Full Text 01

Modeling the Impact of
Sales Promotion on Store Profits

STOCKHOLM SCHOOL
OF ECONOMICS
THE ECONOMIC RESEARCH INSTITUTE
FOUNDATION FOR DISTRmUTION RESEARCH
Address: Sveavagen 65, Box 6501, S-113 83 Stockholm. Tel. +468-7369000
Basic Orientation
The Economic Reserach Institute, at the Stockholm School of Economics (EFI)
is devoted to the scientific study of problems in management science and
economics. It provides research facilities for scholars belonging to these dicipli-
nes, training advanced students in scientific research. The studies to be carried
out by the Institute are chosen on the basis of their expected scientific value and
their relevance to the research programs outlined by the different sections of the
Institute.
Research Divisions
A Management and Organization

B Accounting and Managerial Finance
C Managerial Economics
CFR Center For Risk Research
CHE Center For Health Econonlics
D Marketing, Distribution and Industry Dynamics
ES Economic Statistics
FI Finance
F Public Management
lEG International Economics and Geography
I Information Management
P Economic Psycology
S Economics
T Industrial Production
Research Programs:
Man and Organisation
The Foundation for Distribution Research
Policy Sciences
Additional information about research in progress and published reports is described in our project
catalogue. The catalogue can be ordered directly from the Economic Research Institute, Box 6501,
S-113 83 Stockholm, Sweden.
Modeling the
Impact of Sales Promotion
on Store Profits
Per-Goran Persson
•
i:I~STOCKHOLMSCHOOL
\"¥i~J' OF ECONOMICS
~.-¥-'~ THE ECONOMIC RESEARCH INSTITUTE
FOUNDATION FOR DISTRffiUTION RESEARCH
A Dissertation for the
Doctor's Degree in Philosophy
Stockholm School of Economics 1995
© EFI and the author 1995

ISBN 91-7258-396-7
Graphic Systems i Stockholm AB, 1995
Keywords:
Retailing
Sales Promotion
Profit Measurement
Promotional Effects
Scanner Data
Distributed by:
The Economic Research Institute at the Stockholm School of Economics,
Box 6591, S-113 83 Stockholm, Sweden
To Mother and Father
PREFACE
This report is a result of a research project carried out at the Foundation for Distribution
Research at the Economic Research Institute at the Stockholm School of Economics. It is
part of a larger project concerning how the information potential in supermarket scanner
data can be used in decision making.
This volume is submitted as a doctor's thesis at the Stockholm School of Economics. As

usual at the Economic Research Institute, the author has been entirely free to conduct and
present his research in his own ways as an expression of his own ideas.
The Institute is grateful for the financial support provided by the sponsors of the
Foundation for Distribution Research.
The present volume would not have been possible without the cooperation with the owners
and personnel of the leA store where the data were collected. The Economic Research
Institute wishes to warmly thank the store for their generosity and openness.
Stockholm, April 1995
Claes-Robert Julander
Professor,
Director of the Economic Research Institute at
the Stockholm School of Economics,
ACKNOWLEDGEMENTS
I am grateful for all the support that I have received during the work with this
dissertation; without it I would never have been able to finish this book.
First I would like to thank Dr. Nils-Erik Wirsall: this research would not have been
possible without your drive and vision to build up the Foundation for Distribution
Research, now a lively research institute with ten full time researchers.
The generous financial support from the sponsors of the Foundation for Distribution
research is gratefully acknowledged. I am grateful to ICA for financial support and for
providing the store where we collect the data. I would especially like to thank the owners
of the store and their personnel for their friendly cooperation.
My thesis committee has guided me through my work in an admirable way. Thank you
for always taking time to answer questions and discuss the thesis, you have helped me a lot.
I would especially like to thank Professor Per Davidsson for valuable insights concerning
the structure and the purpose formulation. Professor Per-Olov Edlund has shared his
profound understanding of statistical matters, and has helped me to correct errors in the
formulas.
Professor Claes-Robert Julander, chairman of the thesis committee, has contributed
greatly to this research by initiating the receipt data project and by generating many of the
ideas that are central to this thesis. Thank you Claes-Robert for being such a great coach.
My colleagues and friends at the Foundation for Distribution Research have been a great
support. I am grateful for your help and friendship. Thank you very much Anne, Carina,
Jonas, Mikael, Odd, Per, Richi, and Ulla. I am also thankful to all who participated in the
data collection.
Magnus Eliasson, friend and former colleague, now at ICA: I very much appreciate your
advice and your detailed comments on earlier drafts of this book. Tamara Carlin at the
Professional Communications Skills improved the language of the book. Rune Castenas
helped me to make it to a book.
My wife and closest friend, Ruth, has helped me more than anyone else. I am grateful
for all your support and want you to know that you have helped me more than you believe.
Stockholm, April 1995
Per-Goran Persson
TABLE OF CONTENTS
1. INTRODUCTION
1.1. The Problem 1
1.2. What is Sales Promotion? l
1.2.1. Defining Sales Promotion 1
1.2.2. Types of Sales Pron10tion 2
1.3. Why is there Sales Proll10tion? 4
1.3.1. Sales Promotion Objectives 4
1.3.2. Economic Rationale for Sales Promotion 5
1.4. Planning Sales Promotion 7
1.4.1. Sales Promotion Planning Process 8
1.4.2. Store Level Promotional Strategy 9
1.4.3. Selecting Items to Promote 11
1.4.4. Deciding the Promotion Frequency 12
1.5. Importance and Growth of Sales Promotion 14
1.6. Complex Planning Problem 15
1.6.1. Unknown Profit Impact 15
1.6.2. Easy in Principle 16
1.6.3. Impossible in Practice 17
1.7. Purpose and Delimitations 18
1.7.1. Research Purpose 19
1.7.2. Delimitations 19
1.7.3. Steps to Achieve the Purpose 20
1.8. Organization and Overview of the Report 21
2. SALES PROMOTION AND THE CONSUMER

2.1. Introduction 25
2.2. Implicit Decisions in Grocery Shopping 26
2.3. Sales Promotion and the Store Choice 27
2.4. Sales Promotion and the Category Choice 31
2.5. Sales Promotion and the Purchase Quantity Choice 33
2.6. Sales Promotion and the Brand Choice 34
2.7. Combining the Decisions 36
2.8. Dynamic Effects of Sales Promotion 38
2.9. Summary 40
3. PROMOTIONAL SALES RESPONSE

3.2. Sources of Retailer's Incremental Item Sales 41
3.2.1. Other Brands 42
3.2.2. Other Categories 43
3.2.3. Other Stores 44
3.2.4. Other Periods 45
3.3. Response Modeling: Direct and Indirect Approach 46
3.4. Brand Sales Response Models 47
3.4.1. Linear Models of Brand Sales 48
3.4.2. Multiplicative Models of Brand Sales 48
3.4.3. Exponential Models of Brand Sales 49
3.4.4. Time-Series Models of Brand Sales 50
3.5. Market-Share Models 51
3.5.1. Attraction Models 52
3.5.2. Examples of Attraction Models 53
3.6. Category Sales Models 56
3.7. Store Sales and Store Traffic 58
3.8. Sllmming Up 61
4. PROMOTION PROFITS
4.2. Profit Impact of Sales Promotion 63
4.2.1. Manufacturer Profits 63
4.2.2. Retailer Profits 65
4.3. Manufacturer Induced Promotions 67
4.3.1. A Normative Framework 67
4.3.2. Trade Promotion Profits 68
4.3.3. Couponing Profits 70
4.4. Retailer Promotions 72
4.4.1. Item Profits 72
4.4.2. Category Profits 74
4.4.3. Store Profits 76
4.5. Sllmming Up 78
5. MODELDEVELOPMENT
5.2. Decomposing Retailer Sales 79
5.2.1. General Considerations and Modeling Approach 79
5.2.2. Selecting the Quantity Measure 80
5.2.3. Decomposing Retailer Item Sales 81
5.2.4. Degree of Cannibalization Defined 84
5.3. Specifying the Response Models 86
5.3.1. Model Assumptions and Constraints 86
5.3.2. Specification of Response Models 89
5.4. Modeling Retailer Promotion Profits 92
5.4.1. Retailer Promotion Profits 92
5.4.2. A Sin1plified Promotion-Profit Model 95
5.4.3. Three Promotion Profit Measurement Levels 97
5.4.4. The Profit Impact of Cannibalization 99
5.5. Maximizing Promotion Profits 101
5.5.1. Profit Maximization and Optimal Deal Discount 102
5.5.2. Display Profits 104
5.5.3. Myopic Optimization 104
5.5.4. Determinants of Optimal Deal Discount. 106
5.5.5. Determinants of the Deviation from Optimal Deal Discount 107
5.6. Summary 109
Appendix 111
5.A.1. Relationship between Cannibalization and Model Parameters 111
6. SIMULATING PROMOTION PROFITS

6.2. Definition of the Promotional Response Models 113
6.2.1. Model Structure and Parameters 113
6.2.2. Sales Response 115
6.2.3. Basket Response 118
6.3. Promotion Profits Measured at Three Levels 122
6.3.1. Level of Analysis, Deal Discount, and Special Display 122
6.3.2. The Effect of a Deal Requirement 126
6.4. Promotion Profits of Four other Response Patterns 129
6.4.1. Scenario 2: Moderate Cannibalization 130
6.4.2. Scenario 3: Store Traffic Generation 133
6.4.3. Scenario 4: Intra-Category Cannibalization and Store Traffic Generation 135
6.4.4. Scenario 5: More Cannibalization 137
6.5. Determinants of the Size of the Deal Discount 139
6.5.1. Level of Analysis, Trade Deal Discount, and Response Pattern 139
6.5.2. The Cost of Myopic Optimization 144
6.5.3. The Effects of a Deal Requirement. 149
6.5.4. Margins and Optimal Deal Discount 150
6.6. Meta Analysis of the Deal Discount and the Cost of Myopia 152
6.6.1. Procedure 152
6.6.2. The Impact of the Level of Analysis 152
6.6.3. Scenarios, Trade Deals, Deal Requirements, and Margins 154
6.7. Conclusions 155
Appendix 157
6.A.1. Gross margins, Trade Deals, and Optimal Deal discount 157
6.A.2. Simulation Results 159
7. EMPIRICAL DATA AND ESTIMATION METHOD

7.1. Illtroduction 161
7.1.1. Purpose of Empirical Illustrations 161
7.1.2. Selection of Cases 161
7.2. Data Sources 162
7.2.1. Managerial Judgment, Historical Data and Experiments 162
7.2.2. Information Technology 164
7.2.3. Scanner Data in General 165
7.3. The Scanner Data Project 167
7.3.1. Data Collection History and Procedures 167
7.3.2. Problems in Data 168
7.3.3. The Database 169
7.4. Description of the Store-Level Data 169
7.4.1. Definition of Variables 170
7.4.2. Data Frequency and Temporal Aggregation 170
7.4.3. Description of Store-Level Variables 171
7.5. Parameterization of the Response Models 175
7.5.1. Selection of Independent Variables 175
7.5.2. Linearizing Transformations before Estimation 176
7.5.3. Estimation Procedure 177
7.4.4. Regression Diagnostics 178
7.5.5. Parameter Significance 179
7.5.6. Modeling Approach and the Null Hypothesis 180
8. EMPIRICAL ILLUSTRATION 1: COFFEE

8. 1. Introduction............................................................................................................ 181
8.2. Description of the Coffee Category 181
8.3. Building the Promotional Response Models 182
8.3.1. Model Building Procedure 182
8.3.2. Initial Tests of the Model 183
8.3.3. Store-Level Response Models 184
8.3.4. Category-Level Response Models 186
8.3.5. Item-Level Response Models 189
8.3.6. Overview of the Regression Results 193
8.4. Deal Discount, Sales Response, and Profit Impact. 195
8.4.1. Item A 196
8.4.2. Item B 198
8.4.3. Item C 200
8.5. Determinants of the Size of the Deal Discount. 202
8.5.1. Level of Analysis 202
8.5.2. Trade Deal and Margins 203
Appendix 208
8.A.1. Summary Statistics 208
8.A.2. Store and Category-Level Time-Series 209
8.A.3. Item-Level Time-Series 211
9. EMPIRICAL ILLUSTRATION 2: CEREALS

9.2. Description of the Cereals Category 213
9.2.1. Selection of Items for Analysis 215
9.3. Building the Promotion Response Models 217
9.3.1. Data Manipulations 217
9.3.3 Store-Level Response Models 218
9.3.4 Category-Level Response Models 219
9.3.5 Item-Level Response Models 222
9.4.1. Item A 230
9.4.2. Item B 232
9.4.3. Item C 234
9.4.4. Item D 236
9.4.5. Item E 238
9.4.6. Item F 240
9.4.7. Summing Up 242
9.5.2. Trade Deal and Margin 245
10. EMPIRICAL ILLUSTRATION 3: PASTA

10.2. Description of the Pasta Category 251
10.2.1. Selection of Items for Analysis 253
10.3. Building the Response Models 253
10.3.1. Data Manipulations 253
10.3.3. Store-Level Response Models 255
10.3.4. Category-Level Response Models 256
10.3.5. Item-Level Response Models 258
10.4.1. ItemA 266
10.4.2. Item B 268
10.4.3. Item C 270
10.4.4. Item D 272
10.4.5. Item E 274
10.4.6. Item F 276
10.4.7. Summing Up 278
10.5.2. Trade Deal and Margins 281
10.5. Conclusion 285
11. CONCLUSION
11.1. Summary 287
11.1.1. Promotion Profit as Change in Gross Profits 287
11.1.2. The Sales Model 288
11.1.3. The Profit Model 289
11.1.4. Myopic Measures and Myopic Optin1ization 291
11.2. What is New? 292
11.3. Generalizations 293
11.4. Limitations 293
11.5. Suggestions for Further Research 294
11.6. Managerial Implications ' 295
REFERENCES 299
1. INTRODUCTION
1.1. The Problem

Sales promotion is frequently used in grocery retailing. It takes up a large share of the
resources grocery retailers spend on marketing. The heavy usage of sales promotion has
recently become questioned by retailers and manufacturers. Despite the large sums yearly
spent on sales promotion, the economics of sales promotion is poorly understood.
Consequently, there is a great knowledge gap to fill, and this study is an attempt to
contribute to this effort.
As will be seen later in this study, measurement of store profits of retailer promotion is
difficult because retailers carry large assortments and sales of the individual items in these
assortments are interdependent.
The aim of this study is to improve our knowledge of the impact of sales promotion on
retailer profits. More specifically, the study develops a framework model for measuring the
impact of sales promotion on store profits. The model is then used in a simulation study
and in three studies using empirical data.
This chapter first defines and describes sales promotion in order to give the reader the
necessary background to better understand the problem. Readers already familiar with the
concept of sales promotion may skip sections 1.2. to 1.6. and continue directly with section
1.7. Section 1.7 states the research purpose and with the relevant delimitations. The chapter
concludes with a description of the organization of the report.
1.2. What is Sales Promotion?

This section defines and describes sales promotion, starting with a formal and general
definition of sales promotion. Then, a typology of sales promotion is presented along with
examples of each type of sales promotion.
1.2.1. Defining Sales Promotion
The literature provides several definitions of sales promotion, some of which are
presented by Blattberg and Neslin (1990). These are:
- 1-
Sales promotion consists of a diverse collection of incentive tools, mostly
short-term, designed to stimulate quicker and/or greater purchase of a
particular product by consumers or the trade (Kotler 1988, p.645).J
Sales promotion is the direct inducement or incentive to the sales force,

the distributor, or the consumer, with the primary objective of creating
an immediate sale (Schultz and Robinson 1982, p.8).2
Sales promotion, deals, and display can be defined under the general
term of 'short-term inducements to customer buying action' (Webster
1971, p.556).3
Sales promotion represents those marketing efforts that are

supplementary in nature, are conducted for a limited period of time, and
seek to induce buying (Davis 1981, p.536).4
Based on these definitions, Blattberg and Neslin (1990) develop their own definition of
sales promotion. They choose to exclude short-term from the definition because of the
dispute concerning the long-run effects of sales promotion. Their definition reads:
Sales promotion is an action-focused marketing event whose purpose is
to have a direct impact on the behavior of the firm's customers.
We shall not consider another definition of sales promotion here. For the present study,
any of the above definiti'ons will suffice. The next section presents different types of sales
promotion.
1.2.2. Types of Sales Promotion

Sales promotion can be classified in three types, depending on the initiator and the target
of the promotion. Retailer promotion and consumer promotion are directed toward the
consumers by retailers and manufacturers, respectively. The manufacturers direct trade
promotion to the retailers. The three types of promotions are illustrated in Figure 1.1.
1Kotler, Philip (1988), Marketing Management: Analysis, Planning, Implementation, and Control, 6th
ed., Englewood Cliffs, NJ: Prentice-Hall.
2S chultz, Don E. and William A. Robinson (1982), Sales Promotion Management, Chicago: Crain Books.
3Webster, Frederick E. (1971), Marketing Communication, New York: Ronald Press.
4Davis, Kenneth R. (1981), Marketing Management, 4th ed., New York: John Wiley.
- 2-
Manufacturer
Trade promotion
Consumer
Retailer promotion
Retailer promotion
Consumer
Figure 1.1. Types ofsales promotion.
A large number of different promotional tools are used by retailers and manufacturers.
Table 1.1 gives some examples of these tools.
Retailer Promotions Trade Promotions Consumer Promotions

Price cuts Case allowances Couponing
Displays Advertising a.IIow ances Sampling
Feature advertising Trade coupons Price packs
Free goods Spiffs Value packs
Retailer coupons Rnancing incentives Refunds
Contests/premiums Contests Continuity progra.ms
Double couponing Financing incentives
Bonus packs
Special events
Sweepstakes
Contests
Premiums
Tie-ins
Table 1.1. Examples of sales promotion tools. Source: Blattberg and Neslin (1990)
-3-
The most important and most frequently used retailer promotions in grocery retai~ing are
price cuts, special displays, newspaper feature advertising, and coupons. Very often,
combinations of two or more promotional tools are used.
1.3. Why is there Sales Promotion?

In this section, we look at why sales promotion exists. First will come a review of some
of the more precise objectives that marketing managers have when using sales promotion.
Then, an overview of some previous research concerned with the economic rationale for
the existence of sales promotion is presented.
1.3.1. Sales Promotion Objectives

Given the definition above, the existence of sales promotion might not seem
problematic; sales promotion simply exists "... to have a direct impact on the behavior of
the firm's customers" (Blattberg and Neslin 1990). However, the definition is a very
general statement of the objective of sales promotion, i.e., what it is supposed to do. To
understand why manufacturers and retailers employ sales promotional tools, we need a
somewhat n10re specific description of their objectives with sales promotion.
Manufacturers and retailers have different-and, to some extent, conflicting-objectives
with their sales promotions. Concerning the manufacturers' objectives with trade
promotion, Quelch (1983) states:
When manufacturers offer trade promotions, they expect that the
financing costs associated with taking additional inventory will persuade
retailers to provide special merchandising support to accelerate product
movement. This might include passing the manufacturer's price reduction
through to the consumer, featuring this price cut in store advertising, and
displaying the product prominently.
Blattberg and Neslin (1990) list the following objectives of trade promotion:
• Inducing retailer merchandising activities
• Retail sales force incentives
• Loading the retailer with merchandise
• Gaining or maintaining distribution
• Avoiding price reductions
• Competitive tool
The general objective of trade deals is to push merchandise through the channel while
consumer promotions are used to pull merchandize through the channel. The
-4-
manufacturers' specific objectives with consumer promotions according to Blattberg and
Neslin (1990) are:
• to increase brand awareness
• to attract new customers
• to increase sales to present customers.
Blattberg and Neslin (1990) state that common retailer objectives with retailer
promotions are:
• to generate store traffic
• to move excess inventory
• to enhance the store's image
• to create a price image.
Rossiter and Percy (1987) distinguish two types of pron10tion action objectives: trial and
usage. From a retailer's point of view, the trial action objective refers to attracting new
customers to the store. The usage action objective refers, according to Rossiter and Percy,
to inducing the present customers to visit the store more often. More frequent store visits
are in themselves unlikely to be the retailer's true objective; they might indeed not even be
desirable if the customers only increase their shopping frequency while their total purchase
amount in the store remains constant. The retailer's usage objective would rather be to
induce the present customers to spend a larger share of their household budget in the store.
This objective may be reached in two ways: (1) the customers may choose the store more
often for their shopping trips; or (2) the customers may spend larger amounts on each visit
to the store. The usage action objective is likely to be the more important action objective
for an established grocery retailer while the trial objective may be n10re important for
recently established stores.
1.3.2. Economic Rationale for Sales Promotion
The existence of sales promotion has been analyzed from a micro-economic point of
view. The literature taking this approach is mainly theoretical, with only occasional
empirical evidence presented as illustration.
Blattberg, Eppen, and Lieberman (1981) find the frequent dealing-behavior in grocery
retailing problematic from a theoretical point of view. Why do retailers offer price deals,
i.e., reduced prices? The common explanation of dealing as a means of increasing store
traffic is not enough. It would lead to reduced profits, since the total consumption cannot
be increased much by dealing. They argue that a prisoners' dilemma situation, where no
retailer can stop dealing as long as the others continue, is not probable; price wars seldom
last long.
-5-
That retailers would use price deals to take advantage of manufacturers' trade deals is
countered with the argument above applied to the manufacturers. The explanation
forwarded by Blattberg et ale is based on differential inventory costs of retailers and
households. If some households have lower inventory costs than the retailer, then dealing is
economically rational because deals shift inventory from the retailer to the households,
reducing the retailer's average inventory of the product. The inventory holding cost that
differs between the retailer and (at least some) households is the cost of storage space.
While storage space is a scarce resource for the retailer, it is, within reasonable limits, an
almost free resource for the household. Blattberg et ale do not claim that differential
inventory costs are the only explanation for deals but they provide a rationale for the
observed behavior.
Using theoretical mathematical models, Lal (1990a; 1990b) investigates why
manufacturers prefer to offer substantial price discounts for a short period and then raise
the price to its normal level. According to Lal (1990b), national firms use sales promotion
to compete with local brands for the price-sensitive, brand-switching segment. If the
switching segment is large enough, it will be optimal for the national brands as a group to
price deal in such a way that there is always one and only one national brand on promotion.
Lal (1990a) shows that such a pattern of price promotions of national brands can
represent long-run equilibrium strategies for those brands in their defense against the threat
from local brands. Lal assumes that the local brands have no loyal customers and therefore
constantly compete for the switching segment.
Wernerfelt (1991) builds a mathematical theoretical model. He defines two types of
brand loyalty; inertial brand loyalty results from time lags in awareness while cost-based
brand loyalty results from intertemporal utility effects. The effects of these types of loyalty
are modeled at the market level. It is found that inertial loyalty leads to equilibrium with
price dispersion. Cost-based loyalty can also lead to equilibrium with price dispersion but
single price equilibria are possible.
Salop and Stiglitz (1977) explained the existence of price dispersion among stores by
showing that differential search costs among consumers can lead to a two-price
equilibrium, where some stores charge a high price and some stores charge a lower price.
Consumers with high search costs will find it more economic to remain uninformed, while
those with low search costs will be informed. The high-priced stores sell only to unlucky,
uninformed consumers while the low-priced stores sell to informed consumers and lucky,
uninformed consumers. Their model, which gives one explanation for the existence of
price dispersion across stores, was used as point of departure by Varian (1980) who used
differential search costs as the economic rationale for sales promotion. According to
- 6-
Varian, firms use price deals to discriminate between informed and uninformed customers.
This explains the observed temporal price dispersion
Salop and Stiglitz (1982) showed that price dispersion across stores or over time may
result from the mere existence of search costs, even if all customers are identical with
respect to search costs. If searching is expensive enough, so that consumers only search
once per period, stores with high prices will sell less than stores charging the lower price.
This is because consumers arriving at a low-price store will buy more than consumers
arriving at a high-price store. The price difference between the two types of stores will
offset the disadvantage of lower sales in the high price stores.
Narasimhan (1984) builds a model where coupons are shown to serve as a device for
price discrimination. The more price-sensitive consumers can be given a lower price by
coupons. Less price-sensitive consumers do not find the effort of using coupons
worthwhile, and pay the full price.
Lazear's (1986) explanation for temporal price dispersion is that the existence of more
than one time period gives the retailer a richer set of pricing strategies. Retailers will find it
profitable to charge a high initial price and deal the product in later periods if the product
did not sell. The economic rationale is that the retailer is uncel1ain about the consumers'
reservation prices.
Mulhern and Leone (1991) note that the retailer's rationale for sales promotion is fairly
different from that of the manufacturer. They state that price bundling, Le., charging one
price for the combination of two or more products, is the economic rationale for retailer
promotions. The argument is that visiting retail stores is costly for the consumers.
Consumers consequently buy a basket, or bundle, of items when visiting a store and prices
of individual items are of less importance than the price of the bundle. Although the
consumer may pick single items from different stores, the cost of visiting more stores
induces the consumer to buy a bundle of items when visiting a store.
In summary, the theoretically oriented research has suggested that the ecomomic
rationales for sales promotion are: (1) differential inventory holding costs, (2) price
discrimination, (3) search costs, (4) uncertain demand, and (5) price bundling.
1.4. Planning Sales Promotion

