This action might not be possible to undo. Are you sure you want to continue?

.

Sloan School of Management 50 Memorial Drive Cambridge. 253-2861 (617) mMm THE MARKETING CENTER Massachusetts Institute of Technology Alfred P. SALES PATTERN AND PROFITABILITY* by Leigh McAlister** Sloan School of Management Working Paper # 1622-85 February 1985 Massachusetts Institute of Technology 02139 Cambridge. Mass. Massachusetts 02139 (617) 253-6615 .THE IMPACT OF PRICE PROMOTIONS ON A BRAND'S MARKET SHARE.

.

THE IMPACT OF PRICE PROMOTIONS ON A BRAND'S MARKET SHARE. Mass. SALES PATTERN AND PROFITABILITY* by Leigh McAlister** Sloan School of Management Working Paper # 1622-85 February 1985 Massachusetts Institute of Technology Cambridge. 02139 (617) 253-2861 / .

.

Glen Urban and Lester Young. Others who have contributed include Alden Clayton. Peter Guadagni and John D. (617) 253-2861 Preliminary Draft: Please Do Not Quote * The author would like to thank Carnation and Rudolph Struse for providing financial support for this research. thank Selling Areas Marketing. Little for providing data and acknowledge Peter Fader's valuable contributions in analyzing the data. Richard Staelin. Michael Dermody. John Roberts. Mass. SALES PATTERN AHD PR0FITABILIT7* Leigh McAlister** Sloan School of Management Working Paper # 1622-85 February 1985 Massachusetts Institute of Technology 02139 Cambridge. Jim Lattin.THE IMPACT OF PRICE PROMOTIONS ON A BRAND'S MARKET SHAKE. John Hauser. Assistant Professor of Management Science .C. Scott Neslin. Inc. Subrata Sen. Diane Schmalensee. (SAMI).

.

offering price promotions is shown to depend upon the mix of consumers in the marketplace. SALES PATTERN AND PROFITABILITY ABSTRACT A model of market response to price promotions reflecting consumer heterogeniety in preferences and promotion responsiveness shows that a brand's market share is effected by the relative frequency of its own promotions and by its preference profile among consumers: widely accepted. brand's unpromoted "baseline" sales The overall profitability of are depressed by competitive brands' promotions. Consumers who do not respond to promotions and those who respond to promotions by stockpiling the brand they would have chosen anyway decrease the overall profitability of offering promotions. If consumers respond to promotions by increasing the a probability of choosing the promoted brand. The model is operationalized using UPC . scanner panel data on coffee purchases. smaller preference brands are more likely to gain long run market share through promotion than are widely accepted higher preference brands. Consumers who respond to promotions by switching to the promoted brand or by buying and consuming more of the promoted brand than they would have had there been no promotion tend to increase the overall profitability of offering promotions. ceteris paribus .THE IMPACT OF PRICE PROMOTIONS ON A BRAND'S MARKET SHARE. a brand's unpromoted "baseline" sales are depressed by its own promotions. If consumers respond to promotions by stockpiling.

.

Industry reactions to this trend are mixed. Faulty Management. at the Top in Promotions. The aggregated model is explored to develop intuition concerning the impact of price promotions on a brand's market share. This paper adds to that body a model of market response to price-off promotional offers. Hertz' chair. Roughly following the categorization schemes those studies. The model is built up from a collection of models of individual response to promotional offers which reflect heterogeniety in consumers tastes and in their response to promotional offers. An Ad Age article titled "Room least. finds them essential and productive. 7-Up's chair. at 2 By-standers attribute the confused spending in part. sales pattern and profitability. Literature Review Several studies have acknowledged and explored the heterogeniety in consumers tastes and in their responses to promotions 3 (Blattberg and Sen 1974.INTRODUCTION Expenditures for promotions (price deals. The magnitude of this problem and the relative confusion surrounding it has given rise to some interesting academic research. Peacock and Sen 1978). we suggest that consumers might respond to promotions in one of - 1 - . in Blattberg.) have grown dramatically in recent years. rebates. In "Sales Promotion . Olson." suggest that promotion management has remained clerical despite the fact that promotion expenditures have grown from 5X to 602 of a typical marketing budget.Fast Growth. coupons." Strang (1976) suggests that promotion expenditures have been used by brand managers to optimize their career paths rather than their firms' profits. to the management of promotion dollars. 1976. reaching an estimated ^60 billion in 1982 (Bowman 1981). finds such expenditures counter profitable. etc. The approach is applied to UPC coffee purchase panel data to derive insights and suggest guidelines for managing promotional expenditures in that product class. Winter.

g. such effects. Several researchers have investigated Rothchild and Gaidis (1981) draw on behavioral learning theory to a suggest that well designed promotions can serve as reinforcers and can increase brand's repurchase rate. Neslin. McAlister and Pessemier 1982) that have to do with the brand chosen but not with whether that brand was on promotion or not. Narasimhan 1984). and Lieberman Alternatively. (One function of promotion for new products is to In that situation. Fraser 1983. Cotton and Babb (1978) found that post-promotion purchase rates for dairy products were generally higher than pre-promotion purchase rates. a buying a promoted brand. the probability of choosing a particular brand because it is on promotion.e. Eppen. we eliminate the change in purchase probabilities (a manifestation of changed preference) that arises because the brand chosen was on promotion. (i. consumer might be enticed by the promotion to make a purchase (or to buy and consume more) from the product category.g. We operationalize this restriction by assuming that a consumer's response to a promotion for a brand does not effect that consumer's subsequent preferences among brands...) By assuming the choice of a promoted brand does not effect the consumer's subsequent preference. we eliminate the traditional time series effects of "learning" (e. Quelch and Henderson 1983). a consumer might increase 1981. Keuhn 1958) and "variety seeking" (e. both: A consumer might also do increase the probability of choosing a promoted brand and stockpile when Finally.several ways. stimulate trial and bring about awareness. A consumer might buy more units of a brand than she normally would stockpile) because the brand is promoted (Blattberg. those to be second order effects in this problem. we have assumed away two types of effects. Shoemaker and Shoaf (1977) found that the probability of repurchasing - 2 - . We restrict ourselves to the modeling of consumer response to the promotion of mature products. (Guadagni and Little 1983. We assume Second. In contrast. subsequent preferences and hence purchase probabilities very likely are altered. First.

"not on promotion". Little and Klein 1983. Theoretical and empirical evidence can be marshalled for either increases or decreases. and coupons (Irons.3 - . elicit different magnitudes of response. in-store display (Chevalier 1975). promotions on almost all brands occur so frequenlty that. to assuming that the promotional price reductions in the product class . reduction by a retailer. Neslin and Shoemaker 1983. price reductions of different magnitudes should The assumption we make is tantamount. Finally. The instrument we have chosen is a short term price We further assume that a brand is "on promotion" or is Clearly. Studies of short term price reductions by the retailer (Massy and Frank 1965. It is is offered (Guadagni this short run change in the probability of purchase that occurs at the time of the promotion on which we focus. Tybout and Sternthal (1978) propose self-perception theory as an explanation of this phenomenon. should long range preference changes exist. Guadagni and Little 1983). long run preferences are not altered by promotions for several reasons. That theory suggests that the probability of repurchase is lowered because the individual is uncertain whether her selection of the promoted product is attributable to preference for the purchased product or to a desire to take advantage of the promotion. We further restrict ourselves to the modeling of consumer response to a particular type of promotion. premiums (Seipel 1971). Scott (1976) and Dodson. such effects may cancel one another out. Narasimhan 1984) illustrate the differences that exist among promotional instruments. First there is no consensus as to the force driving or even the direction of such changes in preference.a brand actually dropped if the previous purchase was on promotion. therefore. with mature products. Second. We opted for the assumption that. the magnitudes of the demonstrated long run preference changes are very small compared to the magnitude of the demonstrated short run change in probability of purchase that occurs at the time the promotion and Little 1983).

