. Man-agement Science, 41, 957–978.

R. Schmalensee (1978). Entry deterrence in the ready-to-eat breakfast cereal industry. Bell Journal ofEconomics, 9, 305–327.

R. Sethuraman (1992). The effect of marketplace fac-tors on private label penetration in grocery prod-ucts. Marketing Science Institute, Working PaperNo. 92–128.

R. Sethuraman (1995). A meta-analysis of nationalbrand and store brand cross promotional priceelasticities. Marketing Letters, 6, 275–286.

M.E. Slade (1995). Product rivalry with multiplestrategic weapons: an analysis of price and advertis-ing competition. Journal of Economics and Manage-ment Strategy, 4, 445–476.

L.W. Weiss (ed.) (1989). Concentration and Price,Cambridge: MIT Press.


Documents

Share, price and category expenditure—geographic market effects and private labels