This section describes the promotion-planning process and some tools used for planning
sales promotions. The description starts with promotion planning as a part of the corporate
planning and proceeds to the planning of promotional tactics and the choice of items to
promote.
-7 -
1.4.1. Sales Promotion Planning Process
The retailer's promotion-planning process should be based on the progression from

corporate objectives through marketing strategy, promotion objectives, and promotion
strategy, to promotion tactics (Blattberg and Neslin 1990). A corporate objective can, for
example, be to capture 20 percent of the market in a certain area. A retailer's marketing
strategy has several dimensions. The most important dimensions of retailer marketing
strategy are: price level, promotional strategy (high/low or everyday low pricing), service
level, variety, focus, and convenience (multiple locations, few locations). The dimensions
combine into marketing strategies (Blattberg and Neslin 1990). The chosen marketing
strategy determines the relevant promotion objectives. Important promotion objectives for
retailers are to generate store traffic, to move excess inventory, to enhance the store's
image, and to create a price image.
The promotion strategy follows from the marketing strategy and the promotional
objectives. Examples of promotional strategies given by Blattberg and Neslin (1990) are to
maintain the promotional to total sales ratio at a certain level, to maintain a certain margin
on promoted products, and to promote items with a broad appeal. The promotion strategy is
implemented by translation into promotional tactics. Promotion tactics relevant to the
retailer are event planning, selection of iten1s to promote, setting of the discount level,
merchandising and feature advertising of the promoted items, and retailer's forward buying.
Event planning refers to the creation of a special event or a theme for the promotion.
The events are typically tied to the seasons and holidays and make it easier for the retailer
to communicate the promotion to the customers. The retailer must then select the item or
items to include in the promotion. For a grocery retailer, the selection of items to promote,
the discount level, and merchandising and feature advertising are influenced by
manufacturers' trade promotion (Chevalier and Curhan 1976; Walters 1989). The retailer
may also use the manufacturers' trade promotions to purchase products that will be sold
after the promotion. This is referred to as forward buying.
Holmberg (1992) describes the promotional planning of the largest Swedish grocery
retailer, ICA. The planning is done at different organizational levels. TV-advertising and
sponsoring are planned nationally for the whole chain, whereas a weekly leaflet is planned
at the regional level. The planning horizon is one year for dry products and six months for
perishables. The dry products' plan is revised after six months. Negotiations with
manufacturers are part of the planning of the leaflet'. At the local level, the newspaper
feature advertising is planned based on the regionally produced leaflet; products are added
and deleted from the plan depending on local agreements with manufacturers. In Swedish
grocery retailing, each department usually receives some space in the feature advertisement
-8-
and in the leaflet. The leaflet and the feature advertisement thus reflect the organization of
the retailer (Holmberg 1992).
1.4.2. Store Level Promotional Strategy
Blattberg and Neslin (1990) present a model for planning retailers' margins by focusing
on markups and markdowns. The model is aimed at helping the retailer plan a promotional
strategy. Blattberg and Neslin define the following variables:
S = total retail sales over the planning period

N = total retail sales for product sold at regular price
D = total retail sales for product sold at reduced price
X = total markup for product sold at regular price
Y = total markup for product sold at reduced price
They then define the average fraction total markup as:
(1.1)
where P is the fraction of total sales made at reduced price, M r is the average fraction
markup on product sold at regular price, and M p is the average fraction markup on product
sold at reduced price. The equation shows that the average fraction total markup is a
weighted average of the fraction markup on products sold at regular and reduced price.
Blattberg and Neslin suggest that the equation is useful for comparing store strategies, for
example, high/low-pricing (Hi-Lo) versus everyday low-pricing (EDLP). They note that P
is dependent among other things on the markups and the frequency of deals. This can be
written:
(1.2)
(1.3)
where F is the frequency of deals and E is the percentage of the merchandise that is
offered at reduced price. The retailer's problem is to maximize the store's profits:
max 1r == S . M t == N . M r + D· M p (1.4)
-9-
If the response functions for Nand D would be known, the profit-maximizing strategy
could be determined. However very few attempts to specify and estimate the response
functions have been reported, according to Blattberg and Neslin (1990). The model is used
in practice to compare store strategies. Table 1.2 shows an example provided by Blattberg
and Neslin of a comparison of two hypothetical supermarkets.
Standard Everyday-Law-Price
Retailer Retailer
Percentage of sales sold at reduced price (P) 20% 100/0
Average regular price markup (M r) 250/0 200/0
Average reduced price markup lM p) -10 % 00/0
Total average markup (Mt= l1-P1Mr+ PMo> 18% 180/0
Table 1.2. Pricing and promotion strategies for two supermarkets
Lodish (1982) uses a model with a similar structure in a marketing decision support
system for retailers. The model was developed for a multi-store retailer to support yearly
planning and resource allocation on a high level. The unit of analysis is an entity, which
might be a store or a product category within a store. Managerial judgment was used to
parameterize the model.
Lodish divided sales of an entity into three parts: (1) regular sales R, (2) price-event
sales E, and (3) m~rkdown sales D. Markdown sales is modeled as a fraction of regular
sales and price event sales. This fraction is modeled as a function of inventory level,
merchandise character, and the markup. Regular sales and price-event sales are thought to
be equally affected by national advertising, inventory levels, and selling space. Competition
and a growth trend are also thought to affect regular sales and price-event sales equally.
Price-event sales are particularly affected by the average price-off percentage, the number
of price events, the merchandise character, price event local advertising, and the average
retail n1arkup. Regular sales are particularly affected by the average retail markup, regular-
price local advertising, and the merchandise character.
The model assumes reference levels for the sales variables as well as for the controllable
and uncontrollable marketing variables. Deviations in any of the independent variables
from their reference level are translated into an effect index by use of a response function.
At its reference level, the effect index of a marketing variable is 1.0. For example, if
regular-price local advertising is expanded 10 percent, the response function may produce
an effect index number of 1.08, meaning that regular sales would be up eight percent. The
effect indices for all independent variables are then n1ultiplied together to give the effect
index for the relevant sales variable. The sales forecast is given as the reference sales level
times the effect index. The multiplicative structure makes the model modular and enhances
- 10-
the adaptivity of the model. New marketing phenomena can easily be included in the model
by creation of a new effect index.
The model produces sales forecasts conditional on the managers' inputs. Cost
relationships are applied to the conditional sales forecasts to provide projected profit and
loss forecasts for marketing plans and scenarios. The projected outcomes of different plans
under various scenarios can then be compared.
Lodish reports that although the system was originally in1plemented using managerial
judgment, it was complemented with a sales and marketing database. The database was
used to improve the knowledge of sales response to the marketing variables and to track the
effectiveness of past decisions.
Hoch et ale (1994) recently investigated the profit impact of the EDLP strategy versus
the Hi-Lo strategy. They found that the EDLP strategy implied lower retailer profits than
the Hi-Lo strategy mainly because the incremental regular sales volume was insufficient to
overcome the negative effect on regular profit margin. However, it should be noted that
their study tested the implementation of the EDLP concept in some product categories,
rather than the whole store.
Mulhern and Leone (1990) used intervention analysis to investigate the effect a change
in promotional strategy had on store sales and store traffic. The strategy change was from a
promotional strategy with many small price cuts to one with fewer but deeper price cuts.
Intervention analysis is a type of time-series analysis where a dummy variable is used to
indicate the intervention. The intervention, in this case, is the shift in promotional strategy.
They found that the new strategy increased store sales but had no impact on store traffic.
In summary, these models provide retail management with useful tools for evaluating
the overall promotional strategy for a store or a chain. The n10dels are less useful for
planning specific promotional activities.
1.4.3. Selecting Items to Promote

Blattberg and Neslin (1990) discuss empirical findings reported by Curhan and Kopp
(1986)5 concerning retailers' selection of products to promote. The seven factors reported
by the retailers as being important when deciding on which items to promote were:
• Item importance
• Promotion elasticity
• Manufacturer's brand support
• Promotion wear-out
5Curhan, Ronald C. and Robert J. Kopp (1986), "Factors Influencing Grocery Retailers' Support of Trade
Promotions," Report No. 86-104, Cambridge, MA: Marketing Science Institute.
- 11 -
• Sales velocity
• Item profitability
" Incentive amount
Blattberg and Neslin note that cannibalization6 and effects on store traffic are not
included in the list. If retailers do not consider these effects when deciding which items to
promote, they may produce less than optimal results.
According to Blattberg and Neslin (1990) most retailers use some type of heuristic for
selecting items to promote. Such a heuristic may start with a classification <)f the items as
high (H), moderate (M), and low (L) volunle items. The retailer then decides on the number
of times a year that the categories should be promoted. H items are promoted frequently,
but are rotated so that given items are not promoted too frequently. The M items are
promoted when the manufacturer offers an attractive deal. M items may also be promoted
to complement a promotion event. The L items are used to fill out the advertisement.
McCann, Tadlaoui, and Gallagher (1990) described a knowledge-based system for
advertising design in retailing. The system decides which items to include in a retailer's
leaflet. The layout of the items within the leaflet is also decided by the system. The system
uses rules of thumb and scanner-collected data to design the retailer's leaflet.
Blattberg and Neslin (1990) summarize the state-of-the-art in the item selection as
follows:
"... item selection is an art form. It is based on the merchandiser's
intuition. A planning model, using decision calculus combined with
statistical models, can be created to simplify the process. Currently,
though, very few retailers use a scientific approach to item selection" ~
(Blattberg and Neslin 1990, p.450).
In summary, when deciding which items to promote, the retail manager currently does
not utilize formalized marketing models to a significant extent.
1.4.4. Deciding the Promotion Frequency
Achabal, McIntyre, and Smith (1990) developed a model for determining the optimal
promotion frequency by maximizing profits from periodic department store promotions.
Their sales-response model for a merchandizing program has a multiplicative structure.
Achabal, McIntyre, and Smith note that the model can be interpreted as base sales level and
four effect indices or lifts. The effects modeled are the price reduction effect, the timing
effect, the inventory effect, and the seasonal effect:
6I.e., substitution within a product line, or, when defined for the retailer, within an assortment.
- 12-
(1.5)
where
S = sales for a merchandise program during one week

So = sales during a normal week for a merchandise program
Po = regular price per item for the merchandise program
P = price per item for the program this week
t = number of weeks since the last promotion of this program
10 = fixture fill (i.e., nominal) inventory level for the program
I = actual (beginning of week) inventory level for the item
d(k) = seasonal scale factor for this week and program
Y,!3, (j = model parameters (elasticities)
This model is interesting and unique because it explicitly includes the effect of the time
between sales promotions. The timing effect is of course important when the optinlal
promotional frequency is determined. They assume that sales in weeks adjacent to the
promotion are not depressed by the promotion.
The gross margin for a department during a week is given by two equations. During
normal weeks, i.e., weeks without a promotion, the gross margin for a department is:
(1.6)
During the promotional weeks, the gross margin is:
G2 (p,! ,/2 ) = (p - c)· S(p,po,/2 ,t)- h· 12 - vo (1.7)
where
c = retailer's marginal cost per unit

h = inventory holding cost per week per unit
/1 = inventory level for normal weeks
12 = inventory level for promotional weeks
! = I-lit = fraction of weeks that promotions are not held
Vo =fixed cost of holding a promotion
The average gross margin per week can be calculated as:
(1.8)
- 13 -
The sales response equation is inserted in the gross margin equation. Achabal, McIntyre,
and Smith (1990) then solve for the set of decision variables that maximizes the average
weekly gross margin.
1.5. Importance and Growth of Sales Promotion

Sales promotion has become institutionalized in Swedish grocery retailing. According to
Holmberg (1992), the largest Swedish grocery retailer, ICA, distributes 217 million leaflets
a year to the 4.5 million Swedish households. A normal-sized ICA-store temporarily
reduces prices of somewhere between 500 and 700 items every week and builds special
displays for around 200 of these items (Holmberg 1992).
The extensive use of sales promotion means that deal-sales represent a large share of the
total sales; in many product categories, deal sales represent half of category sales, while in
some categories, deal sales account for as much as 80 percent of total sales (Hohnberg
1992). Deal sales as a share of sales are especially high in product categories used by the
retailers as traffic builders or loss leaders, i.e., to attract customers into their stores.
Sales promotion has large short-run effects on brand sales in a store. Previous studies
show that weekly brand sales increase ten to fifteen times the weeks when the brand is
promoted. The large-short nln effects imply that retailer support for promotions is essential
for the brand's sales. This has led to fierce competition among brand managers to obtain
retail support for their promotions.
Sales promotion takes a large and growing share of the marketing budget. Spending on
sales promotion now exceeds image advertising expenditures, according to Blattberg and
Neslin (1990). The rapid growth of sales promotion has recently lead to a counter reaction,
with a gro\ving skepticism concerning the benefits of the system. Manufacturers find that
the system is too expensive, that it hands over the control of the marketing budget to the
retailers, and they fear that sales promotion erodes brand loyalty. They are also concerned
that much of their promotional efforts are absorbed by the retailers and not passed through
to the consumers. Procter & Gamble recently reduced their reliance on sales promotion
drastically (The Economist 1992).
Retailers are also starting to question their use of sales promotion. New retail formats
have entered the market with less reliance on special offers. The everyday-low-pricing
(EDLP) strategy is an example of such a counter strategy. Stores adopting EDLP set regular
prices somewhere between the inflated regular prices and the low-deal prices in Hi-Lo-
pricing stores. While the Hi-Lo-pricing stores switch between high and low prices, the pure
EDLP pricing stores never offer price deals.
- 14-
This counter-trend should be seen from the perspective that the use of sales promotion
has reached a very high level. There is no evidence that manufacturers and retailers would
actually stop using sales promotion. As Hoch et ale (1994) note, even the stores that have
implemented the EDLP-concept use sales promotion rather frequently. Sales promotion as
a topic, thus, continues to be of great importance in retailing.
1.6. Complex Planning Problem

This section proposes that the effects of sales promotion are not fully known and that
methods for measuring the profits of retail promotions are missing. It is also argued that
although the problem is theoretically simple to solve, the large assortments of grocery
retailers make the solution far from trivial in practice.
1.6.1. Unknown Profit Impact
One would assume that the effects of such a widely-used tool as sales promotion would
be well understood. That is however not the case, and there are important knowledge gaps
to fill. Two areas that have been especially neglected are the profit impact of sales
promotion and the retailer perspective. This is problematic because an understanding of the
profit impact of sales promotion is the key to effective planning and use of sales promotion
(Hoofnagle 1965).
The lack of research taking the retailer perspective is disturbing because the retailers and
manufacturers have different objectives. As Rossiter and Percy (1987) put it, while the
manufacturer wants the consumer to buy his brand in any store, the retailer is satisfied if
the consumer buys any brand in his store.
Not only do retailers and manufacturers have different objectives with sales promotion,
the effects of sales promotion also differ between the two groups. The retailer perspective
makes cross-elasticities important, because when cross-elasticities are high, the effects on
store performance are very different from the effects on brand performance. Previous
research has found cross-elasticties to be large in grocery retailing. Bultez et al. (1989)
state:
A retail business is continuously confronted with a need to monitor the
interdependencies generated within its multiproduct assortments. The
word "cannibalism" denotes retailers' concern for multiple forms that
substitution effects may take within their departments: between brands,
either within or across variety-types, and vice versa, between variety-
types, either within or across brand lines. (Bultez, Gijsbrechts, Naert,
and Vanden Abeele 1989)
- 15 -
The neglect of the retailer perspective may be a reflection of the traditional power
balance between manufacturers and retailers. Most models of sales promotion encountered
in the research literature assume a brand manager perspective.
It is believed that the rapid development of information technology will shift the power
balance towards the retailers. However, the retailers need knowledge to harness the power
of the information potential in their scanner systems. Once retailers have this knowledge,
manufacturers will also find that understanding retailers' profits from sales promotion is
important. With such knowledge, the brand manager may craft deals that are profitable for
the retailers and the manufacturer. Without that knowledge, he n1ay find sales go to
competitors with a better understanding of retailers' needs.
1.6.2. Easy in Principle
Kotler (1991) describes the theory of effective marketing resource allocation, an

application of standard micro-economic theory to the brand n1anager's decision problem.
The theory development starts with the definition of the profit equation, where the profits
are the product's revenues minus its costs. The next step is to specify the unit sales volume
as a function of variables under control by the firm and variables not under the firm's
control. Assume that the sales equation is known, Le., we know the relevant variables, their
functional relation to unit sales volume, and the numerical values of the parameters in the
equation. Then the unit sales volume can be estimated for any combination of the
controllable variables. The result can be inserted into the profit equation to estimate the
profit of the con1bination of the firm's controllable variables. We then have a marketing-
plan simulator (Kotler 1991).
A marketing plan simulator allows the marketing manager to simulate the outcomes of a
large number of possible marketing plans without actually having to execute the plans. The
manager then chooses the plan with the most appealing outcome. The marketing-plan
simulator can also be used to find the plan that maximizes profits.
Interesting results can be obtained from the n10del even if the unit sales volume equation
is not known. A general result is that the marketing budget should always be allocated in
such a way that the marginal profit of a marginal dollar spent on a marketing tool is equal
for all marketing tools. Profit maximization implies that the marginal profit for each tool
equals zero.
These results generalize to the case when the firm sells more than one product. If the
unit sales volumes of the firm's products are independent of each others' sales (with the
exception of budget interdependencies), then the marketing budget should be allocated
over the products and marketing tools so that the marginal profits are equal for all tools and
products. On the other hand, if unit sales volumes of the firm's products are interdependent,
- 16 -
e.g., if they are substitutes or complements, then the problem is more complex and is
treated in models of product-line pricing (see e.g., Rao 1984; Reibstein and Gatignon
1984).
The retailer's problem is considerably more difficult than the brand manager's. As noted
by Doyle and Saunders (1990):
For the single brand, the optimum level of advertising depends upon its
margin and the advertising elasticity of demand. But for a retailer's
product three additional factors are important: (1) the advertising
support offered by the manufacturer, (2) the cross effects of the
advertised products on other items in the retailer's assortment, and (3)
the impact on overall store traffic (Doyle and Saunders 1990).
The retailer who wants to maximize store profits from sales promotion needs to find the
promotion portfolio that maximizes gross profits. The gross profit from a promotion
portfolio is calculated in three steps. First, conditional unit sales estimates for all items are
needed. Second, each unit sales estimate is multiplied by the gross profit of that item.
Third, the sum is computed, and the profit-maximizing portfolio is chosen. Easy in
principle, impossible in practice, as we will see.
1.6.3. Impossible in Practice
The conditional sales forecasts are computed from a matrix of direct- and cross-
elasticities and the promotion portfolio. Some calculations will show the enormous size of
that matrix. We assume that the retailer carries 10,000 items, which is the approximate
number of items carried by the average Swedish supermarket. Our retailer uses only one
type of sales promotion (to reduce the number of cross-elasticities). Assuming that we
know the correct model specification, we only have to estimate the cross-elasticity matrix.
For each item in the assortment, we need to estimate 10,000 cross-elasticities. Taken times
the number of items, we have 100,000,000 (100 million) elasticities to estimate. If the
retailer calculates one elasticity per second the job would take a little more than 3 years. A
computer calculating, say, one thousand elasticities per second would need a little more
than a day.
The advent of new advanced computer technology brings many promises and hopes.
Computers are becoming faster and faster and their storage capacity larger and larger. But
fast computers with almost unlimited storage capacity are not enough.
The problem is that we only get 10,000 observations per time period. If we have weekly
observations it would take 197 years of data collection to accumulate the 100 million
observations which would be the absolute minimum. Using daily observations would speed
up the data collection considerably, but it would still take more than 27 years to accumulate
enough data for estimation. We could just hope that the elasticities would remain constant
- 17 -
over that time. This example shows that the problem is lack of data more than lack of
computing power. Not even the fastest computer can do anything about that.
We have two alternatives: the first is to get more observations per time period, the
second is to reduce the nUIuber of elasticities we need to estimate.
More observations per time unit can be obtained in two ways. We could shorten the time
period of the observations, e.g., use hourly observations instead of daily observations. With
hourly data, we would get ten or twelve times as many observations per year. The data
collection would still take more than two years, but that might be acceptable.
Multicollinearity would, however, be a problem if the promotion schedule is not
accelerated.
More observations per time unit can also be obtained by pooling data from different
stores. A chain with 100 stores collects 100 million observations in 100 days. Pooling gives
more observations per time period but assumes that the elasticities are constant across
stores. In other words, we constrain the elasticities.
Our second alternative is to reduce the number of elasticities we need to estimate. The
number of elasticities to estimate can be reduced by assuming that the cross-elasticity
matrix follow a certain pattern or model. There are theoretical reasons to believe that such
patterns exist. For example, it has been found that the more similar two products are, the
closer substitutes they are and the higher cross-elastisities they have. The problem is to find
the "right" pattern, i.e., a theoretically sound explanation for the pattern that corresponds to
the facts. To find the right pattern might not be so easy either, as Lodish (1982) points out:
"...some managers thought that if Entity A would have sales greater than
reference, then Entity B's sales should also go up because of increased
store traffic. Other managers thought just the opposite because of
cannibalization between the two entities. ... Without consideration of
these synergistic effects, many of the marketing activities would not be
profitable" (Lodish 1982).
Difficult or not, the combination of theory-based models and empirical data is powerful
and is used in most fields of study to advance scientific knowledge. There is no reason why
studies of promotion profits in retailing should not benefit from that combination.
1.7. Purpose and Delimitations

Sales promotion has now been described and the knowledge gap concerning the profit
impact of retailer promotion has been identified. This section defines the research purpose,
makes the necessary delimitations, and outlines the steps that will be taken in order to
fulfill the purpose.
- 18 -
1.7.1. Research Purpose
The overall purpose of this study is to investigate the mechanisms by which retailer
sales promotion influences store profits. In more specific terms, the purpose is to develop a
model for measurement of the profits of ret(\iler promotions. Along with the academic
interest in structuring the thinking about the profits of retailer sales promotions, it is hoped
that the present research shall help retailers to make sales promotion more profitable. The
model is supposed to be useful as tool for the retail managers for planning and control of
sales promotion.
A further specific purpose of this work is to use the promotion profit model developed
to:
• Examine how retailer profits from sales promotions and depend on factors such as
the deal size, gross margins, sales response and cannibalization, and the trade deal.
• Define how to maximize promotion profits, identify important determinants of the

optimal deal size as well as to show how the profit-maximizing deal and the
maximum profits depend on these factors.
• Demonstrate the difference between retailer-oriented promotion profit

measurement and item-oriented promotion profit measurement, and how the
measurement level affects promotion profits, optimal deal, and profits from
optimal de'al.
To clarify, it should be noted that the present research is not an attempt to evaluate
whether sales promotion is used in a profitable way in retailing. The emphasis is on the
development of a model for measuring promotion profits to understand how promotion
profits are affected by a number of factors. The model developed here may then be used in
future research to evaluate whether current usage of sales promotion is profitable or not.
Further, the study assumes a retailer perspective. This means that the interest here is the
promotion profits for the retailer. Manufacturers' profits are thus not a topic of the present
study. However, it will however be argued that efforts to maximize manufacturers' profits
from sales promotion must be based on an understanding of the retailers' promotion profits.
1.7.2. Delimitations
The following four delimitations were made to make the study feasible: First, although
the model developed may be useful also for other types of retailers, the emphasis is on the
- 19 -
grocery retailer. Grocery retailing differs from many other types of retailing in that
consumers make frequent purchases of baskets of relatively low priced products.
Second, the model is meant to be used to evaluate the promotion profits from individual
promotions. This means that the model is limited to the measurement of the effect of rather
small changes in the store's promotional mix. There are two reasons for this delimitation,
one related to the intended use of the model and one more pragmatic or technically
determined. The model is intended to measure the profitability of individual promotions,
rather than of the entire promotion portfolio. This is the everyday problem faced by
retailers and forms the basis for a systematic approach to deal design. Design of individual
promotions is supposed to be done within the frame of the store's promotional strategy. On
the pragmatic and technical side, the small changes are more frequent than changes in the
promotional strategy, implying that models evaluating small changes would be more
relevant as a tool in the daily operations. The low frequency of strategy changes also limits
the usefulness of statistical models in the design of promotional strategy, though, statistical
models can be used ex post to evaluate the effect of a change in promotional strategy.
Third, the model considers short-term effects. This excludes long-term and dynamic
effects of promotion. The reason is that long-term effects can be assun1ed to be small for
the type of promotions studied. If long-term effects are negative, as when future sales are
cannibalized, the focus on short-term effects will give a positively biased view of
promotion profits. It will later be argued that this effect should in general be small for the
retailer. The negative bias that could occur as a consequence of positive long-term effects
is also likely to be small since future store traffic hardly would be expected to be
influenced.
Fourth, the model developed here is positioned as a research model. It is also aimed to
be a basis for a decision support system for the retailer, while it is not a ready-to-use
version of such a system. That would make implementation and information system aspects
too dominant in this report. Although the model is not supposed to be an implemented
system, it should be implementable. This means that it should use data available to the
retailer, i.e., scanner data. While the level of detail must be possible to handle, it should not
be too abstract. Another facilitator for implementation is that the model is easy to
understand, ideally intuitively appealing.
1.7.3. Steps to Achieve the Purpose