When a regular consumer is not responding to promotions. MODEL We model individual choice behavior as a probabilistic process.that we consider are all of the same magnitude..4 - . Guadagni and Little 1983). all brands that she finds "acceptable" (ever purchases) are "relevant. has important implications for a manufacturer. Sunoo and Lin 1978)." When a regular consumer is responding to promotions. For a similar reason we do not consider the possible interaction effect of advertising with promotion (Brown 1974. We avoid that parameterizing in Che interest of analytical simplicity." A "regular consumer" has a fixed consumption rate A but may respond to promotions by altering brand choice or purchase timing. such understanding does not completely solve the manufacturer's problem of managing promotion expenditures. Ibraham and Lodish 1984). One could relax this assumption within the modeling context of this paper by parameterizing consumer response with depth of price reduction. We assume that an individual chooses from among a set of "relevant" brands with probabilities proportional to her preferences for those brands (Luce 1959). "category expansion consumer" responds to promotions by altering her consumption rate as well as by possibly altering brand choice or purchase timing.g. We consider separately the choice behavior of "regular consumers" of the product class and "category expansion consumers. An understanding of such response However. The manufacturer must also understand the ways in which retailers respond to promotional offers made by manufacturers (Chevalier and Curhan 1976. all acceptable brands that are on promotion are . it is important to point out that we consider the response of consumers to promotions offered by a retailer. Finally. It is not uncommon for such uniformity of promotional price cuts to occur (e.

> 1 ) . First we have the consumers who would not buy in the product class at all if there were no promotion. l-y • With probability regular consumer i will ignore promotions and choose from among all acceptable brands. (The units whose consumption is not attributable to promotion are considered purchases by a "regular consumer" of the product class c) Our model of market response to promotions is made up of two components."relevant". Second. Category expansion consumers may be enticed by a promotion to make purchases in the product class that they would not have made had there been no promotion. B We consider the profit implications for brand of the likely buying behavior of a collection of consumers described by such We first present the notation to be used and then develop the two models. . components of the model. this second component can represent a consumer who is indifferent to promotions (Y:=0) a consumer who reponds to promotions by stockpiling (y :^0 and n . consumer i With probability y £ (0 1 Y£ 1 1^. we have those consumers who regularly buy in the product class The extra units but who increase their consumption rate because of the promotion. regular will restrict her choice to those acceptable brands on promotion. The The second component describes the response of "regular consumers" of the product class. a consumer who responds to promotions by restricting choice to promoted ti . Depending upon the values taken on by individual specific parameters. that this type of consumer buys and consumes immediately are considered purchases by a category expansion consumer.5 - . Consumers exhibiting two types of choice behavior are aggregated here.< 1 for all brands j) or a consumer who responds with a combination of the above behaviors. first component describes the response of "category expansion consumers".>0. brands (y. The regular consumer's propensity to respond to promotions is also modeled probabilistically. .

Jt ] = Expected value of P. expects from "regular consumer" E t at time C G. J is not on promotion at time C [P. . Probability that consumer finds on promotion during an arbitrary shopping trip Fraction of time that IC B. .. 1* [G. J i ^i Probability that consumer responds to a promotion given that i a promotion is available on some brand acceptable to consumer J TI . 2. = -i^ iJ 1 . I' The set of "category expansion consumers" of the product class. tt . Consumer i's preference for B.6 . It . I The set of "regular consumers" of Che product class.0 E t if B. i-J > - for all j The set of consumer i's acceptable brands (those brands with with IT . It ] = Expected value over time ^ G. . > 0) w.Notation J B. Number of brands in Che marketplace j — brand. 1 Number of units that regular consumer per purchase cycle i of the product class uses average number of units that regular consumer buys per purchase cycle w. J is on promotion i Profit that B.. ^ J V tt . 4 Jt 1 if B .. i of the product class Number of units that category expansion consumer i buys because of a promotion that she would not have bought had there been no promotion. Jt a over time B . j=l. is on promotion at time t . . ..

. expects its promotion i il . then category expansion consumer Ti. It ] = Expected value over time of G. 0< Mp< Mj^.. = Probability that no brand that consumer as competitive to B S. TI. t (Pi^=0). E [ n k?*l (l-P )] i views is on promotion at an kee./( Z will buy w. =1). k. Category Expansion Conumers If B.) yielding M„ per unit to the manufacturer of - 1 - 7 - . ^ B-^ It i Expected profit to per purchase cycle.. P..TT.jti. V = E t [G. is not on i promotion at time B . It Profit that at t Bi ' expects from "category expansion consumer" i ime t G. n.. receives from each unpromoted purchase 1 . t (P. n. 1 il. F kffl i P.. ^^ "-^ to inflate the probability that consumer will choose B .jP je e -^ . from "category expansion consumer" Factor by which a stockpiling consumer increases the number of units purchased in response to a promotion. then category expansion consumer will not buy If B i is on promotion at time B.. Expected profit to B purchase cycle I from "regular consumer" i per G. ^ = E^ [ TT ] = the factor by which B. + .>l. M„ N M = = Profit margin that B. Profit margin that B^ receives from each promoted purchase.. units of with probability B . arbitrary point in time.

+ .TT kfl k£e S. ^'^ (Var^ <= P. periods is given by 3) . I Ot. The profit that the manufacturer of equation ( B. that. '] It = an il w'. We assume consumer i goes to the market to buy w. (tT . where k^j^ is 1 ^1 ll . 1.. '^'^ ) k^l kce !_ + TT xl •.+ . units from the product class to which brand B belongs. ) 1 1 Regular Conaumers of the Product Clama Since a regular consumer of this product class may respond to promotions by stockpiling (buying w-n. kfl k£ e . k . 1 (2. in period t. We consider the n. .n. G.TT . \£. or possibly w. TT Jt IJ .-l. . units) we must consider the implications of the promotion in the period it is offered and in the subsequent n.-l periods in which the stockpiling consumer uses up her inventory and makes no further purchases.. k Ik ll [ a. kt ll I 71. + il /^i P.il M P 1 .. And. = E c [G. assuming the independence of P-^'s across j. J"..8 - .'l. expects from consumer i across the n. ik . .k. .. units rather than w. kfl e . periods between t and t+n. G' = P ( '^i TI ^il I ) M^ (1) ..

For the next n. 1 ll P 1 !• ]} + (1-P.-l)G. ". 1 ktlk Ti.M„+(n. On her shopping trip in period t..M +(n. It 1 ( . I in period t yielding M^ per opr unit to the manufacturer of B.-l)G.)[TT. Consumer i may or may not react to promotions... Hence we develop a recursive relationship in equation (3). + (l-r. units of yielding M_ per unit to the manufacturer of B^ in the subsequent n.w.. 1 1 )M+(n.) consumer i will buy w..-l)G. B in period t. 1 . units (y • = probability she will react). » P. t+n.-l periods consumer 1 "^ i will be out of the market therefore yielding $0 to the manufacturer of B. 1 !• 1 ll 1 N ]} (3) This equation can be explained as follows.w.. during those periods If ( B^ is on promotion in period ti .)[TT. t but consumer i i is not reacting to promotions B.-l)0] + (l-y .n. (Y-[w. kt ))[tt.. then with probability P /(n ll + F ll . il 1 N 1 !• ] kee. Our best guess of the profit that the i manufacturer of ^i* » B^^ can expect to receive from consumer in Chose periods is that quantity for which we are trying to solve.w. = 1) or does not B^ (P^ find brand B. ) then with probability . = 0) consumer i either does If she does (P. ){y. (l-P..-l periods.n.9 - . on promotion. units of 11 B. n .n. find on promotion she may or may not react to promotions by restricting her choice to those brands on promotion and buying w. consumer will buy w. We don't know what will happen may or may not be on promotion. l-y . kfl k£e.( ^ kf 1 kee. is on promotion and consumer TT i reacts to promotions. .M +(n. If B.-l 1 TT / G.

we take the expectation of equation (3) across time. '' ) consumer i will find no acceptable brand on promotion and will therefore choose among the brands she finds acceptable with probabilities proportional to her preferences for those brands.^+(i-Y. 's units of B^ in period t with probability ^ yielding M to B manufacturer.)I3. +(l-Y-)]Mp ll (1-a. . yielding Mj^ per unit to B.+(l-Y. . units of that other brand yielding iO to the manufacturer of B for all n.-l)G. (1-P.)]M_ ' 1 1 L il 'i P (i^^)[Y.n.7T.'s manufac's turer.) + X (l^ 1 . In each of the subsequent (n. With probability K k^l k£ e .-l) periods. = l .(3. !• the profit that B 1 's manufacturer expects from consumer during any given purchase cycle.k.{a1 il [Y•n. . —— TT . is not on promotion in period t and if consumer i is not restricting her choice to promoted brands in period t then consumer Ti .)]Mj^ + (n. 11* = w.units of B.[Y. t+n. B. is not on t promotion in period t. she will buy w. i is expected to buy w.k.Consider now the case in which consumer consumer i B. 1 1 ll +(l-Y-)]Mj 1 (^) N 10 - .'s manufacturer expects to make G.G. for G.a. If is reacting to promotions in period and if any brand acceptable to i i is on promotion during period t then consumer will buy n.-l)Y . -1 L E t r '^ T=t / G. periods. TT • With probability ..(l-(3 11 + {ci. B manufacturer can expect to make G.. To solve i from consumer i. It ] = n.Y-] ll 1 W G. In each of the subsequent n--l periods. 1 !• [(1-Y . + (n. If B from consumer i. )) 1 (l-a..)[Y.

Kt independent across k (i-ct. = n P.. w. ))) to consumer i being in the market 4 (i. she will choose B. (1-a .11 - . cycle. If she reacts to promotions.) )) I il can be viewed as a purchase incidence model. With proability (1-a. . units of B^ with probability yielding M i per unit to B 's manufacturer.. 1 with probability it. (1-13 .> 1 ) 1 Consumer i who stockpiles (has n will be expected to make a purchase during Note that if any arbitrarily selected purchase cycle with this probability. that she will not be forgoing purchase because she has a stockpile at home).. The probability "^ l/(l + (n. k£e • Equation (4) can be interpreted as follows.-l)Y 1 ' . From equation Embedded Purchase Incidence and Brand Choice Models. with probability unit to B. units yielding Mp per she will choose If she is not reacting to promotions. Given that consumer .) where k£j is defined as before and kee.) she will choose w.n.e. ^iA^i *nd buy w. (4) one can extract two probabilistic models. During any arbitrary purchase . IT. per unit to B 's manufacturer. we assign probability l/(l + (n . she will see on promotion with probability a .. units of 1 B. i is in the market.=l) we expect the consumer to make a purchase during every purchase cycle. (l-a . .) consumer to react will not see B on If she wants to a promotion but none of the other acceptable brands is on promotion (y-B-i) or if she does not choose to react to 1 il promotion (l-f. 11 ill .-Dy (1-Q ' ..'s manufacturer.. 1 il yielding M. B. consumer i is not a stockpiler (has n. L assuming B. promotion.

**We can also extract a brand choice model from equation (4). that we assign to consumer
**

i

The probability

choosing B^

*

is

Yi^^'i^i^J *^i^'^i^^i\i

^'^i^^r

i

**Referring back to equation (3) we can infer the probability that consumer
**

will choose

Bj^

at time

t

is

2^[B

i

1

]

=Y

i

^iUrlt ^^

^

+

Y

i

[

n

(1-P

kt

)]

n

il

+

(l-y

i

)

il

^k^t

1

l^'

'^""i

tee.

MANAGERIAL DfPLICATIOIlS

Impact of ProMOtiona on Long Run Market Share:

Saall Preference Brand

Advantage

From equations (1) and (5) we can see that if no brand ever promoted

(i.e.,

B.

if P

.

=0 for all

j

and all t),

then the long run market share for

**(denoted M.) would be expected to equal preference share among
**

(tt^

**regular consumers of the product class
**

TT

.

.),

where

=

(

Z

W.TT

'

•J

iel

.)/ Z W. ' ^J iel

.

However, if consumers are promotion sensitive (some Y.'s>0) and if some

**brands promote much more frequently than others, we would expect (from equations
**

(1) and (5))

that those brands that promote much more frequently will have long run

market shares that exceed their preference shares.

Conversely, we would expect

**those brands that promote much less frequently to have long run market shares that
**

fall below their preference shares.

In this way a brand's own promotion schedule

and its competitors' promotion schedules come together to determine, in part, which

brands gain share and which lose share due to promotion activity.

**Promotion driven share gains and losses might result from promotion
**

effectiveness as well as from relative promotion frequency.

A promotional price

- 12 -

.

cut of i.40 on a il.OO item would probably have a larger impact on share shifts

than would a cut of $.20.

**A promotional price cut that received a great deal of
**

a

**retailer support (feature, display, etc.) would probably have
**

share shifts than would an unsupported price cut.

larger impact on

One might also expect intrinsic

**properties of a brand (level of national advertising support, market share, etc.)
**

to temper promotion effectiveness.

From the models in equations (1) and (5) we can

see that one such intrinsic quality,

preference share (n..) does influence

promotion effectiveness.

Among acceptable brands, smaller preference brands get

**"more bang for their promotion buck" than do larger preference brands.
**

To see this consider the factors by which two brands (B, and B.

1

,

2

tt.,>tt.->0) il i2

i

can expect to inflate, through promotion, the probability with which consumer

will choose them, given the same level of competitive promotional activity.

Inflation factor for

B.

=

1

xl

^

kt ik

kj'1,2

kee

1

1

Inflation factor for B-

= " -I* il

I

'"

P,

^ •, kt ik

k?'l,2

kee

Since

B.

.

tt

>

tt

the inflation factor for B

is

**greater than that for
**

B

B_ can,

therefore, get a greater percentage increase in sales than

for the same promotional effort.

Impact of Proaotlon on Sales Patterna

**From equations (2) and (4) we can infer that
**

expects "baseline" sales of:

B^

,

when not on promotion,

ill "'^

""^""il

l

+ (n.-l>r.(l-(3./l-a,))

1 1

[Yi(3ii ' '^

*

(WOI

^

il

1

- 13

When promoted,

.Z,

i£ I

;

Bj^

**expects a promotional sales peak of:
**

[Y-n.k.,

-,

11

"i^il =-r

;

TT

1

llll

+

(1-Y.)] *.L^,w'.k.,Tr.. lellllll 1

ll

Figure

1

**illustrates these levels of expected baseline and promoted sales.
**

[Figure

1

about here]

.Z , leliil

The point on the expected sales axis marked

w.tt.,

represents the level

of sales that B. would expect if neither B^ nor any other brand ever promoted.

Region E represents sales to category expansion consumers while regions A, C and D

**represent sales to regular consumers of the product class.
**

is

The height of region A

**proportional to expected sales to regular consumers who are not responding to
**

The height of region C is proportional to expected sales to regular

promotions.

consumers who would like to respond to promotions but find no acceptable brand on

promotion.

**The height of region D is proportional to expected sales to regular
**

's

**consumers who respond to B
**

It

is

promotions.

interesting to consider the impact of differing promotional responses on

the "baseline" (region A + region C) and on the height of the promotional sales

peak (region A

+

region D

+

region E).

If regular

**consumers in the product class
**

B

's

stockpiled but did not switch brands in response to promotions, then the

**expected baseline and expected promotional sales peak (functions of a when n.
**

is

greater than

1)

would be driven down by

B

's

own promotions and uneffected by

**competitive brands' promotions.
**

If regular consumers of the product class switched brands but did not stockpile

in response to promotions,

then the B

's

**expected baseline (a function of 3.
**

)

)

and expected promotional sales peak (a function of k.

would be driven down by

**competitive brand's promotions and uneffected by
**

If regular consumers

B

's

B

's

own promotions.

stockpile and switch brands in response to promotions then

baseline and expected promotional sales peak are depressed by both its own

and competitive brands' promotional activity.

- 14 -

FIGURE 1: Expected Sales to Bi Expected Sales -^ €l ^fiTin^kn^il iel ^^' I E^>ected I Unpromoted Depression Due to Promotions i Sales I J/i^i«il^ilf ^/i(l-Ti)^il yk: BI off Promotion B j on Promotion B^ off Promotion Vfi h = l*(nj)Tj(i-6jj(l-oii)) .