Four steps lead to the fulfillment of the purposes described above. The first step is a
review of previous research about sales promotion presented in the literature. The literature
review serves as a platform for the second step, the model development, in that it presents
the existing knowledge about sales promotion as well as identifies the missing pieces. The
- 20-
second step is the development of a model for measurement of promotion profits in
retailing. The model consists of two parts: a sales response model and an income and cost
accounting model. The third step is to use the model of promotion profits in a simulation
study to illustrate the profit impact of sales promotion for hypothetical model parameters.
The fourth and last step is to apply the model to empirical data from three product
categories in a store.
In short, the steps can be described as:
• literature review
• analytical and theoretical work
• simulation
• empirical illustrations
) 1.8. Organization and Overview of the Report

This report is divided into eleven chapters and a list of references. Although the reader
should proceed in a linear path through the chapters from chapter one to chapter eleven, it
may be useful to note that there are groups of chapters with similar function. This structure
is illustrated by Figure 1.2.
Literature Review
Analytical and
Theoretical Work
Simulation
Empirical
Illustration
Figure 1.2. Structure of the report.
- 21 -
Chapter 1 identified the problem, defined and described sales promotion, and specified
the purpose of the research as well as its delimitations.
The following three chapters present models and empirical results of previous research,
mainly concerning sales promotions. Chapter 2 discusses the decisions the consumer
makes when shopping as well as the effect sales promotion has on these decisions. The
purpose of chapter two is to see the consumer behind the aggregate sales-response to sales
promotion.
In Chapter 3 and Chapter 4, the retailer's perspective is assumed; Chapter 3 is devoted to
promotional sales response models while Chapter 4 concerns sales promotions and profits.
Promotion profits are defined, and methods for measuring and evaluating past promotions
are reviewed. Then some models for planning future sales promotions are presented. It
should be noted that some overlapping between these three chapters could not be avoided.
The overlaps between Chapter 2 and Chapter 3 are natural because these chapters
examinine the same phenomenon from different perspectives. The parallels beteen the
consumer's decisions and the retailer's sales are intended to contribute to our understanding
of the promotional effects. The overlaps between Chapter 3 and Chapter 4 are different;
they occur because all profit models contain sales response models.
Chapter 5 develops the model for planning and evaluating sales promotions. First, a
sales response framework model is developed. Then, it is used in the development of a
promotion profit model.
Chapter 6 illustrates the use of the promotion profit model using simulated model
parameters. This chapter can be seen as a blueprint for analyzing the profits of sales
promotions. Important in this chapter is the decomposition of promotion profits and the
three levels at which the retailer can measure promotion profits.
The empirical part of this research starts with Chapter 7, which describes the empirical
data used and discusses some methodological aspects of scanner data as well as some
statistical considerations of great importance in the empirical work. Detailed descriptions
of the procedures used in the empirical studies are given in connection with the
presentation of the empirical results.
Chapter 8, 9, and 10 serve as illustrations of the methods as well as tests of the
hypotheses and models.
Chapter 8 applies the model to the coffee category, probably the most analyzed product
category. Chapter 9 and Chapter 10 are applications of the model to data from the breakfast
cereals and the pasta categories. In these chapters, the method description is cut back to a
minimum.
- 22-
Chapter 11 summarizes the conclusions, discusses what can be generalized and what
cannot, and identifies limitations of the research. Suggestions for further research are given
as well as some managerial implications of the present work.
- 23 -
2. SALES PROMOTION AND THE CONSUMER
2.1. Introduction
This chapter provides a framework describing the decisions implicit in the consumers'
shopping behavior. This framework of implicit decisions links consumer behavior and
retailer's sales as will be seen in the next chapter. First there will be a brief discussion of
the decisions consumers have to make when shopping. Then, the empirical evidence of the
effect of sales promotions on the consumers' decisions is discussed.
Meal production is an important task in n10st households. The well-being or utility of
the household is influenced by the organization of the meal production process. Meals must
meet taste as well as nutritional requirements. To meet such requirements the meals not
only have to be of good quality but also must ensure sufficient variation over time to avoid
boredom. Differences in taste preferences among the household members and changing
preferences complicate the planning of meals. The complex task of meal production takes
place in an environment where time and money are often scarce resources.
The household uses groceries as inputs in the meal production process. Most occidental
households purchase groceries in supermarkets as packaged and branded products. A meal
usually consists of a number of products all of which have to be available at the
consumption occasion. Logistics and inventory management considerations imply that the
household purchases a bundle of products when shopping.
The prices of the individual products in the bundle that the household plans to purchase
may vary across stores. The sum of the prices of the bundle of products may be minimized
by purchasing each product in the store charging the lowest price. However, visiting a store
costs time. The cost in terms of shopping time, for obtaining the lowest prices for all
products in the bundle can be considerable, especially when prices change over time. A
household therefore visits few stores at each shopping trip and buys bundles of products in
the stores visited. The purchases can be made faster when the shopper is familiar with the
store and the brands in the store.
- 25-
2.2. Implicit Decisions in Grocery Shopping
To explore the household's shopping behavior we use the following fictitious example.
It is ten minutes past seven on a Wednesday evening and Sven Svensson has just paid for
his purchases at the grocery store ICA Baronen in Stockholm. He bought two liters of low-
fat milk from Aria, three cans of Pripps Bid beer, 0.358 kilograms of cured meat, one
bottle of Heinz tomato ketchup, 500 grams of bread, a Mars candy bar, and the evening
newspaper Expressen. What are the decisions implicit in this description? As we shall see,
an ordinary shopping trip as the one described here implies a long list of implicit decisions
made by the shopper. Our shopper has decided on: the timing of the shopping trip; the store
where the purchases are made; to purchase beer on this shopping trip, specifically Pripps
Bid in cans, and in total three cans. Similarly, he decided to buy candy-Mars, one bar-in
this particular store on this shopping trip. The store choice, category choice, brand and
variety choice, quantity choice, and timing decision are all implicit for all the products in
Sven's shopping basket.
Did he really make all these decisions for each product? Probably not explicitly. As
Sven is only a fictitious character we can follow his thoughts and actions in detail. This
morning, as Sven had breakfast, he ran out of bread to toast. He also recognized that only a
little milk was left in the last carton in the fridge and that he needed something for dinner.
Thus, he decided to shop on the way home from work. Sometimes he does some shopping
during the lunch break, but he does not like to have the things in the office the whole
afternoon. Walking home through the city, he passed a couple of supermarkets, but he
preferred to shop in one of the stores closer to his home. Then, he did not have to carry his
purchase so far. He chose ICA Baronen over the Konsum one block away despite its higher
price image mainly because he was on that side of the street. Also, he did not intend to buy
very much.
Sven took a shopping basket and went into the store. He placed two liters of low-fat
milk from Aria in the basket. He always buys low-fat milk, and Aria is the monopolist. He
bought two liters because one liter is too little and three liters is too much both because it is
too heavy and he might not consume all before it becomes sour. He found the bread and
chose one with sesame; not that you taste much of the sesame, but he just liked it better. He
bought one loaf because that usually is enough and there were no price deals this week in
this store. Looking for something to have for dinner, he found that the cured meat was on
sale. Why not make some spaghetti bolognese, Sven thought, and took a pack containing
0.358 kilograms. He knew that he had enough pasta at home so he did not buy any.
However, he did not remember how much tomato ketchup he had left so he decided to buy
some, just in case, and put a bottle of Heinz in his basket. Two bottles would obviously
- 26-
have been too much. He could need some beer, also. He put three cans of Pripps Bla into
the basket without much reflection.
Sven would have bought coffee had there been a deal on Classic or Gevalia, because the
Classic package he bought three weeks ago was only half full last time he made coffee.
This week Luxus coffee was on sale in ICA Baronen but Sven did not buy it.
Sven buys the evening newspaper three or four days a week on average. When waiting
in the check-out line, he was looking at the headlines and decided to buy the paper. As
usual, he bought Expressen. There was a special display for Mars next to the check-out.
Sven felt some hunger and could not resist buying some candy, because the pasta would
take some time to prepare. He put one bar of Mars in his basket, paid, and went home. On
the way home, eating the candy bar, he passed the tobacconist's where he often buys the
evening paper. However, today he had already bought the paper in the supermarket.
This example illustrates how a consumer may behave and the decisions implicit in
everyday behavior. Marketing scholars have built formal models of the store choice, the
category choice, the purchase timing, the purchase incidence, the brand choice, and the
purchase quantity. Some models focus on one of the decisions, while other models include
more than one of the consumers' decisions. Relations between the decisions have also been
modeled. The next section describes some of the models and findings about the consumer's
decisions. The models and findings mostly concern sales promotion's impact on the
consumer's implicit purchase decisions. The decisions are discussed in what may seem a
hierarchical order beginning from the top. This order is chosen only to provide some
structure for the discussion. However, it is not assumed that this is a hierarchy or that the
decisions are made in any particular order or that the decisions are sequential.
2.3. Sales Promotion and the Store Choice

The store choice decision discussed here is that of where the household or consumer
buys a basket of products, i.e., the choice of destination for a shopping trip. The consumer's
decision where to buy a specific product is not treated here (see e.g., Grover and Srinivasan
1992, Kumar and Leone 1988 and Bucklin and Lattin 1992). That decision is covered in
Section 2.4 about category choice and in the discussion about indirect competition among
stores (Section 3.2.3).
The store choice decision or the consumer's store patronage decision translates into store
traffic for the retailer. The consumer's store choice has a long-term and a short-term
component. We shall discuss the long-term component first.
Ahn and Ghosh (1989) review store choice models and find that the consumers' retail
patronage decision has been modeled in two similar ways. The first is a form of attraction
- 27-
model called spatial interaction model. This tradition dated back to 1962 when Huff
published his sen1inal work. 7 In this type of model, the number of customers visiting a store
is proportional to the store's attraction relative to the total attraction of all stores. The
second type of model is the multinomial logit model. This type of model predicts the that
probability that a consumer will choose a store is proportional to the individual's utility of
this store relative to the individual's utility of all stores. Mathematically these models are
very similar. The difference between the models is mainly conceptual. The attraction type
models are calibrated on aggregate data on the number of shopping trips while the logit
models are calibrated on individual choice data. (See also Craig, Ghosh, and McLafferty
(1984) for an excellent review of these models.)
Ahn and Ghosh (1989) extend the store choice models to overcome the independence of
irrelevant alternatives (1lA) assumption of the attraction and logit models. They propose
that the store choice is a hierarchical decision process where the consumer first makes a
choice of shopping center and then, within this shopping center, makes the store choice.
The nested logit model is used to model the sequential hierarchical decision. They found
that the nested logit model performed slightly better than the multinomial model on data
from a choice experiment on MBA students where the setting was clothes shopping.
Fotheringham (1988) also argues that store choice is a hierarchical process since the
consumer cannot evaluate all information simultaneously. If choice is hierarchical,
multinorniallogit models are biased.
According to Engel, Blackwell, and Miniard (1986), location is the most important
factor associated with store choice. They point out that the determinants of the store choice
decision may vary by product class but list the following ten attributes that are usually
salient.
• location
• nature and quality of assortment
• price
• advertising and promotion
• sales personnel
• services offered
• physical store attributes
• nature of store clientele
• store atmosphere
• post-transaction service and satisfaction
7 Huff, D.L. (1962), "Determination of Intra-urban Retail Trade Area," Real Estate Research Program,
University of California, Los Angeles.
- 28 -
Kahn and Lehman (1991) study how the assortment enters the consumer's store choice
decision. They find that the consumer's most preferred item is important when evaluating
assortments. However, consumers also want flexibility for future choice. The larger the
number of unique and acceptable items the assortment offers, the better the assortment. An
assortment consisting of a large number of similar or unacceptable items, on the other
hand, makes the consun1er's choice unnecessarily time-consuming.
Price is also an important factor in the store choice. Mowen, Wiener, and Young (1990)
used a conjoint experiment to investigate the importance of sales tax differences for store
choice. Not surprisingly, they found that stores that did not have to charge a sales tax had
an advantage over stores that had to charge the tax.
Malhotra (1986) found the following eight characteristics important for store choice:
• quality of merchandise
• variety and assortment of merchandise
• returns and adjustment policy
• service of store personnel
• fair prices
• convenience of location
• store layout
• credit and billing policies
Although the store choice decision occasionally can be a high involvement decision,
Engel, Blackwell and Miniard (1986) state that "the consumer usually has well-formed
evaluative criteria and shops at the most convenient store as long as the store continues to
satisfy those criteria" (p.492). The well-formed evaluative criteria are often referred to as
store image.
According to Steenkamp and Wedel (1991), store image has in previous research been
found to be related to store patronage, store loyalty, and the share of the household budget
spent in the store.
Mazursky and Jacoby (1986) define store image as:
1. a cognition and/or affect (or set of cognition and/or affects)
2. which is (are) inferred
3. either from a set of ongoing perceptions and/or memory inputs

attaching to a phenomenon (i.e., either an object or event such as a store,
a product, a "sale", etc.)
4. and which represent(s) what that phenomenon signifies to an

individual.
- 29-
How is the store image formed? Louviere and Gaeth (1987) propose a hierarchical
model of consumers' evaluation of retail facilities. The overall judgment of the retail
facility is modeled as a weighted mean of the judgments of four intermediary
macromeasures (1) price, (2) selection, (3) convenience, and (4) quality. Each of these
macromeasures is in tum a composite of a number of specific store attributes.
Price image is an important part of the overall image of the store. Cox and Cox (1990)
state that consumers often use perceptions of the store's price level. They say that although
little is known about how such perceptions are fornled, the managerial retailing literature
suggests that supermarket managers identify some key products on which they must be
price competitive to project a low price image for the store. Cox and Cox find that the
presentation of prices in advertisements influences consumers' price image of the store.
Consumers perceive the store's overall price level to be lower when advertised prices are
presented as reductions from a higher previous price. Thus, sales promotion may be
important for the store's price inlage.
In a theoretical study, Feichtinger et ale (1988) modeled the price image formation
process as a function of observed prices. In their model, the consumer's store choice is
dependent on the store's price image. The purchase amount, however, depends on the
actual prices encountered when visiting the store.
Now we have mainly considered the long term component of store choice and seen that
sales promotion may be involved in the creation of store image. What determines the store
choice in the short-term and does sales promotion affect the short-term store choice? That
is what the next paragraphs will deal with.
In a Swedish study on grocery shoppers' store choice, Hernant (1992) found that 37
percent of the customers had chosen to visit that specific store because it was located close
to the customer. Almost one out of five respondents said that they had chosen that store out
of habit. Nine percent had chosen the store because they happened to pass by. Only three
percent answered that they had chosen the store to take advantage of a price deal, while
four percent had chosen the store because they wanted to buy a special product. Five
percent had made the store choice dependent on the assortment.
Hemant's findings indicate that sales promotion may not be very important for the
consumer's store choice.
Bucklin and Lattin (1992) studied competition among stores using scanner panel data
for liquid detergents. They found no significant store traffic effects of retailers' marketing
activities for detergents. In other words, the consunlers' store choices were not influenced
by the sales promotion.
Blattberg and Neslin (1990) argue that the effect on store traffic of promotions for single
products is likely to be very small since hundreds of products are on sale every week.
- 30-
Walters and Rinne (1986) found that store traffic was generally unaffected by portfolios
of loss leaders (i.e., products are sold at negative gross margins). The store traffic effect of
double couponing was also insignificant in all stores. Walters (1988) studied promotional
effects on store traffic in six product categories. Promotions in four of the categories were
found to have significant positive effects on store traffic. Walters and MacKenzie (1988)
found that most loss leader promotions did not affect store traffic. They also found that
neither double couponing nor in-store price specials had any effect on store traffic.
Julander (1984) states that the failure of promotional activities to affect store traffic does
not imply that they are without effects on store traffic. Their effects might be
overshadowed by competitors' actions. Also, when using historical data, failures to find
effects of sales promotion on store choice may result from insufficient variation as well as
lack of effects.
In summary, the consumer's store choice seems to consist of both long-term and short-
term components. The long-term component gives the consumer's choice set, where store
location, perceived price level, and assortment are the most important factors. Store
location is also the most important factor for the short-term component of store choice.
Sales promotion has been found to have limited effect on the consumer's store choice.
2.4. Sales Promotion and the Category Choice

One of the basic tasks the consumer has is how to allocate his resources. In other words,
the consumer has to decide which products to consume. In micro-economics, this is
modeled as an allocation problem where the consumer maxin1izes his utility subject to a
budget constraint (e.g., Varian 1984). The consumer may then use promotions to come to a
higher utility level. In reality, finding the optimal allocation and optimizing the usage of
promotions are very difficult decision problems (see Meyer and Assuncao 1990). However,
the consumer's allocation of resources gives his purchase rate of the product. The purchase
rate enters marketing models of the timing of the consumer's purchases.
The consumer's product category purchase decision has, in marketing research literature,
been modeled in two ways: (1) as a decision of when to buy the category, or (2) as a
decision of whether or not to buy the category during a specified time period. The models
of when to buy are referred to as purchase timing models while the the models of whether
or not to buy are referred to as purchase incidence models.
Examples of models of purchase incidence are the Poisson distribution, the gamma
distribution, and the NBD model (negative binomial distribution). Recent work has used
the Erlang-2 distribution, where the probability of purchase incidence is dependent on the
time since the last purchase (Blattberg and Neslin 1990). (See also Uncles and Ehrenberg
- 31 -
1990, for an application of the Dirichlet model on coffee purchases at Safeway.) Dalal et al.
(1984) developed a purchase incidence model that can handle arbitrary heterogeneity
among consumers' purchase rates.
Wheat and Morrison (1990) showed in a recent study that purchase incidence models
are preferable to purchase timing models. The reason is econometrical; in a purchase
timing model, individuals with long interpurchase intervals are usually excluded from the
sample for two reasons. First, the number of observations for an individual is smaller the
longer the average interpurchase time, and the number of observations may be too small to
allow estimation. Second, the length of the last, interrupted, interpurchase period is
unknown. This makes estimates of the interpurchase time biased. Including the people with
very low purchase rates in the sample makes assessment of the reliability and goodness of
fit difficult. Still another reason is that the purchase timing for most products is dependent
on the timing of the shopping trip.
Another problem with interpurchase time models is the difficulty in including marketing
variables. However, Gupta (1991) developed a model of households' interpurchase times
that includes marketing variables in a parsimonious way. He found that feature advertising
and display both significantly increased the purchase rate, and thereby shortened the
interpurchase time. More remarkable is that he found that price cuts had no significant
effect on the consumers' interpurchase time. Gupta forwards the explanation that unless the
price cut is advertised or featured, only those who intended to buy the product are likely to
recognize the deal. In the coffee data set Gupta used, he found that large household
inventories lead to longer interpurchase time.
The positive relation between household inventory and interpurchase time was also
observed by Neslin, Henderson, and Quelch (1985). Using an interpurchase time model,
they found that sales promotion does not have much impact on the interpurchase time.
Gupta (1988) also n10deled interpurchase time as a function of sales promotion. Gupta
found that 14 percent of the brand sales increase due to promotion comes from purchase
acceleration. Chiang (1991) modeled purchase incidence as a function of sales promotion,
and found that 13 percent of the brand sales increase as a result of promotion is due to
increased purchase incidence. All three studies used scanner panel data for coffee.
Helsen and Schmittlein (1993) used hazard rate models to estimate the effect of sales
promotion on interpurchase time for saltine crackers. They found that sales promotion
decreased the time between purchases.
Vilcassim and Jain (1991) showed the importance of unobserved heterogeneity among
consumers when modeling interpurchase time. They found that price, display, and family
size variables explained 14 percent of the variance in repeat purchase rate for saltine
crackers. The remaining 86 percent variation was attributed to unobserved heterogeneity
- 32-
among the consumers. Also using salt cracker data supplied by IRI (Information Resources
Inc., a Chicago based subsidiary of Brad and Dunstreet), Bucklin and Lattin (1991) found
promotional activities to have an impact on purchase incidence. Promotional activities had
the most impact on the purchase incidence of consumers in the opportunistic state, i.e.,
consumers who had not planned to buy the product category before entering the store. They
also found household consumption rates and inventories important determinants of
purchase incidence.
Consumers make their grocery purchases when they visit grocery stores. When visiting
a supermarket, the consumer usually buys a bundle of products. The implication of this is
that the exact purchase timing for most products is dependent on the timing of the shopping
trip, rather than the other way around. This was also pointed out by Wheat and Morrison
(1990) as a reason why models predicting whether or not a consumer makes a purchase are
better than interpurchase time models.
Many products used daily in the household can be consumed over a long period.
Examples are detergents, soap, flour, sugar, oil, and salt. Consumers have low purchase
frequencies for these products (Julander 1984). When the purchase frequency is much
lower than the frequency of shopping trips, the household has several purchase occasions
for the product when the stocks approach the repurchase level. In many product categories,
the frequency of promotion is high compared to the purchase frequency. This implies that
consunlers can make their purchase decision partially dependent on the availability of
promotions for a preferred brand when the household's stocks are running low. They may
shop from deal to deal, to use the terminology of Blattberg and Neslin (1990).
To summarize, when household inventories are high, the probability of purchase
incidence is low. Sales promotion shortens the interpurchase time, and thus, increases the
probability of purchase incidence. It is better to model the consumer's category decision as
whether or not to buy than when to buy, because the purchase timing is largely dependent
on the occurrence of a shopping trip.
2.5. Sales Promotion and the Purchase Quantity Choice

When deciding to buy from a product category on a shopping trip, the consumer has in
some sense made the most basic purchase quantity decision, Le., the decision to buy more
than zero units. Having decided to buy a product category the consumer has to decide how
much to purchase. This decision may involve choosing a package size and a number of
units to buy of that size.
The quantity that the consumer chooses to buy depends on the household's rate of
consumption of the product and the interpurchase time. A promotion that lowers the unit
- 33 -
price could make the consumer buy a larger quantity of the good. If the household's
inventory holding costs are low, it could make economic sense the household to take on
extra inventory of the product (Blattberg, Eppen, Lieberman 1981). Iulander (1984) points
out that if the availability of the product in the consumer's home increases the
consumption, this strategy is less attractive for the household that seeks to minimize the
cost of its consumption bundle. Further, the consumer would not use sales promotion to
build inventories of perishable product.
The purchase quantity decision has not yet been subject to much study. Some pioneering,
work has been done by Neslin, Henderson, and Quelch (1985), Gupta (1988), Chiang
(1991), Krishnamurthi and Raj (1988, 1991), and Krishnamurthi, Mazumdar, and Raj
(1992). All used scanner panel data for coffee.
Chiang (1991) found that promotion induced increases in purchase quantity represented
6 percent of the brand sales increase due to promotion. Gupta (1988) found the purchase
quantity's share of the increase to be 2 percent. Neslin, Henderson, and Quelch (1985)
found that increased purchase quantity is nl0re important than shortened interpurchase time
in purchase acceleration due to sales promotion. They defined purchase acceleration as
shortened interpurchase time or increased purchase quantity, or both.
Krishnamurthi and Raj (1988) found that only the price of the chosen alternative
determines the consumer's purchase quantity while prices of other alternatives have no
impact on the purchase quantity. Krishnamurthi and Raj (1991) found that the purchase
quantity elasticity of brand loyal consumers is higher than the purchase quantity elasticity
of brand switchers.
In summary, there is evidence that sales promotion may increase the consumer's
purchase quantity.
2.6. Sales Pronlotion and the Brand Choice