] { [1+Y ./3a. If that derivative B.(n. the fraction of time that B^ is on promotion. From (6) we observe chat 9G .JHY. . 3G 3a r3G.]Mp> [W.k. To establish the attractiveness of promoting to this collection of consumers we consider the derivative of the expected profit function.)]M 1 ll 11 IN } (6) and 3G' 3a . The manufacturer of 1 B is always made more profitable by promoting to those consumers for whom extra consumption can be induced by promoting. W.-l)(l-(3. Impact of Proaotiona on Profit From equations (2) and (4) we can develop an expression for the total profit that the manufacturer of B = expects to make./3a • > 1 if 1 [lni(n.[G. (7) I From (7) we observe that 3G • .-Dd-ra. ie I r9G.. If that derivative negative then gross profit to decreases as the promotion of B. . + [ G.n. .^)[Y. 1 TT . 1 ll . increases.n. 1 is always positive. with respect to a . G^^. = G E. B. [(l + (n.-l)][Y.-l)Y -d-S-^d-a^). +(1-Y.k. Less promotion is better.' ie I ' 3a 3a where 3G * 3a ^ .^ + l-r JMj. 1 - IT .k. is positive then gross profit to B increases as the promotion of is increases.Q. ] I G.(n.M„ ll ll 1 P .w.(n.il +(W •)]Md P 11 il 1 1 1 [1+Y .-l)][Y.^ + l-f .Q. More promotion is better. (8) - 15 .

-l periods. S.units of the brand which they would have bought anyway and then forego purchase of that brand in the next n. by definition.. ) ) at margin M„. Other consumers might be willing to switch brands for stockpile (k. Q. <l.. N If the promotional sales peak times the smaller margin is greater than the base line times the larger margin (i. = 1).)M^ i ..tt of region C + region A ''1 (y +1-Y ill.n. M and.^(Y.6 ill .v .. is too. is > (y . * 1. It is. defined as being smaller than not profitable to promote to unpromoted margin. unsurprisingly.. . = 1. promotes it ex^ pects sales proportional to the height of region D + region A (Iv at margin M . > > 1. M-. a switcher i a switcher If B .^. . The logic behind this condition is illustrated in Figure 1.). (y-k. Consider first those consumers who do not respond to promotions For these consumers inequality (8) reduces to M this cannot hold (promoted margin. It is to simply "measure the bump" since a brand's own promotions are lowering its - 16 - .^(Y. consumers who do not respond to promotions.)Mp>I. •^ a promotion but not l1 il Promotion to these consumers will be 1 profitable if lTT.n. n.Inequality (8) will hold for some consumers and not for others. =0). there is no not appropriate simple rule of thumb for evaluating promotion profitability.. In this case.inr. Consumers might respond to promotions by stockpiling but not switching brands. at a reduced price.k. Promotion to these consumers unprofitable because they take advantage of the promotion to buy.e. . We next consider profitability of different patterns of consumer response to promotions. if the promotional sales "bump" is "big enough") then promoting to these consumers is profitable. +1-Y-)) If B^ does not promote it expects sales proportional to the height (Fw. w. If consumers respond to promotions by switching and stockpiling. (it. M^.^*W. k.«. inequality (8) reduces to M M^.=l).>l.

Folgers lb. lb.S). 1 lb.L). The predictive ability of these models is tested We then reestimate the against the 747 purchases made during the next 20 weeks. 1980. (BUT. - 17 - . 1978. Butternut 3 3 (BUT. one would not expect to find a preponderance of consumers whose consumption rate would be effected by promotion. )] to account for future il 1 X ' 11 unpromoted sales chat will be lost because of stockpiling.S).-l)(l-(3 . Folgers 1 lb. models using 65 weeks managerial guidelines. 3 (FOL. We will initially. is restricted to the 65 weeks between December 20. Category Expansion Consumers Given the nature of coffee consumption. 1 (FOL.-l)]/[l-tY . (MH. Maxwell House (MEL. 1978 to March 12. Maxwell House 1 lb.L). Inequality (8) asks wheCher profit from the promotional sales peak at Mp per unit is greater than baseline sales at M^ per unit inflated by the factor [l^Y .baseline. (Such behavior would probably be more evident in snack food categories). The purchase histories include competitive a SAMI promotional environment at the time of each purchase for panel of 200 consumers. Folgers Flake (MH. Mellow Roast lb.S). lb.(n.S. Because competitive promotional environment data did not become available until December 20. each making more than 902 of their purchases in one of four stores in Kansas City over a period of 78 weeks from September 14. 1 (FOL. In fact.S).(n.. our analysis 1978 and March 12. 1980. APPLICATIOH Data The data used to illustrate the procedure just described is the same as was used by Guadagni and Little (1983).F. use the 2223 purchases from the first 44 weeks to estimate models for 100 of the consumers.). of data and use these reestimated parameters to develop The brands among which consumers chose are Butternut lb.L).

5 who make every purchase One of these consumers makes more than twice as many purchases as does the average panel member (48 for this consumer versus 21 for the average consumer). gory expansion consumers should either: a) b) Cate- Make all purchases on promotion.75 times out of an average 21. The other four consumers purchase only infrequently.) This implies that the other 59 consumers who ever purchased multiple packages of coffee on a shopping trip did so an average of 1. Among the 200 members of the SAMI panel we find only on promotion.we find very little evidence of such altered consumption rates in this data. This consumer also buys six of the eight brands offered. The purchase of multiple packages of coffee on s Only 68 of the 200 consumers in the sample ever Nine of those 68 purchased multiple packages of coffee on a single shopping trip. Among all 4. It is also important to note that the multiple package purchases were not especially associated with promotional offers. consumers accounted for 60Z of the shopping trips in which multiple packages of coffee were purchased. making a total of 33 purchases (less than 1% of all purchases). 45% were made on promotion. 51% were made on promotion. (Examining these nine consumers' purchase histories reveals that these consumers' regular purchase quantity of coffee included multiple packages. We also found very little evidence of increased consumption by regular consumers of the product class.216 choices made by all consumers in the panel. Among those 151 . This suggests that this consumer might be a regular consumer of the product class who is highly promotion sensitive and willing to switch brands to take advantage of promotions.18 - .1 shopping trips. single shopping trip was rare. or Buy more units on promoted purchase occasions than on unpromoted purchase occasions without effecting interpurchase times. Among the 254 shopping trips that involved multi-package purchases.

We also found no effect of promotions on - 19 - . 502 of the multiple purchases were on promotion. The very infrequent multiple package purchases (1. To estimate n. Christmas. Hence. we again Stockpiling might manifest itself in at least two different ways.. by the ^ faction of time that B. a consumer might manifest her stockpiling behavior by making an a additional (promoted) purchase in the middle of purchase cycle and then forego Such behavior could purchase at the expected purchase time in her purchase cycle. tain an estimate of Y turn to the data. As just noted.-l purchase cycles. Consistent with this we see that 24% of all multiple package purchases occurred around Thanksgiving. and New Years.1) is consistent with consumers having an occasional need to purchase extra coffee for entertaining. We will ob- j€e.75 out of 21. we found no evidence of multiple-package purchases being especially associated with promotion.). units on a A consumer promoted purchase occasion and then skip purchasing for the next n. ct . as the average number of units chosen during purchase cycles in which consumers accelerated purchase to take advantage of promotions. Since there are so few consumers in this database who might be category expansion consumers and since those consumers' behavior is not inconsistent with the model of regular consumers of the product class.. be accommodated in the proposed model by estimating n.n. Estimation: Purchase Incidence Model . J il can be drawn directly from the data: ^ . is on promotion and 6 by II (1-a. we opt to not treat them separately. who purchases at regular intervals might purchase w. it appears that some force other than promotion drives multiple package purchases. Estimates of a and 13. Alternatively.multiple package shopping trips made by the nine consumers who regularly purchase multiple packages. • from the brand choice model. J .

5 parameters (3. average probability between the fit and prediction periods indicates overfitting. and Lieberman 1981) the time between nonpromoted and a promoted purchase (NP*-P) to be shorter than the time between (NP*NP). This pattern reverses that which one would expect if purchases were being accelerated due to promotions. Brand Choice Model We use maximum likelihood to estimate the n. degree of However. on average.656. The average probability assigned to the brand actually selected across all 1. and the average value of the 1264 occurrences of NP^NP was 19. The likelihood function made up by multiplying these probability expressions together is maximized subject to the constraint that obtain estimates of Y . E«tim«tion.'s. NP+-P two nonpromoted purchases In this data base.purchase timing. 14 purchase observations to ' estimate. the model fit very well. The expression shows the probability of consumer i selecting the brand actually chosen at time t.1 days. we create a probability expression using equation (5). 4. on average. we found that the average value of the 727 occurrences of was 19. we take n. If consumers accelerated their purchases for promotions.4 days. Unsurprisingly.20 - ..'s and y. the predictions made by the proposed model are quite good. . We began by estimating the proposed model using the first 44 of the 64 weeks of purchase histories of the 100 randomly chosen consumers on whom Guadagni and Little (1983) fit their model. The average probability assigned to the brand The decline in a actually selected during the 20 week holdout period was 0. Given that promotions seemed to affect neither the number of packages purchased nor the timing of purchases for the consumers.5 n s and 1 y) P^r consumer.539. ^ Y^ ^ 1 to and the ti 1 . ij . For each purchase occasion. This gave. = 1 and the purchase incidence model loses importance in this application. one a would expect (following Blactberg. Eppen.417 purchases made by the 100 consumers during Che 44 estimation weeks was 0. for each consumer based on her purchase history and the promotional environments she faced.

We also used a Monte Carlo procedure like that proposed by Chapman and Staelin (1982) to test for possible bias in model parameters. p< 0.Y-" .When compared to predictions made by Guadagni and Little's (1983) model that used several other predictor variables (e. it is 1- iJ meaningless to consider each individual. are passed through the maximum likelihood procedure to get estimates.001. the Appendix) takes as input a set of "true" n T . and y .. at least . and y.g. . Bias for the tt . E tt . we estimated the parameters for all 200 consumers using the entire 64 weeks of data. size of price cut. The procedure (detailed in T It . Wilcoxon statistic = 173851. ij and y • iJ ~ Y 1 • Since theZ j tt. . = 1. whether prior purchases were on promotion). gammas are biased downward.21 - . we choose a random brand for T . the proposed model was shown to make statis- tically significantly better predictions (average predicted probability for Guadagni and Little = 0. That history. The sample is made up of 100 randomly chosen individuals. Gamma bias is measured by perfonning a two-sided t-test on y is E . When five or fewer observations are used to estimate about three parameters. Instead. t-statistic associated with the difference between predictions = 6.477. This increase in the ratio of observations to estimated parameters should ease the problem of overfitting. T tt . ' s appear to be unaffected by bias . From Table 1 we see that the tt . . regular price. p< 0. The average number of parameters estimated per consumer remains unchanged while the average number of choices available to estimate those parameters rises to 21. for all j.. along with the related promotional environments.001).. then chooses a set of random promotional environments and generates a choice history.'s measured analogously.5. Having thus gauged the predictive ability of the proposed model. .5.04. The reason for this bias is closely related to the issue of overfitting discussed earlier. E . We are interested in the differences n . . . but when sample sizes become very small. - tt . - E tt .

. Figures 3B-3E show the distribution of values of Y in this data base. Each brand-size has a few consumers with values of are thin. and then generally fewer consumers with lower levels of promotion sensitivity until we get to extreme promotion insensitivity where we see another.22 - .'s alone explain most of the variance When a consumer makes so few purchases. made up of consumers who never buy the brand-size. greater than . chain stores is higher than in the warehouse store. . The bulk of the non-zero tt . lacks a spike of promotion Each store has consumers that display a range of values of On average. . . . Fortunately. but in general the upper tails Folgers 1 lb. the level of promotion sensitivity in the promotion sensitivities. Store 4 is a warehouse store. 2. present in such small samples. .S oae parameter becomes useless.5. The stores appear to have distinct consumer promotion sensitivity In particular only the warehouse store (Store 4) has a spike of 1 promotion insensitive consumers and only Store sensitive consumers. .. The Tr. (Figure 2 About Here] Figure 3A displays a discretized frequency distribution of values of YWe see a spike of highly promotion sensitive consumers across the total sample. TT . from 53 consumers for Folgers 1 The size of that group ranges 1 lb.5. most members of our sample make enough purchases to avoid these small sample problems. to 18A consumers for Mellow Roast lb. smaller spike.330. 's have values less than . We see that each brand-size has a This mass is largely mass of consumers with values of between and . A discretized frequency distribution of estimated ' s across all 200 consumers for each brand j is given in Figure it .05. • in each of the four stores Store 1. [TABLE 1 ABOUT HERE] tt . 2 and 3 are members of national chains. it is difficult to make strong conclusions about her promotional sensitivity. has the thickest upper tail which accounts for the tt relatively high value of = • FUL . profiles.

TABLE 1.001 -0.003 0.01. .0002 * Significant at p < 0.1.108* -0.005 0. All others significant at values of p > . AVERAGE VALUE OF BIAS IN MODEL PARAMETERS Number of Simulated Choices 5 8 10 20 Average (y'^-f^) -0.047 -0.002 -0.002 Average (n'^-rT^) -0.

f.5 1.S .s .0 .L 131 n FOL.5 ^ —t— —I— —|— —I— — < I t f .5 1.137 •'"' "but.— 104 — FIGURE rr 2 : c r FREQUENCY DISTRIBUTION OF VALUES OF » I II ACROSS THE 200 SUBJECTS 150 53: p ^ 40-1 u u 3 C/1 00 40- II„.L -"^ ^'' "hh-s - ^MH. '1 CD "•5 40 40- 40- o •"-1 npoL.l = W.5 1.'^' "mel.^ ^ = BUT.f.0 ^fol.0 n FOL.L 176 n MH.s = .068 "mel.L 4 u 40-1 40 3 n D B 3 "fol.5 1.L = '''' 3 V3 w <u 20H 20 20- JH z a a k U" -I 1 — — I \ I I I I \ —— q 1.0 .S -.0 .5 1.5 1.0 I -t— I — —r— I T ^lJ^ ^ — — 1 I .s.s 20- = -^^o u 20- 20- i 3 Z "lJ-^ ^ —^— —I— — — — I I I ^-1 in 1.s n BUT.s 20 ' °^' J3 20H Z 1 I 1 ' I I I I 'I I I — I .0 .S 184 n MH.0 n but.

[Table 2 About Here] Having aggregated parameters to estimate market response we can now investigate a number of questions of interest to managers. competitive brands' We then consider the impact of promotional activity and retailer cooperation. If i . we can estimate market level statistics by a simple weighted averaging. ij ij .. the share of consumer i If no brand ever promoted or if consumer f. ij is the share of consumer i's choices that go to brand j.-TT . on a brand's prof ibability Finally.23 - . j Consumer i less frequently than she would have had there been no promotions by any brands. . . *^ Oi then brand j's promotion calendar was ineffective on consumer bought brand relative to competitive brands' promotion calendars. promotion on expected sales patterns. . We begin by considering the effect of promotion on a brand's long run share. and n i's preference that goes to brand j. ~ ''^ . [Figure 3 About Here] MAMAGERLAL DfPLICATIORS Since these consumers do not display stockpiling behavior in response to promotions. then f . brand j Consumer i was induced to buy more frequently than she would have had there been no promotions by any If f-i brands. were insensitive to promotions. we consider the impact of promotion I«pact of ProBotiona on Long Run Share The basic unit of analysis will be f . > ij ij 0. . . . . where is f . ' then brand j's promotion calendar was effective on consumer relative to competitive brands' promotion calendars. We investigate several forces behind the resulting gain or loss in share: brand characteristics. . - tt . . . . would be expected to equal tt .

trt .

.

and Maxwell House 3 lb. Note that Mellow Roast has no manufacturer dummy and therefore the coefficients of BUT. . and Mellow Roast 3 1 are shown to have lost share while 3 Butternut share.. i-' the standardized frequency of brand j's promo- tions in store s. . 1 lb. and lb.24 - .. across consumers for each brand 1 i. This effect seems not to be driven by consumer preferences among manufacturers. and Maxwell House 1 lb. the weighted average of 1 - TT . the average values of y. The dependent variable in this regression is fii"^... consumers to promotions. Q independent variables are PROM . Small Preference Brand Advantage . Folgers lb. to which consumer brand i is loyal . lb.. Folgers 3 lb. and for consumers who chose Maxwell House lb. 1 both gain share despite the fact that they promote less frequently than their lb. counterparts.'s share and Mellow Roast 1 Ib. As reported in the fifth column of Table 2.'s share.. . - rr . . consumer i's preference for j. lb. Folgers 1 lb.. promoted much less frequently than did other brands. FOL and MH that indicate the brand's manufacturer. for consumers who chose 1 Folgers 1 lb.* With one observation for The each brand in each consumer's acceptable set. on the other hand. Folgers 3 lb. of Folgers 1 The effect seems also not to arise from a differential sensitivity lb.. FOL and MH should be interpreted as the advantage or disadvantage to those . and Maxwell House 2. are in the middle of the range of values that that statistic takes on. are among the heaviest promoters and yet they too lost share..'s and Maxwell House Ib. Maxwell House lb. . The following regression sheds some light 1 on the reasons behind Folgers 1 Ib.'s share losses. we have 723 observations.The third column of Table 2 reports M f. it is not surprising that the net effect of all brands' promotions was to lower Folgers Flake 1 Ib. and Maxwell House 1 lb. and Mellow Roast lb. and dummy variables BUT. Given this. all gained 1 From the fourth column of Table 1 we see that Folgers Flake lb. Folgers Flake 1 lb. Ti .