The consumer's brand choice is the favorite modeling object in the consumer behavior
and marketing literature. Bass (1974) viewed the consumer's brand choice as stochastic.
Bass et al. (1984) investigated the order of the brand choice process. They found that
consumers' choices of soft drinks are zero-order, Le., a consumer's choice probabilities are
the same in each period. Givon (1984) used a stochastic brand choice model to study
consumers' variety-seeking through brand-switching.
Lattin (1987) found that consumers seek variety on some product characteristics but are
loyal to other characteristics. The product category was soft drinks. Bawa (1990) found the
consumer's purchase history to explain the consumer's variety-seeking and inertia in the
brand choice. The product categories Bawa studied were facial tissue, paper towels, and
- 34-
ready-to-eat cereals. Dodson et ale (1978) found sales promotion to increase the consumer's
choice probability of the brand. They also found evidence that the retraction of the deal had
a negative effect on the consumer's repeat purchasing of that brand. Dodson et al. used data
for margarine and flour.
Corstjens and Gautschi (1983) reviewed formal (individual level) choice models in
marketing. These models are derived from four theories: the neoclassical economic theory
as extended by Lancaster, the risk-preference theory, the strict utility model, and the
random utility model. The strict utility model and the random utility model are the most
important brand choice models. The logit model is an example of a model that can be
interpreted as a strict utility model or as a random utility model.
Guadagni and Little (1983) used a logit model to measure the effect of sales promotion
on brand-switching. The model was tested on a set of scanner panel data for coffee. They
found that household specific variables for brand and size loyalty were important
determinants of variation across households in brand choice. The household-specific
variables accounted for about one quarter of the variance in the time-series, cross-sectional
data set. Sales promotion was found to explain another quarter of the variance. They
showed that the logit model provides a parsimonious way of describing the consumer's
brand choice.
Bucklin and Lattin (1991) modeled consumer's brand choice of saltine crackers using a
multinomiallogit model. Sales promotion had a significant impact on the consunler's brand
choice when the consumer was in the opportunistic state. Vilcassim and Jain (1991) also
found sales promotion to affect consumers' choice of saltine cracker brand. They found that
the effect of sales promotion to be asymmetric between brands. Asymmetric switching
between brands was also modeled by Allenby and Rossi (1991) who used margarine data to
illustrate their consumer-level, brand choice model.
Gupta (1988) and Chiang (1991) both used the multinomial logit model to model the
consumer's coffee brand choice as a function of sales promotion. Gupta found that brand
switching accounts for 84 percent of the brand's incremental sales due to sales promotion.
Chiang estimated the same figure to be 81 percent.
Krishnamurthi and Raj (1988, 1991) modeled the consumer's choice of coffee brand as a
function of sales promotion. In their 1988 article, they found that the brand choice, in
contrast to the quantity choice, is dependent on the attributes of all the choice alternatives.
In 1991, they showed that the difference between brand loyals' and brand switchers'
elasticities in the brand choice decision is larger than the overall difference in their
elasticities. This is because the brand loyal consumers' purchase quantity elasticity is higher
than that of the brand switchers.
- 35-
Moore and Olshavsky (1989) found that very deep price discounts (75 percent) lead to
fewer consumers choosing the brand than a moderate (30 percent) price discount when
consumers were not familiar with the brand. When the consumers were familiar with the
brand, the deep price discount attracted more consumers than the moderate price discount.
The product was white long sleeve men's dress shirts.
In consumer behavior models in marketing, price is often included as an attribute that
the consumer uses when forming an attitude toward the brand. The consumer chooses the
brand for which he has the highest (most favorable) attitude. A promotion leads to an
improved attitude toward the brand and thereby improves the brand's position in the brand
choice process of the consumer.
The consumer's brand choice can come out as a repurchase of the brand last bought or as
a purchase of another brand. The latter is referred to as a brand switch.
Closely related to brand-switching as a consequence of promotions is the impact of
promotions on repeat purchasing. This is one link between short-term effects and long-term
effects of sales promotion. It has been suggested that two different psychological processes
operate simultaneously in connection with the impact of promotions on repeat purchasing.
On the one hand, classical conditioning is thought to imply that the purchasing of a product
in itself leads to a higher repurchase probability. Blattberg and Neslin (1990) call this the
purchase effect. On the other hand, the attribution theory implies that the consumer's
attitude toward the brand may be negatively affected by a purchase when the product is on
deal. This is dubbed the promotion usage effect (Blattberg and Neslin 1990). Studies
reporting decreased repurchase probabilities after promotional purchases have failed to
recognize the importance of measurement of this phenomenon on the individual level
(Blattberg and Neslin 1990).
In summary, the literature provides ample evidence that sales promotion affects the
consumer's brand-choice decision.
2.7. Combining the DecisiollS

Blattberg and Neslin (1990) recognized a trend toward modeling efforts combining
purchase incidence, purchase quantity, and brand choice models. Whether the decisions
should be regarded as independent or correlated is being debated and as yet unresolved.
Neslin, Henderson, and Quelch (1985) model interpurchase time and purchase quantity
as functions of sales promotion and previous inventory (i.e., the consumer's inventory
directly after the previous purchase). Further, they investigate the possible impact of
interpurchase time on the consumer's purchase quantity. They find this effect to be small in
their sample.
- 36-
Wagner and Taudes (1986) model brand choice and purchase incidence. Their integrated
model was tested on panel data for detergents and was found to outperforn1 simpler
models.
Currim and Schneider (1991) outline five expected purchasing strategies that consumers
might use as response to sales promotion. The consumer may exhibit: (1) brand preference
without purchase acceleration, (2) promotion-motivated brand-switching without purchase
acceleration, (3) promotion-motivated brand-switching and purchase acceleration, (4)
promotion-motivated purchase acceleration of preferred brand, and (5) switch to other
brand, not motivated by promotion. Their empirical illustration (scanner panel data for 200
heavy users of coffee) provides support for four of the five hypothesized purchase
strategies.
Gupta (1988) developed a model for purchase timing, brand choice, and purchase
quantity. The three decisions are independent in Gupta's model. Chiang's (1991) model
differs from Gupta's in that Chiang models purchase incidence rather than interpurchase
time. It also differs in that Chiang regards the consumer's three decisions as interrelated.
Krishnamurthi and Raj (1988, 1991) and Krishnamurthi, Mazumdar, and Raj (1992)
model the consumer's purchase quantity decision and the brand choice decision. They
regard the quantity decision as dependent on the brand choice.
Bucklin and Lattin (1991) model purchase incidence and brand choice using a nested
logit approach, where the probability of purchase incidence is dependent on the maximum
expected utility of the brand choice. Vilcassim and Jain (1991) also model purchase timing
and brand choice as correlated decisions.
Bucklin and Gupta (1992) segment households buying liquid laundry detergent by their
sensitivity to sales promotion. They model the brand choice and purchase incidence
decisions using a nested logit approach. Sales promotion is found to affect some
households' purchase incidence decision and others' brand choice decision. More
specifically, they found that households whose purchase incidence is accelerated by sales
promotion do not switch brands. Other households, whose brand choice decisions are
affected by sales promotion, do not accelerate their purchases because of the sales
promotion.
Dalal et ale (1984) modeled the purchase incidence and brand choice. Their model is
very general and assumes neither dependence nor independence between the decisions.
Keng and Ehrenberg (1984) modeled the store choice and the brand choice using the
NBD model. They found that the individual consumers tend to spread their purchases of
the product to over stores and over brands in proportion to the market shares of the
products and the stores.
- 37 -
Grover and Srinivasan (1992) investigate the multiple effects of sales promotion using a
brand choice model, a store choice model, and a purchase quantity model. The decisions
are dependent in their framework.
In summary, there is a trend toward modeling two or more of the consumer's decisions
simultaneously. However, there is no model that has integrated all four decisions.
2.8. Dynamic Effects of Sales Promotion

The dynamic effects of sales promotion are of a different kind than intertemporal effects
or lags caused by inventory effects of sales promotion. Dynamic effects of sales promotion
here refer to its impact on the effectiveness of sales promotion in later periods.
Winer (1986) proposed and found empirical support for a model where demand is a
function of current observed price and the consumer's internal reference price. The data set
used in the empirical test was panel data for heavy users of coffee. The model implies that
frequent sales promotions make the deal price to the consumer's reference price. The
regular price is then regarded as a prenlium price.
Lattin and Bucklin (1989) built a model incorporating reference effects of sales
promotion. They model the consumer as perceiving a brand as either promoted or
nonpromoted. If the consumer perceives the brand as promoted, i.e., a brand that regularly
is offered on a deal, the consumer is unlikely to purchase at regular price. Sales promotions
for brands that are perceived as nonpromoted are regarded as purchase opportunities.
Empirical support for the model was found in a test using scanner panel data for regular
ground coffee.
Jacobson and Obermiller (1990) stress the importance of consumers' expectations of
future prices and promotions. Using grocery store data for canned tuna, they found
empirical support for their serial correlation model of price expectation formation.
Kalwani and Yim (1992) found that the promotion frequency and the depth of the price
discounts have impact on consumers' price expectations. The consumers' price and
promotional expectations were found to affect brand choice. Kalwani and Yim conducted
an experiment with 200 undergraduate students as respondents. The product stimuli were
two hypothetical brands of liquid laundry detergents.
Krishna, Currim, and Shoemaker (1991) used a survey to study consumers' perception of
promotional activity. They found that about half of the respondents were reasonably
accurate in their perception of the deal frequency. The expected deal frequency was found
to be positively correlated with the perceived past deal frequency. The accuracy of recall of
sale price was better the higher the frequency of promotion.
- 38 -
Bawa, Landwehr, and Krishna (1989) found that the in-store marketing environment is
correlated with consumers' sensitivity to sales promotion. Consumers tended to be more
price sensitive in stores with a high price variability. Bawa et al. (1989) used panel data for
regular ground coffee in their study.
Krishna (1991) showed that the price which consumers are willing to pay is correlated
with the actual deal frequency of the brand. However, the perceived deal frequency was
more important for consumers' willingness to pay than the actual deal frequency. She used
a computer-simulated experiment with 159 undergraduate and graduate students who made
12 weekly purchase decisions for two hypothetical brands of soft drinks. Using a similar
computer-simulated shopping experiment, Krishna (1994) showed that consumers who
have information about future prices behave differently from consumers without such
information.
Bolton (1989) studied the relationship between market characteristics and promotional
price elasticities. Promotional price elasticity is found to be positively correlated with
feature advertising activity in the category. The display activity in the category and for the
brand were found to be negatively correlated with the elasticity. The correlation between
category price activity and elasticity was insignificant. The study was conducted on store-
level scanner data for three brands in four categories in 12 stores. The product categories
were waffles, bleach, tissue, and ketchup.
Raju (1992) studied the determinants of variability in product category sales. Using
scanner data from 63 homogeneous product categories Raju found that the magnitude of
deals increase the variability in category sales. The frequency of deals had the opposite
effect, Le., more frequent deals lead to less variability in category sales. Raju did not study
the effect on variability of brand sales.
Helsen and Schmittlein (1992) found that the deal frequency is positively correlated
with the price sensitivity. They found forward buying to be more common in product
categories with high mean depth-of-discount. The study employed scanner panel data for
ketchup, canned tuna, margarine, toilet tissue, and peanut butter.
Summarizing the literature on dynamic effects of sales promotion, it seems that the
frequency of sales promotion and the size of the price deals have an impact on the
consumers' deal sensitivity. The sign of the dynamic effects has not been conclusively
determined. The correlational nature of the studies implies that the causal direction
between promotional sensitivity and promotional pattern cannot be determined. The
implication for the retailer of dynamic effects of sales promotion is that the consumer
response to sales promotion cannot be assumed to be constant. Retailers trying to improve
the effectiveness of their sales promotion would consequently need to constantly monitor,
or track, the sales response.
- 39-
2.9. Summary
This chapter started with an informal illustration of the shopper's decisions before and
during a shopping trip. These decisions have been modeled formally in earlier research in:
(1) store choice models, (2) purchase incidence and interpurchase time, (3) purchase
quantity models, (4) brand choice models and (5) models of two or more decisions.
Previous research has found that sales promotion may affect all four decisions. However,
its importance varies across the four decisions and across products. The weakest effects of
sales promotion were found for the store choice.
There is a trend toward an integrated modeling approach of the consumer's decisions.
The integrated n10deling efforts described in the literature have modeled two or three of the
consumer's decisions. None of the models has integrated all four consumer decisions.
Dynamic effects of sales promotions have been found in previous research. The
consumers sensitivity to promotion may depend on the actual and perceived promotional
pattern. Changing the promotional pattern may thus lead to changing promotional
effectiveness.
- 40-
3. PROMOTIONAL SALES RESPONSE
3.1. Introduction
This chapter examines the outcome of the consumers' responses to sales promotion from
a retailer's point of view. It starts with a discussion about the sources of a retailer's
incremental item sales and present some empirical findings from previous research.
Thereafter, different modeling approaches are discussed and some important promotional
sales-response models used in previous research are presented. These models include
brand-sales models, market-share models, category-sales models, store-sales models, and
store-traffic models.
3.2. Sources of Retailer's Incremental Item Sales

Previous research has shown that the impact of sales promotion on the promoted brand's
sales in a store is often substantial (e.g., Chevalier 1975; Curhan 1974; Holmberg 1992;
Wilkinson, Mason, and Paksoy 1982). According to Blattberg and Neslin (1990), retailer
incremental brand-sales volume comes from four sources:
• other retailers
• competitive brands
• purchase and quantity acceleration
• increased consumption
This classification of the sources of incremental sales volume provides a good
framework for analyzing promotional effects. However, we need to modify the
classification slightly to make the sources mutually exclusive and collectively exhausting.
Since total consun1ption can be regarded as fixed, increased consumption of the
promoted item must logically come from other items and product categories. Thus, sales
promotion for an item in a particular store may steal sales from other items during the
promotional period, from its own future sales, as well as from other items' future sales in
that store and other stores. The sources can be seen as relating to the three dimensions
product, time, and store, as illustrated by Figure 3.1.
- 41 -
Other stores
Focal store
Focal item
Other items
in the focal
category
Other
categories
Past Present Future
Figure 3.1. Sources of retailer's incremental item sales volume.
Figure 3.1 shows a 3x3x2 cube. The shaded compartment represents the present sales
volume of the focal brand in the focal store. The remaining 17 compartments represent
possible sources of incremental sales volume. Assuming that the total possible sales are
fixed, sales promotion may move sales from one compartment to another. We shall
examine these effects one at a time, concentrating on the question where the retailer's
incremental sales of the promoted item come from. In other words, we are here concerned
with the substitution effects. The possible positive effect on the retailer's sales of other
products through increased store traffic was treated earlier, in the discussion about the
consumer's store choice.
3.2.1. Other Brands
The retailer's sales increase of a promoted product may come from other brands. Figure
3.1 suggests that the promoted brand may steal fronl other brands' past, present, and future
sales in that store as well as in other stores. What we will discuss here is the effect on the
store's present sales of other brands in the same category.
Using an in-store experiment, Chevalier (1975-76) investigated substitution patterns
caused by special display. The increase in category unit-sales was found to be slightly
larger than the increase in brand unit-sales on average. Chevalier concludes that most of the
increased unit sales of the item due to special display come from additional category sales.
- 42-
Wilkinson, Paksoy, and Mason (1981) reported elasticities and cross-elasticities from an
in-store experiment. Special displays were found to have a strong positive impact on the
promoted brand's sales and significant negative impact on other brands' sales. The product
categories studied were pie shells, rice, apple juice, and soap.
Kumar and Leone (1988) found price deals, feature advertising, and special displays
significant determinants of the promoted brand's sales. The only significant cross-
elasticities were those for price deals. Feature advertising and special displays did not have
any significant impact on sales of other brands than the promoted.
Blattberg and Wisniewski (1989) found asymmetric patterns in the promotional cross-
elasticities. When higher-price and quality brands are offered on a deal, they steal sales
from other high-priced brands and from lower-priced brands. Lower-priced brands, on the
other hand, were found to steal sales only from other lower-priced brands. Blattberg and
Wisniewski used store level scanner data for flour, margarine, bathroom tissue, and canned
tuna.
Bemmaor and Mouchoux (1991) studied short-term deal-price elasticities using an in-
store experiment. They found that while the direct elasticities ranged between 2 and 11 in
absolute value, cross-elasticities fell in the range from 2 to 2.7.
Walters (1991) found that price promotion for one brand in a store reduced sales of
other brands in the same category. The product categories were spaghetti, spaghetti sauce,
cake mix, and cake frosting.
To summarize, previous research has found that sales promotion for one brand may
have negative effects on sales of other brands in the product category. The importance of
such negative cross-effects may vary across product categories and between promotional
tools. Asymmetries in cross-elasticities between products within a category have been
found in several studies.
3.2.2. Other Categories

The impact of sales promotion in one product category on sales of other categories in a
store has received little attention in previous research. The category management concept,
as a specific example, category space management (see e.g., Bultez and Naert 1988)
assumes that space redistribution within a product category has negligible effects on other
categories.
No study was found that explicitly dealt with the promotion-induced substitution across
product categories within a store. The extent of such substitution is unknown, but at least
two reasons speak for its existence. First, product categories are often narrowly defined,
which places fairly close substitutes outside the category. For example, when a product
category is defined as regular ground coffee, then decaffeinated coffee, coffee beans, and
- 43-
instant coffee are not part of the category. However, substitution could occur. Second, the
household's total consumption of grocery products is restricted by biological and pecuniary
constraints. Incremental consumption and purchases of one category would thus have to
imply decreased consumption and purchases of other categories.
One piece of evidence for cross-category substitution was provided by Walters (1988)
who was surprised to observe a negative relationship between promotional sales of notions
and unrelated sales in a study using data from two fabric stores. The study mainly focused
on the substitution within categories and the positive effect on other categories' sales
expected from complementarity and from the effect on store traffic.
Walters (1991) investigated the effects on sales promotion on sales of complementary
categories. He found significant positive effects on spaghetti sales when spaghetti sauce
was offered on a deal and vice verse. The results were replicated on the cake mix and cake
frosting pair of complementary product categories.
In summary, the substitution effect between product categories in retail stores has been
neglected in previous research. Research investigating the extent of such substitution, as
well as the complementarity among product categories, is needed.
3.2.3. Other Stores

Some of the promotion-induced sales increase of an item in one store may come from
other retailers. As Keng and Ehrenberg (1984) note: "Most customers of a given chain also
buy the product elsewhere." Empirical evidence of such interstore competition was
provided by Kumar and Leone (1988) who reported that price dealing in one store affected
sales of that brand and sales of competing brands in closely located stores. They also found
that feature advertising and special display in one store had negative effects on the same
brand in other stores. The product category was disposable diapers.
Walters (1991) found some evidence that a price deal on a brand in one store reduced
sales of that brand and its close substitutes in another store. However, Walters hesitates to
accept the evidence.
Grover and Srinivasan (1992) found that the promotional attractiveness of a product
category in one store significantly affected that store's share of the sales in that category.
The product category used in the study was coffee.
Bucklin and Lattin (1992) note that store competition comes in two flavors. The direct
component of store competition is the effect on the consumer's store choice and leads to
reallocation of store traffic. The indirect component of store competition is the effect on
category sales in another store in the absence of any direct effect. The indirect effect occurs
because consumers are not loyal to a single store but rather shop regularly in a set of stores.
A pron10tion in one store may then increase the consun1er's probability of purchase
- 44-
incidence when visiting that store. If the increased probability of purchase incidence results
in a purchase, then the consumer's probability of purchase incidence is decreased in
subsequent periods when other stores are visited.
The concept of indirect store competition integrates the concept of cross-shopping and
the purchase acceleration concept. Indirect store competition is more subtle than direct
store competition, but might be more important. Bucklin and Lattin (1992) found no
significant direct effect but observed significant indirect competition in their empirical
illustration using data on liquid laundry detergents.
In summary, the literature shows that sales promotion in one store may reduce other
stores' sales of that product and its substitutes. Such store competition can be either direct,
affecting store traffic, or indirect, affecting the consumers' purchase timing. The indirect
effect might be the more common.
3.2.4. Other Periods
Sales promotion may cause substitution of current sales for other periods' sales. Note
that shifts in sales may be due to changes in the timing of the consumption, inventory
build-up, or a combination of the two.
Purchase acceleration is the shift of sales volume from future periods to the present
period. Previous research has found that the purchase acceleration effect exists at the
consumer level (e.g., Blattberg and Neslin 1990; Neslin, Henderson, and Quelch 1985). It
seems that consumers do take advantage of promotions to build inventories. A post-
promotional dip, or trough, would consequently be expected. Neslin and Shoemaker (1983)
produce a post-promotional dip in sales in a model which simulates consumer behavior.
However, studies using sales data have seldom observed any trough after the promotion
period (Blattberg and Neslin (1990). The existing evidence comes from panel data, i.e.,
individual level data.
The high frequency of promotions may mean that the inter-promotion period is too short
for sales to return to their normal level. Doyle and Saunders (1985) suggested that
promotions might be anticipated, causing a pre-promotion dip, which would make the post-
promotion dip difficult to detect. Blattberg and Neslin (1990) note that if a promotion does
accelerate households' purchase quantity, then the promotion will steal sales from several
periods. That would make the post-promotional dip difficult to observe.
Another explanation for the difficulty in observing troughs after sales promotions when
using store level data is that other brands' and other stores' sales in future periods are
affected. The effect on the store's future sales of the brand is diluted by brand- and store-
switching, making the effect difficult to measure. The exception would be a store with a
large proportion of store loyal customers. In that case, the store would steal mostly from its
- 45-
own future sales. Such a retailer may observe post-pron10tional troughs, provided that not a
large number of the store loyal customers shop from deal to deal.
In summary, although sales promotion leads to household level substitution of current
purchases for future purchases, the effect for a retailer might be small.
3.3. Response Modeling: Direct and Indirect Approach

Response models or demand models constitute a special class of predictive models used
in marketing (Naert and Leeflang 1978). The purpose of a response model is to relate
changes in marketing variables (and other decision and environmental variables) to
changes in demand. Response models are important for evaluating the effectiveness of past
promotions and for making conditional predictions for future promotions.
The performance measure in a sales response model is a measure of demand. Sales
response models may be classified according to the level of demand. Three l,evels of
demand are usually recognized. Product class demand (or primary demand) refers to the
demand for a product class and is modeled by product-class sales models. The second level
of demand (secondary demand) is demand for a particular brand. This is modeled in brand-
sales models. The third level is the relative number of units that is sold by a particular
brand relative to the total number of units sold in the product class. This is modeled in
market-share models. The relationships between the different levels of demand are given
by the following equations.
Ij
Qjt = L,qijt (3.1)

i=l
and
qijt
m.. = - (3.2)
lJt Q
jt
where,
= total demand in product class j period t

qijt = demand for brand i in product category j period t
= market share brand i in product category j period t
= number of brands in category j
- 46-
The definition of market share implies that if we know demand at two of the three
levels, then demand at the third level can be calculated. That is, if the product-class
demand and the market share are known, then the demand for the brand can be calculated.
This makes two modeling alternatives available for building brand-sales models. The first
alternative is to model brand sales directly as a function of a set of explanatory variables,
X ijt •
(3.3)
The second alternative is to model brand sales indirectly, by building models of primary
demand and brand market-share as functions of some sets of explanatory variables. Brand
sales are then computed as the product of primary demand and the brand's market share.
(3.4)
mijt == m( Xijt) (3.5)
(3.6)
According to Naert and Leeflang (1978), modeling brand sales indirectly is preferable
for a number of reasons. The indirect modeling approach distinguishes between changes in
brand sales caused by changes in product-class sales and changes caused by changes in its
market share. The different levels of demand may be influenced by different sets of
variables. For example, environmental variables and seasonal or cyclical variables may
cause changes in primary demand but may leave the market shares unaffected. The market
share model need not include such variables and focuses attention on the con1petitive
interaction among the brands (Naert and Leeflang 1978).
The indirect modeling of brand sales is an example of hierarchical modeling. The
category or product-class sales is modeled in the higher level model while market shares
are modeled in the lower level model. When the outcome at the higher level is modeled as
a function of the outcome at the lower level, we have a nested hierarchical model.
3.4. Brand Sales Response Models

The focus here is retail level brand-sales models of response to sales promotions. This is
the direct modeling approach of sales response. Four types of models will be described in
this section. Each of these models will be exemplified with a model used in previous
- 47-
research. The four types of models are: (1) linear models, (2) multiplicative models, (3)
exponential models, and (4) time-series models.
3.4.1. Linear Models of Brand Sales
Kumar and Leone (1988) developed a model for studying the effect of store competition
on brand sales in the stores. They used the following linear sales-response model.
St(AP) == bo +b1 • ~(AP)+b2' ~(AS)+b3' ~(CP)+b4' ~(CS)

(3.7)
+bs . [p; (AP) - P; (AS)] + b6 • [Dt(AP) - Dt(AS)]
where
St(AP) = sales of brand A in the primary store P period t

~(AP) = price of brand A in the primary store P period t
~(CS) = price of brand C in the secondary store S period t
F; (-) = dummy variable indicating feature advertising period t
Dt (-) = dummy variable indicating in-store special display period t
=model parameters
This model must be seen as primarily a research model used for testing whether or not
sales promotion has a statistically significant impact on sales of the promoted brand as well
as other brands in the promoting store and in a competing store. The model does not seem
to be suited for managerial use because of its lack of detail and its linear structure. A
drawback of the Kunlar and Leone model is that store traffic is not included in the model.
3.4.2. Multiplicative Models of Brand Sales
An important multiplicative model is the SCAN*PRO model developed by Wittink et

al. (1987). According to Blattberg and Neslin (1990), the model was designed to help
manufacturers understand the retailers' decision-making. The SCAN*PRO model is:
J
S(i,l,t) == a 11. . rr[p(i,j,t){3ij . rD~I:i,j,t).
1,I,J
rD~2:i,j,t) . rD~3:i,j,t)]. exp(u. )
2,I,J 3,I,J 1,I,t
(3.8)
j=1
where
S(i,1,t) unit sales for brand 1 in store i, week t