016 FOL . however.-Tr. ij IJ . widely accepted.39) . is f. .-it . The regression 9 is: f. We see that larger values of tt.. Retailer gavoritisa.31) + (c=2. . f ...012 (9) R^ n = » . this favoritism should show up as a significant coefficient on the dummy variable(s) corresponding to the favored manufacturer(s) The results of those regressions are: . result in 10 smaller values of and Maxwell House . J ij (t-1.. - .g. it ..038 + .-n . gives deeper price cuts or better display to one of the manufacturer's brands). . manufacturers' brands relative to Mellow Roast. We explore this retailer heterogenity by If a retailer favors one manufacturer's running the regressions within stores. appears to have a slight advantage over other manufacturers' brands.98) + (t=-18.36) MH (C=0. The most important predictor of .. brands over another's (e....88) + BUT (t=0.. lost share because they are both relatively high preference brands. (9) reflects in part. ij 1 ij This regression suggests that Folgers 1 lb.040 (t-1.47) . The value of R^ in the regression reported in equation the differences in how different manufacturers' brands are treated in different retail stores. . a brand manufacturered in the Kansas City area. . As was pointed out earler in this paper.261 tt . lb. and ij that Folgers.-n.25 - .025 PROM 3 ^ . 344 723 ij Here we see that more frequent promotion leads to higher values of f. smaller preference brands have an advantage in promoting.

40) (t=. Store favors Folgers.55) 165 .050 (t PROM 3^J it BUT ij FOL MH .021 -.34) 4 86 .53) (t-. Store favors Folgers and Maxwell House and Store 4 favors Folgers and Butternut.41) .007 -.066 (t=2.21) (t=-1.343 (t=-10.100 .103 .124 . lapact of ProMOtion on Sales Pattern* The heavy lines in Figure 4A-4H represent the expected levels of promoted and unpromoted purchase share 12 for the eight brand-sizes in this study.060 . shows The results also suggest that 3 Store 1 2 treats all manufacturers' promotions equally.51) (t». Discussions with the sales force could help corroborate these indications and might lead the sales force to discover and hopefully remedy problems of which they might have been unaware.107 .046 (t=-1.68) (t=1.22 -.44) (t-.53) 362 .45 .263 (t=-13.21) .07) .40) (t-2.35 .096 (t=-3.297 (t»-8.62) (t«.39) (t»2.23) (t-.29) (t=-1.042 (t=-1. tt ^ .17) (t=1. A peak represents the expected level of the brand-size's purchase share when that brand-size is being promoted.44) — .061 -.02) -.020 -. expects a - 26 - .141 (t = 1.20) -. in Figure 4A we see that Butternut 1 lb.031 -.STORE n R^ CONSTANT -. (For example.047 -. The lower level of the heavy line represents the expected level of the brand-size's baseline purchase share when that brand-size is not being promoted. Such results should be used only as indications of possible retailer bias.50) (t=3.022 (t»-.035 (t-1.084 .157 110 .93) -.44 .92) From these results we see that the negative coefficient for itself as the strongest effect within each store.

and Folgers Flake lb. the coefficient of promotional leverage for brand j. Mellow Roast 1 lb. Maxwell House and Butternut owe their large values of L.. tend to be less promotion sensitive than average (average the smallest value in for consumers of MEL. L . Similarly. Folgers Flake to find 1 lb. 1 owe lb. Maxwell House 3 and IL.F. 2 fact that those consumers who choose Maxwell House lb..F. then lb.) their small values of L . the 1 largest such value in the column. since I^ql.27 - . the expected purchase share peak for Butternut lb.) .for consumers of these two brands is 0. Mellow Roast 1 and Folgers 1 lb.purchase share of 0.50. was promoted three times as frequently as Folgers Flake 3 lb.S = 0. lb.S' lb.25 when The width of the expected promotional purchase share peak reflects the (For example. and Butternut lb. ^"'^ W. and Butternut 3 tend to be more promotion sensitive than average. 3 lb. implies that B. (The fifth column of Table reveals that the average value of y. can expect to be made more profitable by promoting If all brands have the same values of 3 >L. should expect to find promotion profitable r if any brand-size does..) a purchase share of 0. since L-„ .S and FOL. last column of Table 2..is the ratio of the expected promoted share to expected unpromoted share for brand j. 3 able.11 when unpromoted and promoted. the fifth column of Table 2. is three times as wide as the expected peak for Folgers Flake 1 lb.. and L-^.) [Figure 4 About Here] Impact of ProMOtion on Profitability L. to the fact that consumers who choose Mellow Roast J and Folgers Flake value of Y • 1 lb. 3 to the lb. Since consumers aren't stockpiling. is reported in the upper right-hand corners of Figure 4A-4H. should expect promotion profitable only if all other brand-sizes find promotion profitI^^^^ g. lb. Butternut 1 ' relative frequency with which the brand-size was promoted. Consequently.71.S 3 ^^^^ '^^ smallest values. than by not promoting.^ take the largest values.) (These values are also reported in the L.Mp > Mj.

• '"' .0 .1.s - .S ''iu •-'* .20 .L . ncuu *A: urccTZD njROiAte shau rot Birmimrr i ui.s • ^-'o* W.L ncuM 41: latcTto pvtauat shau PATTtBi rot sumiuBrr ] u Vt.157 •but.S . nucTioi or luauT COIK TO lUT.30..i 'N PROMOTION at • 8UT.30 .10 BLT.Ix*«cta4 Fractloa af Narkat Cotn| CO SUT.

S ON PHOWTION "mh.5 • ^^" MM.MTCRll FOR MAXWELL MUSE I U.L "«. HOUSE } P.s"-'" . corf*c TO .3--'" KH.L MH. S«.riGUlt tt: UPICTtD PUIOLASI SSAlt nCL'U exrecTEO 3 F«ACTIOK OP TMI >tA«KrT .40 <>f: UPICTZD PUXOUaE SHAU PATTEM FOI NAXWEU.

. has a relatively small value of LpQL. Mp-/Mj^. Mp. it would require a much larger share increase than the other brand-sizes to expect a similar percentage increase in share as those brands expect.L 3 ^""^ ^..S ^ °-^^' ^P. lb.BUT.74 is "large"). and M .S Maxwell House Folgers Flake promoting.BUT.That Folgers 1 lb.S^^N. for Mp^MH.s *=^"°°^ ^® 1 completely explained by Che promotion sensitivity of Folgers that value is in the same range as values of average Y- lb.. will be "large" (in this case M„. and Maxwell House 3 lb../M will be "small" (in this case.F.S ^ °*'^^' ^P.L/^.MH. make more profit by promoting than by not then ^pi/^wi If it costs a manufacturer very little to promote. the percent increase that B.l/^. lb. preference brand effect" noted earlier.. profit margin per unit will differ across brand-sizes./Mj.s/"n. > 0. Folgers 1 lb.BUT.MEL./M„.BUT.s/"n. 1 °-^^' "p.FOL. 0..: [FIGURE If a brand-size must give up almost 5 ABOUT HERE] its entire profit margin in order to be < promoted. make Similarly.S^'^N.29. > for B. respectively.F. Mellow Roast 1 lb.MEL. Figure 5 graphs the optimal promotion policy for each brand-size as a function of the ratio of its unpromoted to promoted margin.FOL. then 1/L.S ^ °*^^' lb. J The driving force here is another manifestation of the "small L. Butternut 1 lb. It is reasonable to expect that promoted profit margin per unit and unpromoted Letting M_.FOL. 1 Since the expected unpromoted level of Folgers lb.. B. in share is.MH. will find promotion profitable is M„. ./M„. For values of Mp.s/^N. expects when it promotes. essentially.L S^^^er more profit by promoting than than 0. consumers since for brands with higher values of L.28 - .S ^ °*'^^' ^ "p.MH.29 is "small") and any of the eight brand-sizes in this study will make less profit promoting than by not promoting. and I respectively.. and it will be more Pj Nj profitable for any of these eight brand-sizes to promote than to not promote.FOL. be the promoted and unpromoted profit margin per unit. then M .. Butternut by not promoting. share is so high.

42 Promotion Profitable Maxwell House 3 lb Promotion Unprofitable .48 Maxwell House 1 lb Promotion Unprofitable .FIGURE 5 : BRAND-SIZE SPECIFIC OPTIMAL PROMOTION 1 POLICIES AS A FUNCTION OF M /M^ / Butternut 1 lb Promotion Unprofitable .54 ^% P Folgers 3 lb Promotion Unprofitable .45 Promotion Profitable VS. Butternut 3 lb Promotion Unprofitable 29 Promotion Profitable VS^ Folgers 1 lb Promotion Unprofitable I Promotion Profitable 1 .29 Promotion Profitable .

29 - . we found that a brand's expected baseline level of unpromoted sales and its expected promotional sales peak are depressed by its own promotions if consumers react to promotions by stockpiling. Investigation of response parameters showed that brands are more likely to gain share due to promotions if they promote more frequently and if they are widely accepted but smaller preference brands. The models were shown to have predictive validity and the parameter estimates were shown to be expected to be unbiased. Eppen and Lieberman 1981 and Fraser 1983) were shown to "get more bang for their promotion buck". Expected sales levels are depressed by competitor's promotions if consumers react to promotions by increasing the probability of choosing the brand on promotion. promoted price to those consumers who would have bought the brand anyway. Smaller preference brands (rather than larger preference brands as suggested by Blattberg. Promotion becomes profitable if the number of consumers who can be persuaded to switch to the promoted brand is sufficient to make up for the profit foregone by offering the lower. We also showed that promotion is unprofitable if consumers are not promotion sensitive or if they respond to promotions by stockpiling the brand they would have bought anjrway. Further. Analyses at the level of the retail stores showed that different stores appear to favor different manufacturers and that favoritism also affects promotion effectiveness.SUMMARY AHD CORCLUSIONS Market response to promotion was shown to be a function of competitive promotional activity and heterogeneous' consumers tastes and promotional responses. We aggregated individual level response parameters to estimate market level promotion response. The ratio of a brand's expected promoted share to its expected . Using data from 200 store loyal households in the 1978-1980 SAMI Kansas City scanner panel we estimated individual level models of consumer response to promotion.

display). is more profitable for that brand to promote than to not Among the widely accepted brands. . would have found promotion profitabile despite the fact that it lost long run share due to promotion. This modeling effort should be extended in several ways. manufacturer coupon.^) it promote. on the other hand. Finally. Maxwell House lb. The model form should be adapted to account for differing responses to differing promotional instruments (e.unpromoted share defines its coefficient of promotional leverage.30 - . Butternut 1 lb. It is is interesting to note that the effect of promotion on long run market share If M /M^^ » .. would have found promotion unprofitable despite the fact that it gained long run share due to promotion. would also be useful to estimate the model in different product classes to develop intuition concerning the product class determinants of promotion response. that coefficient was larger for smaller * preference brands. If this coefficient is larger than the ratio of the brand's promoted margin to unpromoted margin (M^/M. 1 these brands. The resulting promotional response profiles could then be correlated It with demographic data. price discount. once again consistent with the small preference brand advantage.44 for independent of the effect of promotion on profitability..g. The model should accommodate increasing levels of market response for increasing promotional offers. the interaction between promotion and advertising in stimulating consumer response should be investigated.

and _. (5) After steps 3-4 have been performed for each of the NENV synthetic purchase occasions. E tt . set of all j. E TT . NENV. = y T . (j«l through 8). (2) Randomly choose a specific number of store environments from the 260 different environments (4 stores x 65 weeks) present in the database. When there is no bias . and tt .APPEHDIX The Bias Test Procedure (1) Choose a y • and set of tt • -'s by drawing a random consumer from the panel of 200. . Estimate the proposed model using all 65 weeks. 1 1 ij ij for .'s the "true parameters" y T • and T tt . . (3) For each of the NENV purchase occasions. Call the resulting y • . calculate choice probabilities for ali eight brands by applying equation (1) to the parameters from step 1 and the promotional environments from step 2. s. ' ij y . = T tt . will be set to different values. . we have a full purchase history. (4) Simulate a brand choice decision by using the choice probabilitaies from step 3 and choosing a random number. There is now The end result is a y^ and a sufficient data to run the MLE procedure. . . E .31 - . The number of 5 environments used. ranging from to 20. .

V. income and other household resource variables do effect promotion sensitivity.(y . if consumer i is "in the market" at time (V^^=l) then she will- only be "in the market" at time t+1 if she wants to restrict her choice at time t to promoted brands brands but finds no brand on promotion t (y • 13.y 11 1 Al-a. 1982. From Robert L. Buesing.-l)) 11 il 1 1 . 1 1 or V.Al^.) yOOTBOTES 1 "Car-Rental Firms Sling it Out with Golf Balls and Tote Bags". December 15.-l)) 1 t That is. December 15. = V. 1 il 1 1 It )(l/(n. It =0. Narisimhan 1984). Wall Street Journal . p. ) or if she is not reacting to promotions at time not "in the market" at time a t (l-y)* 1-V.)) + (l-V.)(l/(n.I3.(l-a. these studies show that education. = 1 It if consumer i is "in the market" at time t t and » if consumer i is not "in the market" at time (i. Frank and Lodahl 1968. though often in complicated ways. we get V. In general.e.32 - .(3. If we solve the above equation for an equilibrium value of v. 1982. Blattberg. Montgomery 1971.t + l It (y ' Then .. is consuming stockpiled product rather than buying more units at time t). 3 Other studies have attempted to characterize promotion sensitive consumers with demographic and psychographic variables (Webster 1965.)) + (l-V. i. Wall Street Journal 2 . 4 To see this let V. Simison's article "Why are Auto Sales Strictly for the Birds? Just Ask Any Pigeon". Massy. therefore = 1) then there is t l/(n.. = ^ l + (n. )+(l-t'. Peacock and Sen 1978. 1 = V.-l) chance that she will exhaust her stockpile in period 1 and return to the market at time t+1.)) il 1 .. page 1..-i. It ^^ consumer i is (V.(l-a. 29.) + (l-f . . (1-Qt. .

1978 to March 12.) 5 The period December 29. that is passed to the algorithm is i the frequency with which consumer chooses brand j. 79 40 2453.: FOOTNOTES (cont.. 7 Relevant n's.'s.0 . Wilcoxon statistics and p values associated with the differences are Brand test n Wilcoxan Statistic p Butternut 1 lb. The initial guess for n . 6 The lack of bias in the estimates of Tr.712 . . Folgers 1 lb. even for small sample sizes probably results from the choice of starting values for the maximum liklihood procedure.0 438. f.. Butternut 3 lb. .000 . 1980 is made up of 65 weeks. We ignore the last week of this data in testing the predictive ability of the model in order to be consistent with the data usage pattern in Guadagni and Little (1983).