P(i,j,t) unit price for brand j in store i, week t
D(1,i,j,t) indicator variable for feature only, brand j, store i, week t
- 48-
D(2,i,j,t) indicator variable for display only, brand j, store i, week t
D(3,i,j,t) indicator variable for display and feature, brand j, store i, week t
u(i,l,t) error tem1
a,{3,y model parameters
The SCAN*PRO model is estimated at the store level, giving store specific estimates
for the promotional response parameters. This is to avoid the aggregation bias that would
occur if the model was estimated on data aggregated across stores.
An interesting feature in the SCAN*PRO model is the way display and feature
advertising are modeled. This modeling approach recognizes that a special display in
combination with a feature advertisement may cause an interaction effect on sales.
Prices of competing brands are included in the model at the store level. That is, the price
competition is a competition within the store. Only one price variable is used per brand.
The price variable gives the actual price of the brand. The different effects of regular prices
and price deals are not modeled separately.
The marketing research giant A.C. Nielsen uses the SCAN*PRO model. According to
Parsons et al. (1993), A.C. Nielsen's model has been used in 500 different commercial
applications. Another model that has had considerable commercial success is the Abraham
and Lodish (1993) PROMOTIONSCAN model. Information from PROMOTIONSCAN is
bought by approximately half of all major packaged goods marketers in the u.s. (Parsons
et al. 1993). We shall discuss the PROMOTIONSCAN model in the description of time-
series models.
3.4.3. Exponential Models of Brand Sales
The Blattberg and Wisniewski brand-sales response model was developed to be used by
retailers. The sales response function is:
Sit = exp[a /31' Rit /32 .~+

+ + 1 Dit
I(8 J+ ±r
k:Fl
k ._1
Itt J=1
j ' A;,j,t
(3.9)
+l/11 . Xi,t + l/12 . E9i,t + l/13 . NF;,t + A· 1;,t + l/14 . Ci,t + Cit]
where
= unit sales of brand i in period t

regular price of brand i in period t
actual price of brand i in period t
deal discount, defined as (~t - Rit ) / Rit
- 49-
Ai,j,t = dummy variable for A, B, and C advertisements
Xi,t = display activity (number of stores that display the brand)
E9 i,t = dummy variable for promotion ending in 9 (e.g., $1.99)
NFi,t = dummy variable for an "Nfor" promotion (e.g., 3 for $1.19)
1;,t = time since the promotion began with 0 being the first week
Ci,t = scanner week/deal week correction
£i,t = error term
a,f3,r,~,l/J,A = model parameters
Blattberg and Wisniewski include regular price and the deal price in the model, because
the responses to these differ. The price deal discount enters the nl0del as a fraction of the
regular price. This is important because problems with multicollinearity could arise
otherwise. Other brands' prices in the store are also included in the model. Only one price
variable was used for each of the other brands. Advertising was modeled using dummy
variables indicating the type of the advertisement.
The model was developed for a retailer with seven stores. Therefore, the special display
factor was modeled as the number of stores that displayed the brand. An alternative would
have been to have store-specific dummy variables. Dummy variables were used to model
the effect of "N for" (Le., e.g., 3 for $1.19) and "Ending in 9" promotions. The deals often
nln over several weeks, and a decay variable was included to account for the decaying
effectiveness of the promotion over the weeks.
A variable was included in the model to adjust for the mismatch between the scanner
week and the promotion week: the scanner week started on Sunday and ended on Saturday
while the promotion week started on Thursday and ended on Wednesday. The weight used
for the adjustment was the share of weekly sales that Thursday, Friday, and Saturday
account for on the average.
Blattberg and Wisniewski (1989) used this model when studying price-induced patterns
of competition among brands (or items).
3.4.4. Time-Series Models of Brand Sales
PROMOTER is a system for automatic evaluation of manufacturers' trade promotions. It

was designed by Abraham and Lodish (1987). The model in PROMOTER is a one-product
model based on time-series decomposition.
Set) = T(t)· SI(t)· X(t)· [bet) + pet) + e(t)] (3.10)
where
- 50-
Set) = sales level at time t
T(t) = trend of sales not due to promotion at time t
SI(t) = seasonal index of consumer demand at time t
X(t) = exception index representing the effect of special factors at time t
bet) = base-level sales after trend, seasonal, and exception index adjustment
pet) = promotion effect at time t
e(t) = noise term at time t
Adjusted sales is defined as:
AS (t) = Set) (3.11)

T ( t ) . SI (t) . X ( t)
Inserting (3.10) into (3.11) gives:
AS (t) = bet) + pet) + e(t) (3.12)
In other words, adjusted sales is the sum of baseline sales, incremental sales due to sales
promotion, and a noise term. The promotion effect index pet) is defined to be zero during
normal, i.e., nonpromotion, periods. The exception index X(t) is modeled as the product of
a number of exception sub-indices. Each sub-index equals unity during periods when that
exception does not occur. When the exception occurs, the sub-index assumes a value
representing the effect of that exception on sales. Examples of events modeled by sub-
index are all commodity volume distribution, the number of trading- days, the introduction
of new items, price changes, major competitive deals, sales force incentives, strikes,
shortages, and product recalls.
Abraham and Lodish (1993) describe an implemented system for improving promotion
productivity. The system, PROMOTIONSCAN, is a brand management model similar to
PROMOTER. The main difference is that PROMOTIONSCAN is inlplemented using
store-level scanner data whereas PROMOTER uses shipment data.
3.5. Market-Share Models

Applying the indirect modeling approach, the market shares of the items or brands nlust
be modeled. Market-share models call for somewhat more complex structures than brand-
sales models in order to be robust. Robustness or logical consistency criteria are important
when specifying market-share models.
A logically consistent market-share model should ensure (1) that predicted market
shares are not negative or greater than one and (2) that the sum of market shares should at
- 51 -
all times be one. These criteria are not satisfied by the linear model, the multiplicative
model, the exponential ll10del, and the time-series model presented in the previous section.
This section presents the attraction model, a type of model that satisfies the robustness
criteria for market-share models. First, we describe the general formulation of the attraction
model. Then, some specific attraction-type models used in previous research are described.
3.5.1. Attraction Models

The attraction model is a class of models satisfying the logical consistency criteria of
market-share models (Naert and Leeflang 1978).
In the general formulation ofan attraction model, the market share of brand i is:
m.=~
It J
(3.13)
Laj=l
jt
where
ait = attraction of brand i period t
A brand's attraction is a non-negative function of the relevant marketing variables. The

market share of brand i is given by the ratio of the brand's attraction and the sum of all
brands' attractions.
The attraction function can be specified in different ways. One specification is the
multiplicative competitive interaction model (MCI) of Nakanishi and Cooper (1974). The
attraction function in the MCI n10del is a multiplicative model. The market share of brand i
is given by:
K
ai • IT Xetk
mit =~
J
k=l
(3.14)
La
j=l
jt
This model was thought to be intrinsically nonlinear until Nakanishi and Cooper (1974)
showed that a linearizing transformation exists. The first step of the transformation is to
divide the brands' market shares pairwise. In this way we get rid of the complicated
denominator, since the sum of all brands' attractions obviously is the same in all brands'
market share expressions.
- 52-
ai·rr X~:t
K
m. _ _ a·l_,t_
_l,_t ~
a'
l _ _ a.l,t_ _ k=1
(3.15)
mi+1,t
-J
~a.
J
~a.
-
a i+1,t
-
a i+lrr K
Xf3k
L..J J,t
j=1 / L..J j,t
j=1
.
k=1
ki+l,t
The second step is to take logarithms. This gives
(3.16)
Rearranging the terms makes the equation look simpler.
In _m.l ,t)
_ = In (_
[ m + 1,t
i
l _ + L..J
a·
a i+1 k=1
J f[ [13k .In __ · t )]
X kl,_
X ki + 1,t
(3.17)
Nakanishi and Cooper (1982) provided a simplified estimation procedure for the model.
They showed that regression analysis using dummy variables can be used to estimate the
MCI model. Cooper and Nakanishi (1988) provide an excellent description of different
formulations of the attraction model. They show that the logit model is a special case of the
attraction models, where the attraction function is an exponential model.
For an interesting debate over the pros and cons of attraction models please refer to:
Naert and Bultez (1973), Beckwith (1973), McGuire and Weiss (1976), Naert and
Weverbergh (1981), Brodie and Kluyver (1984), Gosh, Neslin, and Shoemaker (1984),
Leeflang and Reul (1984), Naert and Weverbergh (1985), and Wittink (1987). The main
advantage of the attraction model is that the logical consistency criteria are satisfied. Its
disadvantage is that the number of parameters to estimate rapidly increases with the
number of brands and marketing variables (Hanssens et ale 1990).
3.5.2. Exanlples of Attraction Models
Surprisingly few market-share models have been reported in the recent literature. Most
recent studies use panel data to estimate individual choice probabilities using logit and
multinon1ial logit models. Important examples of such studies are Guadagni and Little
(1983) and Gupta (1988).
A store-level application of the attraction model to the n1easurement of brand market-
share is the shelf space allocation model by Bultez and Naert (1988). An attraction model
is estimated and used to determine the optimal shelf space allocation. A brand's attraction
is given by:
a.l = a.·
l
S~i
l
(3.18)
- 53 -
where
= attraction of itenl i
= shelf space allocated to item i
= model parameters specific to brand i
Bultez et ale (1989) extended the model to incorporate the effects of asymmetric
competition in retail assortments. This was done by assuming that items of the same variety
are more likely to compete with each other than items from different varieties, ceteris
paribus. Similarly, items from the same brand are likely to be closer substitutes than items
from different brands. They showed that a space allocation based on the symmetric
competition model was inferior to an allocation allowing asymmetric competition.
Grover and Srinivasan (1992) used a logit model to model the variations in store-level
market shares caused by sales promotion. The market share of brand j in period twas
modeled as:
~ (3.19)
J
La
1=1
lt
(3.20)
where
a jt = attraction of brand j period t

Pjt price of brand j period t
F jt
= feature advertising indicator brand j period t
C jt = coupon indicator brand j period t
D jt = display indicator brand j period t
B gjt = dummy indicator of brand, 1 when g=j, 0 otherwise
ak = model parameters
Foekens et al. (1992) described the nested multinomial logit (NMNL) modeling
framework. In the NMNL model, the choice outcomes at each level in the hierarchy are
dependent on the alternatives' attributes defined at that level and a summary measure of the
- 54-
attractiveness of lower levels of that branch. The summary measure is named the inclusive
variable of the alternative. The dependence of higher level models on lower level models
implies that the model estimation must take a bottom up approach, starting with models of
the lowest level decisions. Assuming that the choice of a variety of a brand is the lowest
level choice, the variety its share of brand j's sales is given by:
Sijkt _
eXP(La
m
mjk • X mijkt )
(3.21)
sj/a - ~exp(~amjk • X mhjkt )
where
X mijkt = the mth marketing variable of variety i of brand j in category k

period t
= response parameter associated with the mth marketing variable of
brandj in category k
At the next higher hierarchical level (i.e., the brand choice in this example), brand j's
share of category k is:
(3.22)
where
J:jkt =the nth marketing variable of brandj in category k period t

f3nk = response parameter associated with the nth marketing variable in category k
I jkt = the inclusive variable of brandj in category k, period t
The inclusive variable is a summary measure of the attractiveness of all varieties of the
brand. It is defined as
(3.23)
- 55 -
which is the logarithm of the denominator in the variety share model. In attraction
model terms, it is the logarithm of the sum of all varieties' attractions. Modeling the next
higher hierarchical level is a straightforward extension of this method. At each hierarchical
level, the marketing variables relevant at that level and the inclusive variables defined at
the next lower level are included in the model.
3.6. Category Sales Models

Cooper and Nakanishi (1988) suggest that a store's sales of a product category can be
modeled using the following general n10del:
(3.24)
where
Qt = category sales in period t

Pjt = price of brandj in period t
X kit = marketing variable k for brand j in period t
ut = error term
a,b,c = model parameters
This model allows all brands and marketing variables to have a fferent impact on
category volume. It is a flexible model which contains a general interaction among the
marketing variables. This general interaction may be disturbing when applied to a product
category. An example will show why.
Think of a promotion each for brand A and brand B that, if run when no other
promotion is run, will increase category sales by 50 percent, or., multiply category sales by
150 percent. What would category sales be if both promotions were run the same week?
Intuition says that sales would be in the range of 150 to 200 percent of baseline sales. The
model proposed by Cooper and Nakanishi predicts 225 percent of baseline sales. However,
if the restriction that only one brand at the time may be promoted is imposed on the model,
then the positive interaction is no longer a problem.
The lagged sales term is questio~able since the model does not include last period's
marketing activities. Large category sales in the previous period may be due to a seasonal
factor or to some promotional activity. The implication for the relation between the
- 56-
previous period's sales and sales in the present period would be rather different in these two
cases.
Mason (1990) modeled product-class sales as a function of the product-class attraction.
Although Mason only used product characteristics in her model, she suggested that
marketing mix variables also could be included in the attraction measure. Mason defines
four alternative operationalizations of the total attraction of a product category. Total
attraction as the sum of the individual products' attraction implies that anyaddition to the
product category would increase its attraction. In the averaging model, a category's total
attraction is the average attraction of the individual products. Adding a product with below
average attraction would decrease the total attraction. The maximum model defines total
attraction as the highest individual attraction in the category, while the minimum model
defines total attraction as the attraction of the least attractive product in the category.
Mason used the maximum model to model product-class sales. The sales-response
model was
Q = L+(U _ L)._ex_p(:-b._TA:-)_ (3.25)

B + exp(b· TA)
where
Q = product-class sales
L = lower limit on demand
U = upper limit on demand
TA = total product-class attraction
B, b = model paran1eters, (B>O)
Grover and Srinivasan (1992) modeled a store's share of the sales in a product category
using an attraction model. The total category sales was then modeled as a function of the
category's overall attractiveness. They defined the category's attractiveness as the log of the
sun1 of the stores' attractiveness with respect to that category. A store's attractiveness with
respect to a specific category was modeled as the log of the sun1 of the individual products'
attractiveness. That is, the store's attractiveness is calculated in the same way as the
inclusive variable used as input at lower hierarchical levels of their model
a jt = eXP(fak . X kj,) (3.26)

k=l
- 57-
(3.27)
where
a jt = attraction of brand j period t in the store

at =the store's attraction with respect to the category
X kjt =determinants of brand attractiveness
ak =model parameters
Grover and Srinivasan note that this model has the intuitively appealing feature that a
price cut of 50 cents on one brand contributes more to the total attractiveness than a price
cut of 25 cents on two brands (assuming that the brands are equally preferred).
A drawback of Grover and Srinivasan's (1992) approach of linking attraction at different
levels is that they constrain the effects of sales promotion on the store's category sales to be
directly related to the gain in market share within the category. This seems to be an
unnecessarily restrictive assumption.
Grover and Srinivasan model the store's share of the sales in one product category.
Using a similar modeling approach, it should also be possible to model a category's share
of a store's sales.
3.7. Store Sales and Store Traffic

This section presents ll10dels that have been used to study sales promotion's impact on
store sales and store traffic. Walters and Rinne (1986) used a linear model to investigate
the impact of double couponing and loss leader portfolios on store sales. They estimated
the following model:
10
St = a+ 8· H t + 1]' M t + L "i' P;t + Ct

i=l
(3.28)
where
St = store sales in dollars, week t

Ht = dummy variable indicating holiday weeks
Mt = dummy variable indicating double-couponing in week t
P;t = dummy variable indicating use of loss leader portfolio i week t
Ct = error term
- 58 -
a,D, 17, r = model parameters
The linear model with dummy variables is a rather crude model that would not be useful
as a management tool. However, the aim was to test whether the pronl0tions have impact
on store sales and for that aim the nl0del may be good enough. A methodologically
interesting issue is that they concluded that sales promotion has no effect because the
model parameters were insignificantly different from zero. A better model might have had
the power to reject the null hypothesis of no effect. Store traffic and store profit were
modeled with the same model as store sales, but with other model parameters.
Walters and MacKenzie (1988) provided a structural model for studying the impact of
sales pronl0tion on store profits. Figure 3.2 presents their structural model. This model will
be examined here in some detail because it provides a useful framework for analyzing the
profitability of retail promotions. However, the model has some undesirable features which
shall be addressed.
Figure 3.2. The effects ofprice promotions on store performance according to Walters and
MacKenzie (1988)
- 59-
This model explains the variation in store profits with variation in promotional efforts
and seasonal variation. Loss leader sales, non-promoted sales, coupon sales, and in-store
promotion sales are the variables that directly affect store profits in this model. It is
important to note two things about these four sales measures. First, sales is measured in
dollars and not in quantity. Second, it appears that the four sales variables measure the
dollar value of products sold with a certain type of promotion (or without promotion). For
example, loss leader sales are sales of products promoted as loss leaders that week while
nonpromoted sales are sales of products that were not promoted that week. The weeks an
item is on in-store promotion, its sales are added up together with sales of all other
products on in-store promotion. Other weeks, when the item is not promoted, its sales enter
the calculation of the non-promoted sales variable. This means that part of the variation in
the non-promoted sales variable is due to the composition of the basket of non-promoted
products. Another source of variation is the number of products on deal. If all products
would be on deal, non-promoted sales would of course be zero. This is not a desirable
feature of the model because it makes the results difficult to interpret and makes some
results trivial. For example, that in-store specials increases sales of products on in-store
promotion is not too surprising.
A very interesting feature of the model is that store traffic is explicitly included in the
model. Store traffic is a determinant of sales of all products in the store, and is
consequently an indirect determinant of store profits. The model measures the effect of loss
leader promotions and double couponing on store traffic. In-store specials are assumed to
have no impact on store traffic in the short run. The effect of the December holiday season
on store traffic is measured.
The modeling of non-promoted sales as a function of loss-leader sales, coupon sales,
and in-store promotion sales is questionable because we can expect a positive and a
negative effect simultaneously. The negative effect between promoted sales and non-
promoted sales arises because of substitution and because of the way the variables are
defined (as was discussed above). The positive effect would arise from variation in store
sales that is not caused by store traffic variation. For example, the season may affect the
average purchase. It would be better to model the effect of sales promotion on non-
promoted sales directly rather than indirectly through promoted sales.
Walters (1988) used a n10del similar to the Walters and MacKenzie (1988) model. An
important improvement in Walters's model is that he models sales for each product
category separately. Unfortunately, within each category, he measures sales of promoted
and unpromoted products in the same way as Walters and MacKenzie (1988). The
conceptual model of promotional effects suggested by Walters (1991) (see Figure 3.3) is
from this aspect more appealing.
- 60-
Promotional Activities on Sales of
Store 1
Product i
Store 1
Complements
of Product i
I Product i
Substitutes
of Product i
Complements
Store 2 of Product i
Product i Substitutes
of Product i
Complements
of Product i
Substitutes
of Product i
Figure 3.3. Walters's (1991) conceptual model of retail promotional effects
The drawback of this model is that store traffic is not included in the model. It would
have been very interesting to see how store traffic and store sales were affected by the
stores' promotions.
3.8. Summing Up
The aggregate results of consumers' purchase decisions translate into retailer sales. In
the previous chapter, it was found that sales promotion affects the outcomes of consumers'
purchase decisions. It can therefore be expected that sales promotion affects sales. This
chapter started by noting that previous studies have found that sales promotion has a large
effect on sales of the promoted item. The possible sources of the retailer's incremental sales
of the promoted item were identified as other items, other categories, other stores, and
other time periods. Note that these sources can be visualized in three dimensions as
product, store, and time.
Empirical results reported in the literature suggest that other items (or brands) in the
same category, in the same store, at the same time, constitute an important source of
- 61 -
promotion-induced incremental sales volume. Concerning other categories, previous
research has focused on complements rather than substitutes. Consequently, no study
examining whether other product categories' sales suffer from promotions was found. This
is clearly a knowledge gap that needs to be filled.
Further, other stores have been found to be a source for incremental sales volume.
Promotion in one store has been found to have impact on competing stores' sales of the
promoted item and its close substitutes. Such effects may be either direct, by affecting store
traffic, or indirect by affecting the consumers' timing of the purchase. Indirect effects may
occur because households are not completely loyal to one store.
The previous chapter reported that sales promotion has been found to affect consumers'
purchase timing and that household inventories moderate that effect. However, the research
presented in this chapter found that the retailer's sales of the promoted item in periods
following the promotion did not suffer. This suggests that retailer-oriented models of sales
promotions can disregard effects on future sales.
We have seen that retail sales can be modeled at different levels of demand. Further,
brand sales can be modeled directly as well as indirectly as the product of category sales
and market share. A number of models of retail sales' response to promotions were
described. These included models of promotional brand-sales response, market-share
response, category-sales response, store-sales response, and store-traffic response.
However, none of the models included all of these important responses.
- 62-
4. PROMOTION PROFITS
4.1. Introduction
A basic principle in economic decision theory is that only the incremental or marginal
effects of a decision are relevant. Costs and revenues that are equal across decision
alternatives can be disregarded when evaluating the outcomes of the decision alternatives.
Promotional decision-making should be no exception from this basic economic principle.
Thus, the profit of a promotion should be evaluated on an incremental basis, comparing the
change in profits caused by the promotion. Evaluating a promotion ex post implies that the
profits that would have been if the promotion had not taken place must be estimated.
As will be seen in this chapter, several authors have suggested models for defining and
measuring promotion profits. All the models presented here define promotion profits on an
incremental basis. However, the models differ in the definition of the relevant incremental
costs and revenues.
Although the main focus in this report is profits of retail promotion, models and results
concerning trade promotion are presented as well. The reason for this is that these models
and results provide additional insight because they have important features in common
with store-related models.
4.2. Profit Impact of Sales Promotion