10 One might think that the negative coefficient of tt .08) .005 MH . is subtracted ij .27) - (t=-5. BUT. (FOL . .39) -2 Since the constant has been suppressed. IJ Tl . in the regression tt . ) is . - . we cannot report an R value.04) (t — R^ = .009 PROM ..) Imposition of these constraints requires suppressing the constant and 'placing rei predictors tt .89) .37) . This is not. (BUT .. (PROM . .47) (t-.. - . (a surrogate for tt . IJ 3. . however. PROM . - 200/723). - 292/723). FOL. . reported in equation (9) is driven by the fact that from f . = IJ IJ tt . . (MH . and MH with (tt .054 f..261 ii .FOOTROTES (cont. as a predictor and we get: .016 BUT + (t-. We then get f. J ij (t*2. Notice that the magnitudes and signs of the coefficients are unchanged.040 FOL + (t-1.295/723)..93) + (t-2.J (t-.002 FOL (t=-. to obtain the dependent variable.34 - . ij S. .008 + .15) .193/723).001 BUT (t = .025 PROM .02 n = 723 Here we see that the coefficient of relative promotion frequency is still positive and that the coefficient of f.) 9 (cont. -TT.219/723).98) + ^ (t— 18. . . regression (9) substituting f.46) . the We can rerun driving force behind the sign of that coefficient. - . for f.. .012 MH ..

. S.k' . . for This despite the fact that f^j is added to form the dependent -2 R drops dramatically indicating that . ..J [ r. i.) negative. variable.k . .k .-IT -^ . . constraints can be imposed to insure logical consistency. .J i-" S„„^^ PROM. . ' TT . ' I . f .) (cont. advantage in promoting and that smaller and vice-versa. i. BUT.'' 11 n. = S . [ r. .J = "k Number of Store k 1 Observations in Store k 110 362 165 86 ^PROM.J . ij 11 • » 0. is 11 As before.'• I .J FOL = S^^^ FOL. .k . i-. BUT = S„„. I. In store k we will have .k 2 3 4 The coefficients and t-statistics that result when the logically consistent regressions are run are summarized below.J i. is a poor surrogate TT . . . =» i.J I .FOOmiES 10 (cont. f . tends to be smaller when tt . The implication is that small preference brands have an f . . .^ PROM ..

3* 01 3" n n o n '-> I 9 1 a a (0 H 1 0) rr HI < 3* n 9 3* n o o 3 rr 1 n a a (0 .u n a to rr O •^ n O O rr 3 o rr n JO O — rr N • • r> O O »^ rr II O *• rr II O Ul I-" O N3 I O 3 3 C (^ n (0 U) a N9 a.

12 We present expected levels of promoted and unpromoted purchase share rather than expected levels of promoted and unpromoted sales to avoid distortion due to seasonality.) FOOTNOTES ( cont .37 - . .

"Determinants of Coupon Effectiveness. . "Market Segmentation Using Models of Multidimensional Purchasing Behavior. 19-32. and Subrata K. "A Theoretical and . 3 (Summer). "A Dealing Strategy to Maximize Profits. pp. Robert C." Journal of Marketing (July). Advances and Practices of Marketing Science Proceedings of the 1983 ORSA/TIMS Marketing Science Conference. pp. and Subrata K. Little (1983). Thomas Buesing. Robert C. . Deal Restriction on Brand Switching. "Consumer Response to Promotional Deals. Karl W." Working Paper. pp. . Sloan Management Review (Fall). Tybout and Brian Sternthal (1978). John D. Sen (1976). . Sen (1974). . Brown." Marketing Communications (August)." Journal of Marketing Research 19 (August). Robert. (1981). University of Rochester. "Increase in Sales Due to In-Store Display." Journal of Marketing Research 15 (August). Michel and Ronald C. Peter Peacock.. Robert C. and Emerson M. Guadagni. 157-164.38 - . "Market Segments and Stochastic Brand Choice Models. pp. Russell D. No. Little and Robert L. 288-301.C. Blattberg. and Subrata Sen (1978). pp. 17-28. Babb (1978). Blattberg.. Peter Peacock. ." Working Paper. Chapman. . 38 (October).C. "Sales Response to Promotions and Advertising. . Chevalier. pp. Bowman. Robert C. Gary D. and R. Alice M. 33-45. 109-113." in Fred S. "Identifying the Deal Prone Segment. "MC's Second Annual Report on Advertising and Promotion Expenditures. ed. R. 203-238. Blattberg." Journal of Marketing Research 13 (February).BIBLIOCTAPHnf Blattberg.G. Robert George (1974). pp. Sen (1978). "Exploiting Rank Ordered Choice Set Data Within the Stochastic Utility Model." Marketing Science Vol. and Subrata K. B. Ohio State University. "Purchase Segmentation: Beyond Aluminium Foil. Dodson. "Impact of Deals and 15. . Irons. Curhan (1976). 426-431. C. Peter and John D. pp. 369-377. Joe A. pp. University of Southern California. "Retail Promotions as a Function of Trade Promotions: A Descriptive Analysis." Journal of Marketing Research 12 (November). Cynthia (1983). Zufreyden. Cotton. 2. Eppen and Joshua Lieberman (1981). Empirical Evaluation of Price Deals for Consumer Nondurables " Journal of Marketing 45 (Winter)." Journal of Marketing . 33-39. 116-129. . Chevalier. Staelin (1982). Fraser. pp." Journal of Marketing Research 72-81. Kelin (1983). Michel (1975). "A LOGIT Model of Brand Choice Calibrated on Scanner Data. . Blattberg." Journal of Advertising Research 14 (August).

Massy. . "A Price Discrimination Theory of Coupons." Harvard Business Review 54 (July-August). Lodahl (1968). "Consumer Promotion and the Acceleration of Product Purchases. (1984). "Logically Consistent Market Share Models II." Journal of Consumer Research 9 (December).BIBLIOGRAPHY (cont. Neslin. A Theoretical Analysis . Luce. and Ronald E. Pessemier (1982)." Working Paper. Shoemaker (1983). 70-78. Robert W. "Consumer Characteristics Associated with Dealing: An Empirical Example. Massy. and William C. . Rothschild. 118-120c . William F. ." Journal of Advertising Research 17 (April). pp. Michael L. Gaidis (1981). Frank (1965). pp. Strang." Journal of Marketing Research 13 (August). Shoemaker. Lodish (1984). . 128-147. p." Working Paper #124. pp. "PROMOTER. Scott A. and Doyle L." Journal of Marketing Research 13 (August). Faulty Management. McGuire. 296-302. Carnegie Institute of Technology. pp. Management Decision Systems." Journal of Marketing 34 (April). Carl-Magnus (1971). Scott. 26-34. "The Effects of Trial and Incentives on Repeat Purchase Behavior. 263-269. Seipel. Robert Shoaf (1977). pp." Journal of Marketing Research 8 (February). Timothy W. (1976). Dartmouth. Scott A. . Ronald Frank and T. pp. "An Analysis of the Dynamics of Consumer Behavior and its Implications for Marketing Management. "Variety Seeking Behavior: An Interdisciplinary Review. . pp. William. .39 - . . "Sales Promotion: Fast Growth. . and F. 171-185. 361-388." Journal of Marketing Research pp. Neslin. Montgomery. (1976). "Behavioral Learning Theory: Its Relevance to Marketing and Promotions.Forgotten by Theory. 311-322. "Repeat Rate of Deal Purchases. University of Pennsylvania Press. McAlister. David (1971). pp. "Short Term Price and Dealing Effects 2 (May). Roger A. A. 45 (Spring). Individual Choice Behavior: New York: John Wiley. "A Model for Evaluating the Profitability of Coupon Promotions. in Selected Market Segments. and Robert W. Leigh and Edgar A. Magid M." Marketing Science 3 (Spring). and Leonard M. 47-53.. C. "Premiums . . Weiss (1976). ." Marketing Science 2 (Fall). Carol A. an Expert Promotion Evaluation System. Purchasing Behavior and Personal Attributes Philadelphia. Tuck School of Business Administration. R.D dissertation. Narasimham. John Quelch and Caroline Henderson (1983)." Journal of Marketing . (1958). Duncan (1959). Keuhn.) Ibrahim." unpublished Ph.

pp. 186-189. 2 (May).. Journal of Oc5k / . Journal of Advertising .40 - .) and Advertising. Webster.BIBLIOOtAPHY (cont. pp. "Salea Effects of Promotion Research 18 (October). Frederick E. (1965). Jr. "The 'Deal-Prone' Consumer." Sunoo.S. Don and Lynn Y. 37-40." Consumer Research. Lin (1978).

//'/^-^ MH LIBRARIES 3 TDflD DD3 D D 354 .

.

.

BAStMlflylt^e D ue JUN 8 1990 Lib-26-67 .

jD0Lrcc3d«- 1 5 ov\ yd(Xu\ co^c~v^^ .