This section presents results and empirical evidence concerning the profit impact of
sales promotion as reported in previous research.
4.2.1. Manufacturer Profits
Only a small number of studies investigating the impact of sales promotion on

manufacturers' profits have been reported. The results of these studies are discussed here.
Strang (1976) called attention to the increasing use of sales promotion and the lack of
effect evaluation. The effects on profits were particularly unknown. He showed an example
from a consumer goods market where an increased usage of sales promotion led to higher
sales but lower profits.
- 63 -
Hardy (1986) found that only 30 out of 45 promotions that managers reported as
successful were actually financially successful. At the same time, 19 out of 46 promotions
reported to be overall failures increased profits. Hardy explained the discrepancy between
the reported degree of success and incremental profits with the managers' concern about
sales volume.
Blattberg and Levin (1987) developed a method to measure the profitability of trade
promotions. The model shows the importance of incremental consumer sales for the
profitability of trade pron10tions. If consumer sales do not increase, incremental shipments
only mean increased pipeline inventories. The retailers' pass-through of the trade deal to the
consumers is therefore paramount for trade promotion profits. Blattberg and Levin found
negative net profits for the manufacturer for most trade promotions.
According to Abraham and Lodish (1990), only 16 percent of the trade promotion
events they studied were profitable. They found that the cost of obtaining one dollar extra
sales often exceeded one dollar. The following example of a hypothetical trade promotion
was presented to illustrate the weak results of trade promotions. They assume weekly base
sales of 100 cases and a list price of $10 per case.
Cases Gross
Dollars
Baseline sales that would have occurred during the four weeks 400 4,000.00
+ Incremental sales to consumers 430 4,300.00
+ Ten weeks of forward buying by reta.ilers 1.000 10.000.00
Total sales during promotion 1,830 18,300.00
Cost of promotion ($18,300x 15% discount) 2,745.00
Cost per incremental dollar of sales 9.64
Table 4.1. The cost of incremental sales according to Abraham and Lodish (1990).
In their hypothetical example the cost of one dollar's incremental sales is 64 cents. If the
manufacturer's margin is less than that, the promotion would not be profitable. The high
cost of the incremental sales arises because the manufacturer gives the discount also on the
normal sales. Abraham and Lodish stress that although this is hypothetical example, it is
based on extensive empirical experience. Further, they claim that the example gives a
conservative estimate of the manufacturer's cost of increased sales.
Jones (1990) tabulates the sales and profit outcomes of a 10 percent price discount for
different combinations of variable costs, net profits, and price elasticities. The levels are
chosen to reflect typical costs, profits, and elasticities for real brands. Although the effect
on sales volume is positive and fairly large, the price discount reduces profits in 21 out of
- 64-
the 24 cell entries. The loss would be even greater if the mortgage effect on future sales
were included in the calculation.
Neslin (1990) measured the incremental sales due to manufacturers' coupon promotions.
Incremental sales were shown to be important for the profitability of couponing. None of
the seven instant coffee brands studied by Neslin reached the levels of incremental sales
that would be necessary to make the coupon promotion profitable.
In summary, the literature suggests that trade promotions in general are unprofitable.
One reason for this seems to be that sales volume is used to evaluate the performance of
sales promotion, although sales volume and profits behave differently. Blattberg and Levin
(1987) put forth the explanation that many trade promotions are defensive rather than
offensive. The gain from a promotion would then be that the decline in profits is reduced.
4.2.2. Retailer Profits
Empirical reports on the profitability of retailer promotions are scarce. Doyle and
Gidengil (1977) complained that most in-store experiments up to that time had failed to
consider "even the basic characteristics of retailing economics". These characteristics are
the high cross-elasticities of demand between items in a· supermarket, the low gross
margins, and the high variable costs. These characteristics imply that store profits are
inelastic to volume changes but elastic to price changes. A second implication is that
isolated item elasticities are of limited value for the retailer who is more interested in the
impact on overall store profitability.
One attempt to measure the impact of a retail couponing on retail profits was done by
Chapman (1986) who conducted a low-cost field experinlent for a pizza parlor's
couponing. He found that the couponing increased the pizza parlor's profits by five percent
during a six-week period. The incremental profit corresponded to eleven percent of the
fixed cost of the couponing. A sensitivity analysis showed that the profits could be
increased considerably if the fixed cost could be reduced.
Walters and Rinne (1986) tried to measure the impact of double couponing and loss
leader portfolios on overall profits. They recalculated the gross margins for each product
every week in three supermarkets. The weekly store profit was calculated as the sunl of the
gross margins times sales volume for all products, minus labor and material costs. They
found that double couponing had a significant impact on store profits in one of the three
stores. Loss leader portfolios were found to affect store profits significantly in only four
cases. Walters and Rinne concluded that loss leader portfolios and double couponing might
leave profits unaffected although the impact on store sales may be significant.
This study by Walters and Rinne is interesting because it is the first attempt to measure
the impact of double couponings and loss leader promotions on store profits.
- 65 -
Unfortunately, their model is too aggregate to provide insight into the mechanisms by
which sales promotion affects retail profits.
Dhebar, Neslin, and Quelch (1987) measured the incremental retailer profit from a sales
promotion launched by a automobile dealer. They found that the increased sales volume
more than offset the lower margins and the fixed costs of the promotion. The immediate
result of the promotion was increased profits for the automobile dealer. When adding the
estimated long-term effects on service and repeat purchases, the promotion was even more
profitable.
However, the automobile dealership setting is fairly remote from grocery retailing. The
long-term effects are certainly more important for the automobile dealer than for the
grocery retailer. The effects of the promotion on the dealership's sales of other cars and
sales of the promoted model in the periods after the promotion was negligible and not
included in the model.
Walters and MacKenzie (1988) can be seen as an extension of the study by Walters and
Rinne (1986). They use a similar way of measuring store profits. Weekly store profit is the
dependent variable in a structural equations model relating store profit to loss leader
portfolios, double couponing, and in-store specials. Store traffic is an endogenous variable
that is affected by the independent variable. The model also tests the impact of store traffic
on store profit. They found that most loss leader promotions had no effect on store profits.
The loss leaders that had an effect on store profit did so because they created incremental
store traffic. Double couponing, on the other hand, affected store profits by increasing sales
of products sold with coupons rather than affecting store traffic. In-store specials did not
have any impact on store profits, store traffic, or store sales.
Walters (1988) used a model similar to Walters and MacKenzie (1988) in a fabric store
setting. He found that sales promotion had a positive effect on store profits by their impact
on store traffic. Sales of promoted products had a negative impact on store profit because
of their depressed profit margins.
Inman, McAlister, and Hoyer (1990) found that consumers respond also to signal-only
promotions. Based on this observation, Inman and McAlister (1993) developed a
promotion-policy model for retailers. They found that a retailer can improve category
profits by offering negligible price discounts accompanied with a promotion signal, e.g., a
special display and a one percent price cut. The model draws attention to the effect of the
signal-only promotion. However, their resulta are a consequence of the restrictive
assumptions made. Promotions are assumed to have no effect on category sales volume;
their only effects are on the market shares of the brands within the store. Inman and
McAlister also assume that the retailer can improve gross profits by absorbing the trade
promotion.
- 66-
Gijsbrechts (1993) made an extensive review of pricing and sales promotion research in
consumer marketing. She concludes by giving directions for future research, stating the
importance of studies on pricing and retail profits:
"Theoretical and empirical insight is needed on the impact of price
strategies on intra- and interstore brand substitution, and on sales of
complementary products. The role of price in store choice and the
formation of store price images should be further investigated, since they
are closely linked to retailer profitability".
In summary, it has been found that sales promotion may have a positive impact on the
store profit if it increases store traffic and thus the retailer's sales of other products. Store
profits may also increase if the retailer can improve the margins by absorbing some of the
trade promotion. More research is clearly needed in this important area.
4.3. Manufacturer Induced Promotions

This section presents models for measuring the manufacturer's profits of manufacturer
induced promotions. First, a normative framework developed by Aaker is presented. Then,
trade promotion models are discussed, followed by models of the profits of couponings.
4.3.1. A Normative Framework
Aaker (1973) developed a framework model for normative promotional decision-

making. The core of the framework is the objective function to maximize, the net present
value expected from a promotion
select 0 to max V(O)

(4.1)
subject to: V(O) > 0, Budget Constraints
where fJ is a decision parameter vector describing the promotion and V( fJ) is the net
present value of the promotion. The parameters included in Aaker's decision parameter
vector are (P) the promotional vehicle or type, (s) the size or unit value of the promotion,
(x) describes the target audience, and (t) is the number of months since the last promotion.
The vector can be expanded or modified to fit the problem at hand. The objective function
is specified as:
V(O) = S(O) - A - C(O) + L(O) + W(O) (4.2)
where
- 67-
see) = total gross nlargin contributed by the brand during and just after the
promotion
A = total gross margin that would have been generated in the absence of
the promotion during a comparable time period
c(e) = direct promotion cost, excluding the unit cost
L(e) = the value of any increase in the loyalty among existing customers as
a result of the promotion
wee) = the long-run value of a group of new triers of a brand attracted
during promotion
We are interested in the short-run effects of sales promotion and will concentrate on the
first three terms of the equation. While the direct promotion cost is straightforward to
calculate, the difference S(e) - A nlust be estimated. It is important to note that the
"immediate effect" in Aaker's model is measured "during and just after" the promotion. In
this way, Aaker includes a possible post-promotional dip in the model.
Aaker developed the model from a manufacturer's viewpoint, but stresses that it would
be identical for an analysis of the retailer's situation. The model was developed to structure
the thinking about the immediate and long-run effects of sales promotion. For our
purposes, the model does not contain enough detail concerning the short-run effects of
sales promotion.
The two most important aspects of this model are (1) that it clearly defines the profit of
a promotion as the change in profits caused by the promotion, and (2) that its essence is
that all such changes should be considered.
4.3.2. Trade Promotion Profits

Hardy (1986) also used the change in profits as the measure of the profit of a promotion.
He used the following equation to calculate the profits of a trade promotion:
Incremental Profit = [S(p)- c].q(p)- [S(r)- c]·q(N)- c(p) (4.3)
where
S(P) = selling price during the promotion

S(r) = regular selling price
c = average variable cost of product
q(P) = sales volume during the promotion
q(N) = normal sales volume
- 68 -
C(P) = fixed promotional cost
It is not clear from Hardy's equation how the promotion period and the normal sales
volume are specified. However, it appears that the promotion period refers to the period
when the promotion was run, not including the period inlnlediately after the promotion, as
Aaker suggested.
Blattberg and Levin (1987) used the following method to compute the profitability of a
trade promotion. They computed the average unit price charged per period since not all
units were sold with the promotional allowance.
Econometric models and historical data were used to estimate the sales volume that
would have occurred had the promotion not taken place. The incremental unit shipments
due to the promotion were computed as the cumulative difference between actual unit sales
and projected unit sales. Incremental dollar shipments were calculated by multiplying the
period's incremental unit sales by the relevant price. Incremental gross profit was computed
by multiplying the monthly incremental dollar shipments by each month's product gross
margin. The deal cost was computed by multiplying the total promoted shipments sold
during the period by the dollar discount.
Blattberg and Levin (1987) provided the following example to illustrate the profit
calculation. The regular price of the product was $12, its gross margin $6, and the discount
$1. During normal periods, sales were 20,000 units per period.
Period Total Percent on Deal Cost * Incremental Cum. Incremental

Shipments Deal Shipments Shipments
1 50,000 80% 40,000 30,000 30,000
2 30,000 80°,10 24,000 10,000 40,000
3 25,000 800/0 20,000 5,000 45,000
4 30,000 80% 24,000 10,000 55,000
5 35,000 50% 17,500 15,000 70,000
6 5,000 0% 0 -15,000 55,000
7 5,000 0% 0 -15,000 40,000
8 10,000 0°,10 0 -10,000 30,000
9 10,000 0% 0 -10,000 30,000
10 15,000 00/0 0 -5,000 15,000
11 20,000 0% ° 0 15,000
Total
12 20 000
255,000
00/0 °
125.500
0
15.000
15,000
15,000
Table 4.2. Calculating profits from a trade promotion. *) Deal Cost = Total Shipments x
Percent on deal x Discount
The total deal cost was $125,500. Incremental unit sales due to this promotion were
15,000 units. With a gross margin of $6 per unit the total additional gross margin was
$90,000. Thus, the manufacturer lost $35,500 on this sales promotion.
- 69-
A similar model was used by Abraham and Lodish (1990). Their model emphasizes the
importance of incremental consumer sales for the profitability of trade promotion (see
Table 4.1). Retailers' forward buying explicitly enters the cost calculation. They calculate
the cost of an additional dollar's sales as follows. Total sales volume during the promotion
consists of three parts: (1) baseline consumer sales, i.e., the sales that would have occurred
during the period even without the sales promotion; (2) incremental sales to consumers;
and (3) retailers' forward buying. The total sales value in gross dollars is calculated as the
sales volume times the list price.
Abraham and Lodish calculate the cost of the promotion as the total gross dollar sales
times the discount percentage. The incremental sales dollars are incremental sales to
consun1ers times the list price. Dividing the cost of the promotion by the incremental sales
dollars gives the promotional cost per incremental dollar sales. If the cost per dollar sales
exceeds the manufacturer's gross margin, the promotion is not profitable.
4.3.3. Couponing Profits
This sub-section presents two models developed to measure the profit of couponings.
Neslin and Shoemaker (1983) showed the importance of calculating the net incremental
impact of a coupon on consumer sales volume when evaluating the profitability of a
couponing. Neslin (1990) defined incremental unit sales per redemption as net incremental
unit sales to the number of redeemed coupons. The incremental unit sales per redemption
enters the profitability calculation. Neslin illustrated the profitability calculation with an
example calculation. The couponing profit calculation by Neslin was then formalized by
Blattberg and Neslin (1990). They define the following variables.
C = distribution cost per coupon

D = number of coupons distributed
F = face value of coupon
P = processing cost per redeemed coupon
R = redemption rate as fraction of distributed coupons redeemed
L = fraction of redemptions that are legitimate, i.e., not misreden1ptions
M = profit margin per unit sold (in dollars)
I = fraction of redemption sales that is incremental
The total incremental costs are the sum of distribution costs and redemption costs. The
total incremental profit is given by the number of incremental sales times the profit margin.
The net incremental profit of the couponing is the difference between the incremental
profits and the incremental costs. Expressed as an equation, the net incremental profit is:
-70 -
111T: = (D· R· L· I· M) - [C· D+ D· R· (F + P)] (4.4)
or
111T: = D.[R·(L· I· M - F- P)- C] (4.5)
The equations show that the profit margin from redemptions that represent incremental
sales is the only source of profits from a couponing. Given that the face value and the
processing costs do not exceed the profit margin from the incremental sales, a couponing is
more profitable when the redemption rate is high. If the incremental profit per distributed
coupon would exceeds the distribution cost, then the couponing is more profitable the more
coupons that are distributed. However, it is unlikely that the variables in this equation are
independent. The redemption rate may be lower when the total number of coupons
distributed increases. The distribution cost per coupon, on the other hand, is likely to be
lower for a large couponing than for a smaller drop.
Blattberg and Neslin (1990) show that the break-even point for incremental sales per
coupon is given by setting the incremental profit to zero and solving for I. The necessary
incremental sales per coupon to break even is:
[* = C + F+P =_l_.[C +F+P] (4.6)

R·L·M L·M L·M R
Equation (4.6) shows that the break-even point is reached at lower levels of incremental
sales per coupon when profit margins are high. The incremental sales per coupon must be
higher to break even when the coupon has a high face value, when the processing costs per
redeemed coupon are high, and when the distribution cost per redeemed coupon is high.
Although the model gives the incremental profit of a couponing, it could be used to
evaluate the profitability of a price deal. The number of coupons distributed would then be
equal to the total sales and the redemption rate would be unity. The discount would
correspond to the face value of the coupon.
Chapman (1986) measured the profitability of a pizza parlor's couponing. Although a
pizza parlor is a retailer, the model will be presented in this section because the model is
more related to n1anufacturer couponing than to retailer price deals.
The incremental profit from the couponing was calculated with the following equation.
G = u· c· S + (1- u)· m· S - m· B- F (4.7)
where
G = the incremental profit from the couponing
-71 -
u = the proportion of sales for which a coupon was used
c = the contribution margin (in percent) from coupon sales
m = the contribution margin (in percent) from sales without coupon
S = total sales revenue during the promotion
B = base sales revenue that would have occurred without the coupon
F = the fixed costs of the coupon promotion
The equation can be interpreted as the sum of the gross margin from sales ;with and
without the coupon, less the gross margin that would have been in the absence of the
promotion, less the fixed cost of the promotion. Chapman showed that the equation can be
rearranged to:
G = m· (S - B)- u· S· (m-c)- F (4.8)
The profit from the couponing is the regular margin from the incremental sales less the
face value of redeemed coupons less fixed costs of the couponing.
In summary, the profit of a couponing promotion is calculated as the change in gross
profits due to the couponing.
4.4. Retailer Promotions

This section presents models developed to measure the profits of retailer promotions.
The models are of three types: (1) models measuring only the impact of sales promotion on
the retailer's gross profits from the promoted item; (2) models considering also the
promotion's effect on other items in the category; and (3) models considering effects on
other categories as well.
4.4.1. Item Profits
Dhebar et al. (1987) developed a model for planning retailer sales promotions. The aim
of the model is to help an automobile dealer to design an individual promotion. The
retailer's profit is the focus of the model. Figure 4.1 shows the conceptual model for the
short-term impact of a sales promotion on profit. Dhebar et al. link the conceptual model of
short-term and long-term impact on profit through immediate sales. We will concentrate on
the short-term part of the model, which is relevant to the current purpose.
-72 -
ptions Purchasing R a t \
Model-Specific TraffiC\ Sales of op~
PROMOTION
Y Closing Rate
/
Immediate Sales
Short-Term Profit
Contribution
~ Delivery Acceptance Rate / \ I e s by MOd~1

Model Breakdown
/
Figure 4.1. Promotion's Short-Term Impact on Profit. Source: Dhebar et al. (1987)
Dhebar et at. (1987) model immediate sales of the promoted model (ISALES) as the
product of four factors: TRAFFIC is the nUlTlber of customers visiting the store per day;
RELTRAFFIC is the percentage of traffic that represent a potential sale of the promoted
model; CLOSE is the percentage of n10del relevant store visits that result in a sale; and
DELIVER is the percentage of closed sales accepted by the customer for delivery. Thus,
ISALES == TRAFFIC· RELTRAFFIC . CLOSE· DELIVER (4.9)
Each of the four factors is potentially affected by the sales promotion. They n10del the
sales of various options as percentages of immediate sales. The profit is calculated as
contribution margins less commission cost and fixed promotion cost. The profit of a
promotion is then calculated as the profit with the promotion less the profit that would have
been without the promotion. ill other words, the incremental profit due to the promotion is
calculated.
The model is a powerful tool for evaluating the effects of a sales promotion for an
automobile dealership. However, the effect of sales promotion on other cars than the
promoted model is assumed to be negligible and is not included in the model. Such an
assumption may be relevant for an automobile dealer, but in grocery retailing we must
include such effects in the model. Dhebar et ale indicate that the effect on sales of other car
models could be included by estimating the cross-elasticities of sales promotion.
Blattberg and Neslin (1990) describe how the optimal price discount for a product can
be determined. The profit function is defined as:
1C == (p -c+ DISA + TD)· q - D· q - DISC + FAA - FAC (4.10)
-73 -
where
p = regular retail price per unit

c = cost per unit
q = quantity in units
DISA = display allowance per unit
D = deal discount (in dollars)
DISC = cost of a display
FAA = feature advertising allowance
FAC = feature advertising cost
TD = trade deal in dollars per unit
Specifying a sales-response model to model unit sales (q), the profit maximizing price
can be calculated using standard calculus. The model can be used to analyze the retailer's
passthrough of trade deals. The higher the price deal elasticity, the larger the passthrough.
At high elasticities, the retailer will add their own money to the price deal. On the other
hand, when the elasticity is low, the retailer will absorb most of the trade deal.
In summary, the item profits modelspresented in this section show how the prmotions
affects the profit from sales of the promoted item. This is sufficient when only one item is
sold. However, the item profit models are of limited value for the grocery retailer
evaluating the profit of a sales promotion because these models fail to consider effects on
other iten1s. As there are complex interdependencies among the products in a retailer's
assortment, these models thus have to be extended to include the effects on other items.
4.4.2. Category Profits
Cannibalization within the category is an important issue that the item sales models
failed to consider. Blattberg and Neslin (1990) extended their item profit model (presented
above) to include interdependencies an10ng the items in a category. In this model, the
retailer's objective is to maximize the profits in the category. The category profit equation
is:
I I
n== I,m i ·qi - I,d i ·qi (4.11)

i=l i=l
where
= item i's gross margin per unit in dollars including the trade deal
= item i's deal discount per unit in dollars
-74 -
= quantity sold of item i
= number of items in the category
A sales-response function is defined for each item. The sales of item i are dependent on
other items' deals. The category margin and the degree of cannibalization determine the
optimal price discount for an item. When cannibalization is the dominant source of
incremental brand volunle, the optimal price discount is smaller. Thin margins also lead to
smaller optimal discount.
Blattberg and Neslin (1990) show that a model of this type can be used to determine
whether to display an item or not, or whether to include an item in the feature
advertisement or not. They argue that the direct cost of a display or a feature advertisement
may be approximately the same independent of which products are displayed or featured.
However, the retailer has limited space to display products in the store and the space in the
newspaper feature advertisement is also limited. Thus, the relevant costs of display and
feature advertising are the opportunity costs of display space and space in the feature
advertisement.
Inman and McAlister (1993) developed an interesting model of category profits of sales
promotion. The aim of their model is to maximize a retailer's category profits for a given
week. The model recognizes that signal-only pronl0tions 11lay affect consumers' brand
choice (Inman, McAlister, and Hoyer 1990). Given the brands' margins, available trade
deals, and consumer response to sales promotions for the brands, the retailer has to choose
which brand to promote, if any, and the discount level. The discount can be either
significant or negligible. Inman and McAlister refer to promotions with negligible price
discounts as signal-only promotions.
Inman and McAlister used a very restrictive response function to derive the profit from
the different promotions on the brands. They specified the negligible discount to 1 percent
and the significant discount to 15 percent. At these discounts, the sales response is the
same for all brands in a category, and is only dependent on the market shares of the brands.
The consumers are thought of as members of one of three segments: (1) brand loyals (never
react to any promotion); (2) switchers responding only to significant discounts; and (3)
switchers reacting to all promotions. In the three-brand setting in which the model was
tested, the market shares are conlpletely determined by six parameters: the market
preferences for each brand and the sizes of the three segments.
Inman and McAlister assume that sales promotion only affects brand-switching.
Category sales are thus not affected by sales promotion and other periods' sales are also
unaffected. Also, sales promotion has no effect on store traffic and there are no effect on
sales of other categories. Only one brand may be promoted each period.
-75 -
The retailer receives the offered trade promotion only if he promotes the brand, but
receives the same trade support per unit whether the brand is promoted using a significant
discount or a negligible discount. This means that the retailer can choose to absorb most of
the trade promotion or to pass it through to the consumers.
The most interesting aspect of this model is that it provides the retailer with a two-step
procedure for defining the profit-maximizing promotion in a product category; to select the
brand to put on a negligible discount promotion, the retailer has to calculate, for each
brand, the category profits that would occur if that brand were promoted with a negligible
discount promotion. The maximum category profits that can be obtained by using a
negligible discount can then be found. The same procedure is then repeated for the
significant discount promotion. Having found the maximum category profits from a
negligible discount as well as from a significant discount, the retailer calculates the
category profits that would occur if no promotion were used. Then, the strategy giving the
highest estimated category profit is chosen.
Although allocation of display space is not addressed with this model, it would not be
difficult to extend the model to this decision. The opportunity cost (or the value) of display
space is then important. The value of the display space when allocated to the product
category is the difference between the maximum category profit the retailer would gain
when a display is used in the category and the maximum category profit the retailer could
reach without the display.
In conclusion, the category profit models presented here are much more relevant for the
retailer than the single item n10dels presented above. While the model developed by
Blattberg and Neslin (1990) allows sales promotion to expand the category sales, the model
developed by Inman and McAlister assumes complete cannibalization within the category.
Both models ignore the possible impact on other product categories.
4.4.3. Store Profits
A completely different approach to measuring the effect of sales promotion on retailers'

profits was suggested by Walters and Rinne (1986). They measured weekly store profits by
adding the weekly gross margin for all products sold and subtracting labor and material
costs. The profit impact of sales promotion was measured by regression analysis with store
profit as dependent variable and the promotional activities as independent, or explanatory,
variables. Their model is important because it is the first attempt to measure the effect of
sales promotion on the total store profit. However, their model is aggregate and does not
address the question of how sales promotion works to affect retail profits.
Walters and MacKenzie (1988) carried the analysis of Walters and Rinne (1986) further
by providing a structural model for studying the impact of sales promotion on store profits.
-76 -
They calculated weekly aggregate store profits in the same way as Walters and Rinne
(1986). Their model was presented in Section 3.7.
These models are important because they represent unique attempts to make empirical
observations of the store level profit impact of sales promotion. The profit impact is
defined as the change in a store's gross profits resulting from a change in the promotion
portfolio. Unfortunately, the models are too aggregate to reveal the causal structure
transforming changes in the promotion portfolio into changes in store profits. This also
makes the models unlikely to reject a null hypothesis of no impact of sales promotion when
there is insufficient variation in the overall promotional portfolio.
These models are clearly research models and not models intended to be used by
retailers as decision support or decision models. The aim of these models is to test
hypotheses concerning the existence of a profit impact of sales promotions.
A very interesting modeling framework was suggested by Mulhern and Leone (1991).
They recognize the importance of a multiproduct perspective when studying retailer pricing
and promotion policies. Products within a product line are substitutes in their model.
Across product lines, products can be substitutes, complements, or independent. They note
that complementary relationships between products are of two types: products can be use
complements or purchase complements, or both. The retailer perspective makes purchase
complementarity important, and Mulhern and Leone note that purchase complementarity
may occur for all items in the store because consumers buy baskets of products when
visiting a grocery store.
A price deal thus influences sales of three groups of products in the Mulhern and Leone
model. While a price deal increases the sales volume of (1) the promoted iten1 and of (2) its
purchase complements, it reduces sales of (3) the item's purchase substitutes. The major
weakness of their model is that it fails to explicitly include store traffic. They test the
model on the cake mix and cake frosting categories where direct and cross effects for all
items within and across these two categories were tested for statistical significance. Effects
on other categories' items were implicitly assumed to be zero.
Mulhern and Leone calculate the profit impact of the promotion as the difference
between the profits with the promotion and the profits if the promotion were not run. They
define the promotion profit for the store as the sum of the change in profits for the
promoted item, the change in profits from substitutes, and the change in profits from the
complements.
In conclusion, the model developed by Mulhern and Leone is the most promising of the
models trying to measure sales promotion's impact on store profits. However, it would be
valuable to have a model that more clearly accounts for all effects outside the promoted
item's category.
-77 -
4.5. Summing Up
This chapter presented a number of models and definitions of promotion profits.
Common for all these models is that promotion profit is defined on an incremental basis,
i.e., as the change in profits caused by the sales promotion. A general observation is that
the models are adapted to the problem at hand in order to take the relevant incremental
costs and revenues into consideration in the definition of promotion profits.
The profit impact of sales promotion is to a large extent unknown. However, there is
evidence and logical arguments indicating that many sales promotions are not profitable (at
least not in the short run). It was also argued that consumers usually buy a basket of
products when visiting a grocery store. The consumer's store choice is consequently
important for the retailer as it means store traffic for the retailer. Empirical evidence
indicated that sales promotion's impact on store traffic is important for the profitability of
retail promotions. At the same time, there is evidence that sales promotion has small
impact on store traffic.
None of the models presented here seems to be ideal for a grocery retailer who wants to
measure the profit impact of sales promotion for individual items. The single product
models are clearly much too limited to catch the complex impact of sales promotion on
other items in the store. The category profit models are better but failed to look outside the
focal product category for effects on other categories.
The store level models developed by Walters and his colleagues include effects on other
product categories. The store level models that include store traffic, however, are too
aggregate to help the retailer. The most promising model is the model developed by
Mulhern and Leone. Their model is detailed and recognizes the importance of effects on
other categories, but fails to include store traffic.
The retailer perspective implies that store traffic must enter the model for two reasons.
First, store traffic is commonly assumed to be a driver of sales of all products in the store.
This could be tested, and the result would contribute to the retailers' and researchers'
knowledge. If store traffic is a determinant of sales, then any model of item sales would be
incomplete without taking store traffic into consideration. Second, some sales promotions
are assumed to influence store traffic, and thereby sales of unrelated products. The model
should measure this influence.
Thus, there is a need for a model defined for the retailer's specific decision problem.
Such a model should the consider all costs and revenues relevant for the retailer. It should
also contain considerable detail concerning the effects on other items and should explicitly
include store traffic. A model of that kind will be developed in the next chapter.
-78 -
5. MODEL DEVELOPMENT
5.1. Introduction
In this chapter we develop a system of models for the measurement of the profit impact
of sales promotion. First, a framework model which decomposes the retailer's sales into
four factors is developed. Then, the framework is model is filled with with more specific
sub-models of the four factors. Third, the retailer's promotion profit is defined and and
modeled within the framework model. Last, the profit-maximizing deal is defined and the
factors affecting the size of the profit-maximizing deal discount are examined.
5.2. Decomposing Retailer Sales

This section shows how that the retailer's of an individual item can be decomposed in a
number of factors. First, the important issues the model should include are defined and the
general modeling approach is proposed. Then follows a discussion and definition of the
framework model's dependent variable. Last, the model is specified a formal definition of
sales cannibalization is proposed.
5.2.1. General Considerations and Modeling Approach
The task here is to develop a model of retailer sales of individual items. Such a model
should include the following important aspects:
• Grocery shoppers generally buy baskets of items.
• The average basket size is a source of sales for all items in the store.
• Store traffic is a generator for sales of all items in the store, including the focal item.
• Sales promotion may increase the size of the average shopping basket.
• Sales promotion may increase store traffic and thereby cause sales of other items to
increase.
-79 -
• Sales promotion may cause cannibalization among the items in the retailer's
assortment.
A model of retailer sales should include all these aspects. Hypotheses regarding the
importance or existence of each phenomena can then be tested. Such a nlodel could also be
used to track the effectiveness of the retailer's sales promotional activities.
A very interesting approach was suggested by Cooper and Nakanishi (1988) as an
avenue for future research:
"If we have data on the total transaction of each consumer, someone is
going to try to model them. Extrapolating the approach in this book to
that task, we would divide the market basket into categories, model the
total expenditures as we would a category-volume model, and model the
shares among categories as we would a market-share model. Within each
category we would have a nested pair of models for category volume and
brand shares" (Cooper and Nakanishi 1988, p.264).
This is basically the approach that will be used in this study, but it will be extended and
adapted to suit the present problem of modeling retailer sales of individual items. The
store's sales volumes of all items could be modeled using this hierarchical approach. For
example, on the lowest level, consumers choose between different package sizes (purchase
quantities) or varieties or both. On the next higher level, consumers choose between
brands. There may be a level where the consumer chooses between groups of similar
brands. On a higher hierarchical level, the consumer chooses between product categories.
Again, there may be one or nlore hierarchical levels of groups of product categories before
the store choice level is reached.
This modeling approach can be described as the indirect approach because the ultimate
dependent variable is modeled as the product of variables that are modeled as functions of
the independent variables.
5.2.2. Selecting the Quantity Measure
When building market-share models, it must be decided what kind of shares to use. This
is not an easy decision; there are many aspects that need to be considered when defining a
market share measure for practical use. It will be argued that a share of sales valued at
regular (as opposed to actually paid) prices should be an appropriate market share measure
when modeling the retailer's sales.
According to Cooper and Nakanishi (1988), market shares can be measured as shares of
sales volume (e.g., units or weight) or shares of monetary sales (e.g., dollars or kronor). We
shall here use monetary market shares for the following reason. A necessary condition for
the definition of market shares is that the sales volumes of the items can be aggregated.
- 80-
This implies that sales volumes of all items must be expressed in some common unit,
which becomes harder and harder to find the more heterogeneous the items are. Grocery
retailers have very large and heterogeneous assortments. The most relevant unit common to
all items in such assortments is money. Monetary sales volumes can always be aggregated.
In addition, monetary sales have direct managerial relevance.
The monetary market share measure used in the model will however be modified to
overcome a major drawback of monetary market shares in response models. The drawback
of monetary sales is that it is the product of the number of units sold and the price charged.
This makes changes (and even the absence of changes) in monetary sales difficult to
interpret. It also makes sales-response models difficult to interpret, because price would
occur on both sides of the model equation. The model would thus explain price changes
with price changes, which is not very helpful. A solution would be to get the price out of
the sales measure.
Fortunately, actual price, i.e., the price charged, can be removed from the monetary sales
measure. Instead of using actual prices for valuing the unit sales volume, we can use the
regular price. This gives the sales volume valued at regular price. Sales valued at regular
price can be seen as a Laspeyres quantity index, i.e., a quantity index that uses the base
period's prices as weights (see e.g., Chou 1984, p.669). The base period here corresponds
to the case when the promotion is not used and the current period is the case when the item
is promoted.
What happens when the regular prices changes? Regular price change·s could be handled
as follows. Instead of using the current period's regular prices to calculate sales valued at
regular prices, sales could be valued at the regular price of some base period.
The difference between current period valuation and base period valuation would then
be regarded as a gain or a loss depending on whether current prices are higher or lower than
the base period's. These considerations are mainly of interest when estimating the sales
model parameters from a long time-series with large changes in the price level and in
relative prices. In such cases, however it may be better to use a subset of the data set, since
structural changes in prices can change the parameters as well.
In conclusion, the market-share measure chosen here is a monetary market-share
measure. In the preferred measure, sales is valued at the regular prices valid at a specified
point in time.
5~2.3. Decomposing Retailer Item Sales
This section develops a model for analyzing retailer sales of single items by expressing
item sales as the product of four factors. The n10del can be seen as an extension and
adaption of the indirect modeling approach (e.g., Naert and Leeflang 1978). Recall that
- 81 -
indirect modeling of item sales means modeling product class sales and market shares
separately. Item sales is then received as the product of the two models' output. The model
developed here is an extension in the respect that a hierarchy of sales shares is introduced.
It is also an adaption of the general model to the retail setting in the respect that store sales
are modeled as the product of store traffic and average basket value. A model similar to the
one developed here was used by Persson (1992) for decomposing the promotional sales
effect into a purchase quantity effect and a effect on the number of buyers. That model was
later formalized by Persson (1994).
Following the conclusion of the previous section, we define sales of item i in category j,
period t as:
(5.1)
where
= base period's regular price of item i in category j

= unit sales of item i in category j, period t
Sales valued at regular prices can be aggregated to the desired level of detail and
overview. Three relevant levels of aggregation are assumed: the item-level, the category-
level, and the store-level. Then, category sales of category j in period t is the sum of the
sales of all items belonging to that category. That is,
(5.2)
where I j is the number of items in category j. The J categories can then be aggregated to
yield store sales in period t:
(5.3)
We now have market-share measures that can be applied at all levels of aggregation in a
supermarket. Assuming three hierarchical levels (store, category, and item), sales of an
item (valued at regular price) can be expressed as the product of three factors:
(5.4)
- 82-
In period t, the sales of item i in product category j is the product of three factors: (1) the
total store sales valued at regular prices; (2) the product category j's share of store sales;
and (3) the item i's share of category j's sales.
In this example, we assume that the hierarchical levels are store-level, product category,
and item. Obviously, the hierarchy could be extended to any number of levels. For
example, categories could be grouped into departments of related categories, or categories
could be split into sub-categories based on brands or varieties.
Store traffic is another important consideration for the retailer. The model is not
complete on important issues if store traffic is not included in the model. Using the
hierarchical modeling approach, inclusion of store traffic is a fairly straightforward
extension of the model. The model can be written as:
(5.5)
where Bt is the number of receipts produced by the EPOS in period t. This is a direct
measure of the number of shopping baskets that shoppers purchased. The second factor is
the average value of the shopping baskets (valued at regular prices).
The shopping basket concept makes alternative models of item sales feasible. At each
hierarchical level, the number of shopping baskets that contain an item, or at least one item
from a group of iten1s, can be counted. Sales of an item can then be written:
(5.6)
or
(5.7)
where
bijt = number of baskets that contain item i in category j, period t

bjt = number of baskets that contain at least one item from category j, period t
Although these three models look similar to the previous model, there are important
differences. The ratios between basket counts at different levels give percentages that can
be interpreted as conditional probabilities, i.e., the probability the category j's item i is
bought, given that the category j is bought. However, these ratios are different from the
- 83 -
sales ratios, because more than one sub-class may be chosen. This is most obvious at the
category-level where the sum of category-to-total baskets is unity only when each basket
contains exactly one product category. It is also easy to see that the sum can never be less
than unity, because empty shopping baskets are not counted. Thus we have the following
relation between the basket counts:
(5.8)
and
(5.9)
The ratios are useful because they are easy to understand and highlight the importance
of the number of buyers for the sales of an item or a category. Knowing, for example, that
about five percent of all baskets contain coffee would be useful as a heuristic device when
making back-of-the-envelope calculations to forecast sales.
The hierarchical model proposed here is similar to the model proposed by Grover and
Srinivasan (1992) but differs in the hierarchical structure. Grover and Srinivasan define
store share at the category-level as the store's share of total category sales. In their model,
the category sales are a function of the attractiveness of the category in all stores. The
present model defines category share as the share of a store's sales. The store's sales are
regarded as a function of the attractiveness of all the store's categories.
A further difference is that, while the Grover and Srinivasan model is specified at
consumer segment level, the present model regards the store's market as unsegmented. On
the other hand, the present model is more explicit concerning store traffic than the model
by Grover and Srinivasan.
5.2.4. Degree of Cannibalization Defined
Sales cannibalization is an important issue in grocery retailing, and the retailer thus
needs an efficient way to describe and measure this phenomenon. It is not immediately
clear how such a measure should be defined. Cannibalization could be measured in
absolute and relative tern1s, and relative category or store sales. We shall here use a
measure relating absolute changes at the item-level to absolute changes at the store-level,
and absolute changes at the category-level to absolute changes at the store-level.
- 84-
Define lisijt' lis jt , liSt as the changes in item, category, and store sales respectively
caused by a promotion for item i, in category j, period t. The degree of intra-category
cannibalization can then be described by cI>12' defined such that:
cI>12 liSijt =lis jt (5.10)
This measure is zero when there is complete intra-category cannibalization and unity in
the absence of intra-category cannibalization. Values greater than one would be interpreted
as sales increase multipliers rather than in terms of cannibalization. For example, if the
intra-category cannibalization variable is 1.5, then an increase of SEK 100 in sales of the
focal item would mean that category sales increase by SEK 150.
Inter-category cannibalization can be described by defining cI> 23 such that:
(5.11)
The notation of the cannibalization variables anticipates definitions used later in this
report where the item-level is denoted as the first level, the category-level as the second
level, and store-level as the third level. The subscript 23 is thus based on the logic that the
intra-category cannibalization measure relates the second level to the third.
These definitions of cannibalization are useful for describing the impact of a promotion.
However, it may sometimes be useful to describe the rate of cannibalization, i.e., the
cannibalization that would occur for a small change in the promotion variable. We define
the rates of intra- and inter-category cannibalization l/112 and l/123 such that:
dSijt ) dSjt
lfJ12 - - = - - (5.12)
[ dOijt dOijt
and
dSjt ) _ dSt
l/123 - - - - - (5.13)
[ dOijt dO ijt
where 0ijtdenotes the promotional status of item i, category j, period t. Similar to the
cannibalization measures relating sales changes at different levels, the variables '1'12 and
\¥23 can be defined to relate changes in the number of baskets at different levels. Let,
(5.14)
and
- 85 -
(5.15)
The variable \f23 can be interpreted as the share of the increase in category baskets that
represent increased store traffic. When the variable is zero, store traffic is not affected by
the promotion and all incremental category baskets belong to customers who would visit
the store anyway. A value of unity implies that all incremental buyers of the category
represent increased store traffic. Likewise, the variable \f12 is interpreted as the share of the
increase in item baskets representing incremental buyers of the category.
The corresponding variables defined for small changes in the promotion are:
11£
't'12
ab··t
_lJ J
-
ab'J_t (5.16)
[ aO ijt aOijt
and
If/23 - -
[ aO ijt
J
ab jt ==--
aRt (5.17)
aOijt
Note that the these measures cannot be interpreted in terms of cannibalization. Even jf
the number of category buyers is unaffected by a promotion while the number of buyers of
the promoted item increases, the number of buyers of other items need not decrease. This is
because the incremental buyers of the promoted item may still buy other items in the
category.
5.3. Specifying the Response Models

The framework model developed in the previous section will here be filled with
response models. First, a series of assumptions that guides the model specification is made.
Then, the promotional response models are defined for the factors of the framework model.
5.3.1. Model Assumptions and Constraints

We can impose some constraints on the variables and ratios in the model. The
constraints guide the model specification in two ways. First, some of the assumptions guide
the choice of the response models. Second, the assumptions lead to constraints restricting
the valid range of the model parameters. This is useful when specifying the model
parameters without empirical data. If empirical data are available, the second type of
assumptions are to be treated as hypotheses.
By definition of the variables, the following inequalities are always satisfied:
- 86-
(5.18)
(5.19)
If the retailer promotes item i in category j in period t with the sales promotion eijt' we
can assume that the effect on sales is positive or neutral at all levels of aggregation. That is,
the partial derivatives of sales with regard to the promotion are positive or zero:
as.. as. as
---!l!...- ~ O;-J-t ~ o;_t- ~ 0 (AI)
aeijt aeijt aeijt
This assumption is not very restrictive and simply means that promotion for the focal
item does not hurt sales of the focal item, sales of the focal category, or store sales. At least
at the item-level, it is generally expected that the promotion increases sales of the promoted
item. Moore and Olshavsky (1989) state that the exception to this general pattern is when
an unknown item is promoted with a deep price cut. However, sales of other items in the
focal category, as well as sales of other categories, may be hurt by the promotion without
violating this assunlption. This assumption may be violated if the focal item is considerably
cheaper (at regular price) than the items it cannibalizes. However, results from previous
research (e.g., Blattberg and Wisnievski 1989) show that cheaper products rarely
cannibalize more expensive products when promoted.
The number of item baskets, number of category baskets, as well as the total number of
baskets would be expected to rise or remain constant as a result of the promotion. That is,
ab.. abo aB
----.!l!...-~O;-J-t ~O;-_t ~o (A2)
ae~ ae~ ae~
However, we could make a somewhat stronger assumption for the changes in the
number of baskets. We can assume that the increase in item buyers would be at least as
large as the increase in category buyers. In general it is reasonable to believe that the
promotion would not increase the number of category buyers more than the number of item
buyers, since the new category buyers would be attracted to the category by the promotion
on the focal item. Using the same logic, store traffic would not increase more than the
number of category buyers. We have the following inequalities:
ab.. abo aB
----.!l!...-~-J-t ~--t ~o (A3)
aeijt aeijt aeijt
- 87-
This implies that the ratio of item baskets to category baskets, and the ratio of category
baskets to total baskets increase (or remain constant) as a result of the sales promotion. In
other words, the following two inequalities must be satisfied:
(A4)
Note that if (A3) holds, then (A4) will automatically hold, while the opposite is not true.
We can also assume that the promotion would increase the focal item's share of category
sales, and the focal category's share of store sales (valued at regular prices). This means
that
(AS)
It should be noted that this does not imply that sales of the focal item increases more
than sales of the focal category, or that the focal category's sales increase would be larger
than the increase in store sales. In some cases, this may be the result, but in general we
would expect store sales to increase more than category sales, which is likely to happen if
store traffic increases. The relative sizes of the changes, however, should satisfy the
following inequality:
(A6)
Assumption (AS) constrains the cannibalization rates, ¢23 and ¢12' by the following
boundaries:
O-<At <!.L
'1'23- (A7)
Sjt
and
(A8)
The verbal interpretation is that the multipliers are greater or equal to zero and smaller
than the inverse of the sales share. If the multiplier were equal to the inverse of the sales
share, the sales share would not be affected by the promotion. For example, the multiplier
for an item with a 20 percent share of category sales would be lower than five. If the
- 88 -
\
multiplier were greater than five, the promotion would reduce the promoted items share of
category sales.
Assumption (A3), about changes in the number of baskets at different levels, is stronger
than assumption (AS) about the changes in sales sales at different levels. This constrains
the variables If/12 and If/23 to the zero to unity range, i.e.,
(A9)
and
(AIO)
It should be noted that these assumptions are mainly proposed to guide the specification
of the response models. The modeling framework would, however, not break down if the
assumptions turn out to be unrealistic. The two most basic assumptions, (AI) and (A2),
which state that promotions have positive effects on sales and on the number of baskets,
are not very controversial. Assumptions (A4) and (AS) are used in the specification of the
response functions and guide the estimation of the empirical models. Assumption (A3) is
the strongest of the assumptions; it is used to facilitate the parameterization of the
simulation model. In the empirical part, this assumption has no importance and would
rather be seen as a hypothesis than an assumption. The remaining assumptions, (A6), (A7),
(A9), and (AIO) , were all specified for convenience, and can be derived from (A3) and
(AS). Thus, they should also be seen as assumptions guiding the parameterization of the
sin1ulation model and be regarded as hypotheses in the empirical use of the model.
5.3.2. Specification of Response Models
In this section, we specify seven response models for the factors of the framework
model. Three models relate to store traffic, category-to-store baskets, and item-to-category
baskets. Store sales, category-to-store sales, and item-to-category sales are modeled by
three further models. The last model is a model of average basket value, i.e., store sales to
store traffic. Before defining the models, we shall define the independent variables used to
represent sales pron10tion.
The response models specified here consider two types of sales promotion for the focal
item. Most of the attention will be given to the price deal promotion, which is modeled by
the DEAL variable. The DEAL variable is defined as the deal discount expressed as a
fraction of the regular price. That is, DEAL is zero when there is no price deal for the focal
item and 0.15 when there is a 15 percent price deal discount.
- 89-
The second promotion variable in the models is a dummy variable indicator for in-store
special display of the focal item. The variable is named DISPLAY and assumes the value
zero when there is no special display allocated to the focal item. When the focal item is on
special display, then DISPLAY becomes unity. Note that both DEAL and DISPLAY would
have subscripts for the focal item, the category, and the time period. These subscripts were
however omitted.
There are, of course, numerous other variables that could be included in the response
models. For example, newspaper feature advertising, coupons, leaflet advertising, all types
of sales promotion for other items, seasons, and holidays. In order to avoid overloading the
models with unnecessary detail, such variables have not been included in the model.
However, this should not be seen as a limitation because the response models can easily
accommodate these variables as well.
The four dependent variables, item-to-category baskets, item-to-category sales,
category-to-store baskets, and category-to-store sales, are all logically constrained to the
range of zero to unity. A robust model formulation should ensure that the model output
falls within this range. It was noted in chapter three that the attraction type of model
satisfies this criteria. One well-known model of that type is the binomial logit model,
which is the model chosen here.
The binomial logit model used to describe the ratio of item baskets to category baskets
can be written as:
bijt ----..!!1..- (5.20)

bjt 1+ ~
where
(5.21)
The promotion response parameters are non-negative to satisfy assun1ptions (A4) and
(AS). Thus, the three functions are strictly positive and their partial derivatives with respect
to price deal and special display are non-negative. In the same way, we define the binomial
logit models for category-to-store baskets, item-to-category sales, and category-to-store
sales as follows:
b==-
-
jtU 2- , where U 2 == exp(a f
u2 3 ~
+ u2DEAL+uu2DISPLAY ) (5.22)
Bt 1+ u2
- 90-
Sijt= -VI h af
= exp( 3 ~ )
- - , were VI vl + vIDEAL+uvIDISPLAY (5.23)
Sjt 1+ VI
(5.24)
The remaining three models to be defined are not restricted to the range of zero to one.
They do have zero as a lower boundary, but have no upper boundary. These variables can
be seen as scaling factors in the framework model. We shall here use the exponential
model to model store traffic, store sales, and average basket value. The exponential model
has previously been used successfully to model promotional sales-response by Blattberg
and Wisnievski (1989), for exan1ple. Thus, we define the following 1110dels:
(5.25)
(5.26)
The promotion response parameters are non-negative in the store traffic and store sales
models. This means that the functions (5.25) and (5.26) are strictly positive, and their
partial derivatives with respect to the price deal and the special display are non-negative.
The average basket value is modeled using the follwing model:
(5.27)
The assumptions above do not imply that the parameters in equation (5.27) have to be
positive. In other words, sales promotion may decrease the value of the average shopping
basket. This happens if the store traffic increases more than the store sales. It can be shown
that the model parameters are related as follows:
(5.27)
f3w3 = f3v3 - f3u3 (5.28)
(5.29)
This section specified seven promotional-response models for the factors of the
framework model. In four models, the binomial logit model was used, as it represents a
- 91 -
robust model form when the dependent variable is constrained to the range of zero to unity.
The remaining three models were specified using an exponential structure. The sales-
promotion variables entering the models as independent variables were also specified. The
relationships between the cannibalization measures and the model parameters are presented
in the appendix to this chapter.
5.4. Modeling Retailer Promotion Profits

This section defines a model for measuring retailer promotion profits from a sales
promotion on a single item. First, a general and exact model defining promotion profits of
a promotion for an item as the change in the total gross profits of the store is developed.
Second, the model is simplified by introducing some approximations in order to make it
more parsimonious. Third, it is shown that retailer promotion profits can be measured at
the item, category, and store-level, and that the item and category measures are myopic in
the sense that not all the effects are measured. Fourth, cannibalization's impact on
promotion profits is investigated.
5.4.1. Retailer Promotion Profits
This section defines promotion profits as the incremental profits due to the promotion.
For determining the promotion profits, we first need to define the equations for the
retailer's profits (or gross margin) with and without the promotion. The retailer's profit (n)
in period t can be described by the following equation:
J Ij
1f, = I"I,,[qij'(Pij' - Cjj, - djj, + 'fij') - Dij, + J;j,] - 1'; (5.30)

j=1 i=1
where
qijt = unit sales of item i, category j, period t

Pijt = regular price of item i, category j, period t
C ijt = unit cost of item i, category j, period t
d ijt = deal discount (in monetary units) for item i, category j, period t
'fijt = trade deal per unit of item i, category j, period t
Dijt = lump sum deal cost for item i, category j, period t
T;jt = lump sum trade deal for item i, category j, period t
~ = fixed costs in period t

J = number of product categories
- 92-
= number of items in category j
We will examine small changes in the retailer's promotional portfolio; specifically, the
marginal effect of a change in the promotional status of one item, assuming that retailer
keep the promotional status of all other items unchanged. It is assumed that the retailer
faces a number of trade deal offers which mayor may not be made conditional on the
retailer's promotional activity.
The profit of a suggested pron10tion for item m in category n in period t is the difference
in store profit with and without the promotion. The superscripts * and () denote variables
that relate to the baseline (i.e., without the promotion) and the promoted cases respectively.
So, the profit of the promotion for the focal item is:
(5.31)
where
= profit in period t without the promotion on the focal item in period t

= profit in period t with the promotion on the focal iten1 in period t
Inserting the profit expression (5.30) in (5.21) gives:
(5.32)
The retailer's fixed costs are the same whether or not the promotion is run. Also, other
items' promotional status are supposed to be given when calculating the profit from the
suggested promotion on the focal item. All items' regular prices and the regular wholesale
prices are unaffected by the promotion. It is assumed that the focal item would be sold at
regular price if the suggested promotion did not occur.
Dividing the profit from the promotion into the effect on the focal item, other items in
the focal category, and items in other categories, the promotion profit equation can be
written:
- 93 -
f11C~ = [q~nt(P:"'t - C:"'t - d: t + -r~nt)- D~nt + T:nt ]- [q:"'t(P:nt - c:nt + -r:'t)+ T;nt]
change in gross profit from the focal item m in category n
J J J. (5.33)
+![(qi~t -qi:t)(P:t -C: -di:t+ <nt)] + Lt[(q~t -qj;t)(p;t -C;, -dj;t +-r~t)]
t
i::l=m j::l=n i=1
_------..--------"
'.......
oth~r items i in the focal category n
\
change in gross profits from change in gross profit; from other categories
Define the change in sales quantity of item i as:
(5.34)
Inserting (5.34) in (5.33) gives the profit of the sales promotion for the focal item as:
t = ~qmnt (*
l11C 9 *),+ ~qmnt (9
Pmnt - Cmnt * (9
9),+ ~mnt
1'mnt - dmnt * - dmnt
1'mnt - 1'mnt 9) I
~ J ~
+(T:nt -T;nt -D~nt)+ LI1Qint(P;nt
, (4) I !::I=m
-C: -d t +
t i: 1':nt) + LLI1Qijt(p;t -C;t -d;t +1'~t)
!..·::I=_n_i=_1
.. -------'
(5) (6)
(5.35)
The promotion profit can thus be seen as the sum of six terms, which are interpreted as
follows: (1) the increased sales volume of the focal item makes a positive contribution to
the store's gross profits.
(2) The gross profit of the incremental sales is corrected for the per unit trade deal and
the per unit price discount. Whether this term is positive or negative depends on the size of
the trade deal and the retailer's pass-through of the trade promotion to the consumers.
(3) the gross profit from the sales of the focal item that would have occurred had the
promotion not taken place must be adjusted for the per unit trade deal and the retailer's
price deal. Note that it is the difference between the per unit trade deal the retailer would
receive with the promotion and the trade deal the retailer would receive without the
promotion that is of importance for the promotion profits. The sign of this term is again
dependent on whether the trade deal would be received independently of the promotion or
not, the size of the trade deal, and the retailer's pass-through.
(4) the net effect of lump-sum trade deals and fixed costs of the promotion must be
considered. (5) the sales volumes of other items in the focal category may be affected by
the promotion. (6) the sales of items in other categories may also be affected by the
promotion. The sign of the last two terms can be positive or negative. Term five would in
general be expected to contribute negatively to the promotion profits.
The profitability of a sales promotion thus heavily depends on its effect on sales of the
promoted item as well as its effect on sales of all other items in the store.
- 94-
We rewrite the promotion-profit equation (5.35) in terms of sales valued at regular
prices. Define sales valued at regular price as:
(5.36)
Inserting (5.36) in the equation for the profit of a sales promotion (5.35), we receive the
following expression:
t1n()
t
= t1smnt (1- c: J+ t1s
nt
*
P mnt
mnt
('f~nt* -
P mnt
d: nt
*
P mnt
J+ * S
mnt
('f~nt
*
P mnt
- 'f :nt - d: nt
P mnt
* *
P mnt
J
() *
+(Tmnt - Tmnt - Dmnt )+
()
Lt1s
In
int
( ClOt~
1--*---*-+-*- +
d.*lOt ~
r mt J J Ij
LLt1S~jt 1--* - - * +-*
( ~
CIJt d~IJt 'f~.lJt )
i'#:m Pint Pint Pint j'#:n i=1 Pijt Pijt Pijt
(5.37)
It is useful to define the effective margin for all items except the focal item as:
(5.38)
For the focal item, it is more useful to define the regular margin as:
*
m
mnt
=1- cmnt
* (5.39)
Pmnt
The promotion profits can now be written:
() 'f ()mnt d()mnt J * ('f (mnt

) 'f*mnt d()mnt J
I1n t = I1Smntmmnt +l1smnt ( -*---*- +smnt -*---*---*-
P mnt P mnt P mnt P mnt P mnt
(5.40)
In J Ij
+( T:nt - T;nt - D:nt ) + L t1Sintmint + L L t1Sijt m ijt

i'#:m j'#:n i=1
This is the exact version of the retailer promotion-profit model proposed and used in
this study. The next section will propose a simplified version of the retailer's promotion-
profits model.
5.4.2. A Simplified Promotion-Profit Model
This section develops a simplified or approximate version of the promotion profit

model. The reason for developing an approximate version of the model is that it facilitates
analysis and notation. The model is modified as follows:
- 95-
First, the lump-sum trade deal and the fixed promotional costs are omitted from the
model. This makes the model easier to handle and the loss of generality is minimal. The
lump-sum trade deal and fixed promotional costs can easily be included again in the model.
Second, average effective margins are used for other items in the category and for other
categories than the focal category. This makes the notation of the model much simpler and
facilitates the analysis. However, it introduces the assumption that the margins of other
products are hon10geneous or that the promotional effects are fairly evenly distributed over
the other items and categories. This should be a useful approximation because the retailer's
margin for items within a category tends to be fairly homogeneous. Across categories,
margins n1ay vary somewhat more, but the promotional response may be fairly similar.
However, if the violations to this assumption would be too large, we could split the sales of
other items in the focal category and the sales of other categories into groups of items and
categories with similar margins and sales response.
Third, the manufacturers' promotion requirements are formally modeled. This is done by
defining the size of the trade deal offered and the minimum deal discount the retailer has to
use in order to be entitled to the trade deal. Note that the deal discount and the trade deal
are both expressed in percent of the regular retail price. This facilitates comparisons
between the trade money received by the retailer and what is passed through to the
consumers as price deals. It is important for the interpretation of the results to remember
this definition, because the common way to express the size of the trade deal would be in
percent of the wholesale price. The advantage of a clear interpretation of pass-through was
judged to be much greater than the inconvenience of the somewhat unusual way to express
the size of the trade deal. Lastly, we simplify the notation by omitting the time subscript, t,
from the model.
We define the following variables:
= sales of focal item valued at regular price

= sales of focal category valued at regular prices
= store sales valued at regular prices
= gross margin of item (in percent of regular retail price)
= average effective gross margin of other items in the focal category
~
= average effective gross margin of products in other categories
DEAL = deal discount on focal item (in percent of regular retail price)
DISPLAY = display indicator for focal item

= trade deal received by the retailer (in percent of regular retail price)
TROFFER =trade deal offered on focal item (in percent of regular retail price)
DEALREQ = minimun1 deal required to obtain the trade deal
- 96-
DISPREQ =indicator of display requirement to obtain the trade deal
The manufacturer's deal requirement is modeled using the expression:
t = {TROFFER if (DEAL ~ DEALREQ) and (DISPLAY ~ DISPREQ)

(5.41)
o otherwise
Although not included in this model, the manufacturer could demand that the retailer
includes the item in the feature advertisement. The model could easily be extended to
implement a feature advertising requirement.
Denote the promotion induced changes (i.e., the sales that would occur with the
promotion less the sales that would occur without the promotion) in sales of the focal item,
the focal category, and store-level sales, by the three differences:
(5.42)
The store-level promotion profit, or the promotion profit for short, can now be written:
8 8
117X 3 = &tm1 +&1(t -DEAL)+s;(t -t* -DEAL)+m2(&2 -&1)+m3 (11s 3 -I1S2)
(5.43)
This is the simplified and final version of the promotion-profit model proposed in this
study. This is the model that will be used throughout the analyses in this and following
chapters. Equation (5.43) is an approxin1ation of equation (5.40). The most important
modification is that the average effective margins of other items in the focal category and
the average effective margin of other categories are used instead of the effective margin of
each item.
Note that both the trade deal and the retail deal discount are expressed in percent of the
regular retail price and the net effect of fixed costs and lump-sum trade deals are assumed
to be zero.
It should also be noted that the definition of equation (5.43) is neither dependent on the
specific response models used, nor whether a direct or indirect modeling strategy is used.
The definition of retailer's promotion profits would remain the same, but the numerical
results may be different should the response models be defined qifferently.
5.4.3. Three Promotion Profit Measurement Levels
This section shows that a retailer may evaluate promotional (and other) profits at three
hierarchical levels. Promotion profits may,be evaluated at the item-level, the category-
level, or the store-level. The model developed in the preceding section n1easures the
- 97-
promotion profits at the store-level, the level defined as the correct level of promotion
profit measurement. Item and category-level measures are incomplete and can be viewed as
myopic measures of the retailer's promotion profits.
We now define promotion profits at the item, category, and the store-level. The change
in the retailer's gross profits from the item is the item-level profit from the promotion.
Thus, the item-level promotion profit is given by:
() *
~TCI = TC I -TC I = &l~ +&1 (1() -DEAL) +Sl*(1() -1 *-DEAL) (5.44)
At the item-level, the promotion profit is the sum of three effects. The first effect is the
incremental gross profit from the incremental sales valued at the regular price. This is
adjusted by the second term, the difference between the trade deal and the retail promotion.
The third term is the effect of the change in effective margin on the sales that would have
occurred even if the promotion had not taken place. This is in general negative but may
also be positive if the retailer can absorb some of the trade deal. The third term will be
negative if the retailer would receive the trade deal irrespective of the pass-through.
As already noted, the retailer is more interested in store performance than item
performance. The promotion profits must therefore be measured at the category and store-
level as well. The category-level promotion profit is defined as:
(5.45)
The category-level profit of the promotion is the sum of the item-level promotion profit
and the change in gross profits from other items in the focal category.
The store-level promotion profit is defined by the difference:
(5.46)
Like the category-level promotion profits, the store-level promotion profits is the sum of
the category-level promotion profits and the promotion's impact on gross profits from other
categories than the focal category. This means that the total effect of the promotion, i.e., the
store-level promotion profits, can be written as the sum of the change in profits from the
focal item, the change in the profit from other items in the focal category, and the change in
profits from other categories than the focal category. That is:
other items in focal category other categories
focal item ~ ~
~
~TCI + (dTC 2 - dTC I ) + (dTC 3 - ~TC2) (5.47)

"----y---J
~tem level promotional profit
category level p;omotional profit
store level pro~otional profit
- 98 -
This equation clearly shows that the promotion profit can be decomposed into three
parts: (1) the effect on the gross profit from the focal item, (2) the effect on the gross
profits from other items in the focal category, and (3) the effect on the gross profits from
other categories. Further, this suggests that three measures of promotion profits can be
defined, depending on the number of effects that are considered.
The item-level promotion profit considers only the effects on the focal item. The
category-level promotion profit includes the effect on other items in the focal category. The
store-level promotion profit includes the effect on all other categories as well. Obviously,
the three measurement levels will generally provide different estimates of the profit of a
promotion.
As already noted, the store-level promotion-profit is the relevant measure for the
retailer. The item- and category-level measures are myopic, because they do not include all
the effects of the promotion.
5.4.4. The Profit Impact of Cannibalization
The promotion profit model shows the importance of sales effects on other items in the
focal category as well as on other categories in the store. In the development of the sales
model, formalized measures of the intra-category and inter-category cannibalization were
defined. We shall here use these measures to examine the profit impact of cannibalization
by re-writing the promotion-profit model to include the cannibalization measures. Define
<I> 12 and <I> 23 such that:
(5.48)
and
(5.49)
Insert <1>12 and <1>23 into the store promotion profits equation and rearrange terms:
The equation for promotion profits would be more general if the incremental sales were
expressed as the increase relative to the baseline sales. Therefore, define the relative
incremental sales:
(5.51)
- 99-
The relative incremental sales (5.51) are inserted into the promotion-profit equation
(5.50), which yields:
An 3 == S;{((8 _to :DEAL)+(jl[(~ +(8 ~DEAL)+~+~]}

term 1 term 2 term 3 term 4
(5.52)
This equation can be interpreted by examining its parts in detail. First, it can be noted
that the sales volume valued at regular price if the deal would not be launched, enters s;,
the equation only as a scaling factor. This means that it does not affect whether or not the
promotion is profitable, but determines the size of the profit or loss. All other factors being
equal, the larger the regular sales volume, the more pronounced the impact on the profit.
The expression within the brackets consists of four terms which shall here be analyze
one at a time beginning from the left, after distributing the relative incremental sales to
each of the three terms within the brackets. The first term, (t 8 - t* - DEAL), is the change
in gross margin on sales of the focal item that would have occurred had the promotion not
taken place. This term is in general negative and can be regarded as an opportunity loss
caused by the promotion. The second term, CJ'1 (m1 + t 8 - DEAL), is the effect on profit
from the item that stems from increased sales volume. In this term, (m 1 + t 8 - DEAL) is the
effective margin on the incremental sales while CJ't is the size of the sales increase. The
effective margin can be interpreted as the regular margin adjusted for the deal and the trade
deal received from the manufacturer. Given that the effective margin is positive, the size of
the sales increase is obviously very important for the profitability of the promotion.
The first two terms together give the promotion-induced change in the retailer's gross
margin from sales of the focal item. The third term expresses the effect on the retailer's
gross profits from other items in the focal category, while the last term expresses the effect
on gross profits from other categories.
Assuming that the average gross margin for the other items in the focal category is
positive, the third term, 0'1m2(<1>12 -1), is zero if <1>12 equals one. This would n1ean that
sales of other products in the category are unaffected from the sales promotion on the focal
item. The full increase in sales of the focal item translates into increased sales of the
category, which can be interpreted as the absence of intra-category cannibalization. In the
absence of intra-category cannibalization, the promotion has no impact on profits from
other items in the category. If the focal item does cannibalize on sales of other items in the
category, the value of the third term is negative. With full intra-category cannibalization,
i.e., when <1>12 equals zero, the sales promotion reduces the gross profits from other items
in the category by CJ'1~. Whether or not the promotion is profitable is dependent on the
- 100-
relative size of the effective gross margin on the focal item compared to the gross margin
on other items in the focal category. The third term may sometimes become positive,
implying that gross profits from other items in the category increase due to the promotion.
This would be rare if the category is narrowly defined with only substitute items, but more
common if the category contains complements to the focal item. The third term is positive
when cI>12 is greater than one.
The effect on the retailer's gross profit from other categories than the focal category is
given by the fourth term, (Jlm3cI>12(cI>23 -1). This term is more likely than the third term to
be positive because there is more complementarity in the purchase of products at this level
than within a category. The complementarity in purchase may come from complementarity
in use or simply from the fact that it is convenient to buy a basket of groceries rather than
making one store visit for each product. The fourth term may also become negative. This
occurs when the promoted category cannibalizes on other categories' sales, indicated by a
cI> 23 between zero and one. When cI> 23 is exactly one, the incremental store sales are equal
to the incremental sales of the focal product category. The effect on the gross profits from
other categories than the focal one is zero. The fourth term can be zero for another reason,
too--when there is complete intra-category cannibalization in the focal category, i.e., cI>12
equals zero. This is because complete intra-category cannibalization means that all
incremental sales of the focal item come from other items in the category. The effect on
sales in other categories is consequently zero.
5.5. Maximizing Pronlotion Profits

The previous section showed how the effect of sales promotion on a retailer's profits can
be computed using changes in sales and margins caused by a promotion. Ex post, that
model can be used to analyze the components of promotion profits. The model can also be
used to analyze the profit impact of different values of the variables in the model. This
section defines the optimal deal and how that can be found, starting with the optimal size
of a deal discount. Then, the value of a special display is discussed as well as how to
determine whether or not to allocate a special display to an item. The concept of myopic
deal optimization is defined together with a discussion of its costs. Last, we discuss the
impact of cannibalization, margins, and deal elasticity for the size of the optimal deal as
well as for the myopic deal. It should be noted that the results in this section are general
and not specific to any particular response model, except for the last discussion where a
particular response model is used.
- 101 -
5.5.1. Profit Maximization and Optimal Deal Discount
This section describes how the profit-maximizing retailer would set the size of the deal
discount for the focal item. It is assumed that the retailer wants to maximize store-level
gross profits. Note that the retailer's gross profits can be written:
(5.53)
This can be rewritten as:
(5.54)
Define the rate of intra-category cannibalization, ¢Jl2' and the rate of inter-category
l
cannibalization, ¢J23' such that:
aSl aS2
(5.55)
¢J12 aDEAL aDEAL
and
l/> dS2 aS3

(5.56)
23 aDEAL aDEAL
Assuming that the trade deal received by the retailer is not dependent on the size of the
deal, differentiation with respect to DEAL yields:
------'---
an 3 = -s + aSl ( m +t- DE'r1AL) + (aS 2 _.asl ) m + (aS 3 aS2 )
aDEAL 1 aDEAL 1 aDEAL aDEAL 2 aDEAL aDEAL m3
(5.57)
Inserting ¢J12 and ¢J23 and sin1plifying gives the following equation:
(5.58)
If the gross profit is a continuous function of the deal discount, then the derivative of the
gross profit function with respect to DEAL is zero at the profit maximizing deal discount.
This means that the rate of change in store-level gross profits is zero when the profit-
maximizing deal is used. Thus, the profit maximizing deal, DEALill, satisfies the equation:
(5.59)
- 102-
When DEAL '* 0 we can define the deal elasticity as:
DEAL as.
TJ DEAL =-s-.-. aDEAL (5.60)
Inserting the deal elasticity and rearranging the equation gives:
OJ In, + t + (If>.2 -1)m2 + If>12(lf>23 -1)m3

DEAL = 1 (5.61)
1+--
TJDEAL
This expression looks fairly simple and may be used to evaluate whether a profit
maximizing deal discount has been found. Using the sales response models defined earlier
in this chapter, it seems unlikely that the profit-maximizing deal discount can be found
analytically by solving for DEAL OJ. This is because If>12 and If>23' as well as the rate of
change of sales due to the price deal, are all functions of the price deal in that model. Using
that model, numerical methods are needed in order to find the profit-maximizing deal
discount.
However, we can evaluate the derivative for a given size of deal discount in order to
determine whether this is a profit maximizing deal discount. If the derivative does not
indicate a profit maximum, its sign tells whether a larger or smaller deal would be better. In
order to check whether a price deal could be profitable we can evaluate the derivative when
the deal discount is set to zero. A non-positive value indicates that nothing can be gained
by a price deal (given that the trade promotion is not conditional on a certain size of the
deal discount). A negative derivative indicates that a negative deal discount, i.e., a higher
price, would be more profitable than the current price. However, since we cannot have
negative deal discounts and the response function is specified for price deals (not for
regular price changes), it cannot be concluded that the regular price should be increased.
If the trade promotion is made conditional on a certain deal discount, we need to
evaluate the profit when the deal discount is zero and when it is at the minimum level at
which the retailer receives the trade deal. If the gross profit is higher when the deal is zero
than when set equal to the deal requirement, it is sufficient to evaluate the derivative when
deal is zero. If the profit would be higher when the retailer receives the trade deal, the
derivative at the smallest deal discount that still gives the retailer the trade deal should be
evaluated. If this is non-positive, this is the optimal deal discount. If it is positive, a larger
deal discount would be more profitable.
A perhaps less elegant but simpler and fail safe method is to evaluate the promotion
profit for all possible deal discounts. This is the method that will be used in the illustrations
of the model in Chapters 6, 8,9, and 10.
- 103-
5.5.2. Display Profits
In principle, the profit of a display is defined in the same way as the profit from a price
deal, i.e., as its impact on the retailer's store-level profits. However, there are two important
differences: First, display does not have any impact on the margin of the focal item.
Second, display is not a continuous variable. This means that derivatives are not useful for
determining whether the display should be used or not.
As already noted, the value of a display allocated to an item is equal to its impact on
retailer profits, measured at the store-level. Its value is dependent on whether or not the
size of the deal discount has already been fixed. If the deal discount is pre-determined, for
example by an agreement with the manufacturer, the display profit is the difference
between the retailer's gross profits with the display allocated to the item and the gross
profits without the display. Although this approach gives a correct estimate of the value of
the display given that the deal discount is fixed, it is not useful when the deal discount can
be changed. In fact, setting the deal discount first and subsequently determining the value
of display is likely to lead to sub-optimal performance. The better alternative when the
retailer can set the size of the deal is to determine the optimal deal discount with and
without a display. The value of the display is then the difference between the promotion
profit of the optimal deal discount with the display and the promotion profit of the optimal
deal discount without the display allocated to the focal item.
The manufacturer may offer the retailer a lump-sum display allowance that is made
contingent on the retailer allocating a special display to the item. Such a lump-sum
allowance can easily be handled by the model. The effect is to increase the value of the
display by the lump sum. Note that a lump-sum display allowance that is not contingent on
the actual allocation of a display for the focal item has no effect on the value of the display,
because the retailer receives the money irrespective of the display. The lump sum would
then be added to the gross profits with the display as well as without the display. Thus, the
difference, i.e., the value of the display, is unaffected by the allowance.
5.5.3. Myopic Optimization
It has been argued that the retailer should measure promotion profits as the change in
store-wide gross margins caused by the promotion. The optimal .promotion is thus the
promotion that has the greatest positive effect on the store's profits. Using the more myopic
approach of measuring promotion profits as the change in gross margin of the promoted
item and designing promotions that maximize these profits is likely to be sub-optimal. We
shall here investigate what the myopic approach means for the size of the deal discount and
what it costs in terms of forgone profits.
- 104-
If the retailer uses a myopic approach when determining the size of the deal discount,
then the deal discount will be determined by maximizing the item-level gross profits:
(5.62)
This implies differentiation with respect to the size of the deal discount:
ale} as} ( )
- - - = -s} + m} +t-DEAL (5.63)
aDEAL aDEAL
The myopic deal discount, DEALJ1, is obtained when the rate of change in item-level
profit is zero.
O=-s! + as! (~+t-DEAI!') (5.64)

aDEAL
Inserting the price deal elasticity defined above (5.60) and rearranging the terms means
that the following equation holds at the myopic optimal deal discount:
DEALJ! = ~ +t (5.65)
1
1+--
17DEAL
Note that this equation in general cannot be used to determine the size of the myopic
deal discount unless the elasticity is constant. Myopic deal optimization and store-level
deal optimization occasionally lead to the same deal discount. However, in general, myopic
deal optimization cannot be expected to set the same deal discount as store-level
optimization.
Whenever the deal discount determined at the item-level deviates from that determined
at the store-level, it is likely that the store-level gross profits would be lower if the myopic
deal discount were used. This means that there is an opportunity loss associated with
myopic optimization, i.e., there is a cost to myopic deal optimization.
The cost of myopia can be estimated as the difference between the profits from the
optimal deal and the profits from the myopically determined deal. Zenor (1994) used a
similar approach to measure the manufacturer's profit from category managen1ent as
compared to brand management. Symbolically, the cost of myopia can be written:
(5.66)
- 105 -
Trying to find a general analytical solution to this cost leads to a complex equation
which is difficult to interpret. The cost of myopia should be evaluated using the relevant
sales-response models and the promotion profits equations.
Myopic optimization was defined as the deal set by maximizing item-level gross profits.
The cost of such myopic optimization was defined as the difference between store-level
promotion profits from the optimal deal and the store-level profits from the myopically
determined deal. Maximization of category-level profits is also a myopic approach and that
the cost of category-level deal optimization is defined in the same way as the cost of item-
level deal optimization.
5.5.4. Determinants of Optimal Deal Discount
In this section, conlparative statics are used to examine how the optimal deal discount
changes when the degree of cannibalization, the margins, and the deal elasticity change.
We also examine the effect that changes in these factors have on the myopic deal discount.
To simplify the analysis, we assume that lP12' lP23' and 17 DEAL are exogenously determined
and independent of one another. This means that the results presented here are not directly
applicable when the factors show complex interdependencies, as in the sales model
developed earlier in this chapter.
Recall that the optimal deal discount satisfies the equation:
OJ m1 + t + (lfJ12 -1)m2 + lfJ12(lfJ23 -1)m3

DEAL = 1 (5.67)
1+--
17 DEAL
As the factors on the right side of the equation are assumed to be exogenously
determined, i.e., not dependent on the deal discount, the equation determines the optimal
deal discount. Thus, we can investigate how the optimal deal discount changes when the
factors on the right side change. Beginning with the intra-category cannibalization, the
following partial derivative gives the rate of change in optimal deal discount for small
changes in lP12:
JDEALOJ _ m2 + m3(lfJ23 -1)

(5.68)
JlfJ12 1+_1_
17 DEAL
The derivative shows that as long as there is no inter-category cannibalization, i.e., as

long as lP23 is not lower than one, increasing lP12 leads to larger optimal deal discounts. In
other words, the optimal deal discount will be larger the less cannibalization within the
category when sales of other categories are positively affected by the deal. When lfJ23 is
- 106-
lower than one and the average margin of other categories is higher than the average
margin in the focal category (excluding the focal item), increasing l/J12 may lead to a lower
optimal deal discount.
Increasing the value l/J23 means to reduce the degree of inter-category cannibalization (or
to increase the store sales increase multiplier if its value is greater than unity). The effect
on the optimal deal discount of increasing l/J23 is always positive (or zero); that is, the
optimal deal discount will be larger. The rate of change is given by:
In:3l/J12 (5.69)
1
1+--
17DEAL
Larger trade deals and higher margins on the focal item obviously lead to larger optimal
deal discounts. ~rhe effect on the size of the optimal deal discount of increasing margins of
other items in the category is in general negative as can be seen from the following
derivative:
l/J12 -1
(5.70)
1
1+--
17DEAL
The effect of higher margins on other categories than the focal category on the size of
the optimal deal discount depends on l/J23. When l/J23 is lower than one, i.e., when the focal
category cannibalizes on other categories' sales, increased margins on other categories
mean lower optimal deal discount. On the other hand, when sales of other categories
increase due to the promotion, the optimal deal discount will be larger. The rate of change
of the optimal deal discount for changes in the average margin of other categories is given
by:
l/J12(l/J23 -1) (5.71)

1
1+--
17 DEAL
The effect of increasing the price deal elasticity is an increase in the size of the optimal
deal discount.
5.5.5. Determinants of the Deviation from Optinlal Deal Discount
Next, we examine the effect of cannibalization, margins, and deal elasticity on the
deviation of the myopic deal discount from the optimal deal discount. We still assume that
- 107-
l/J12' l/J23' and 1J DEAL are exogenously determined and independent of each other. The
myopically determined deal satisfies the equation:
DEALJ.L = ~ +t (5.72)
1
1+--
1J DEAL
The difference between the optimal deal discount and the myopically determined deal
discount is:
DEAr' _ DEAl!' = (l/112 -1)m2+ l/1¥(l/123 -1)m3 (5.73)

1+--
1J DEAL
This difference will be zero when l/J12 and l/J23 are both equal to one, indicating that the
promotion has no effect on items or categories in the store other than the focal item. The
difference may be zero for other reasons, too. If the average margin of other items in the
focal category and the average margin of other categories relate as
(5.74)
the difference will again be zero. In these cases, there will be no loss incurred by using
the myopic approach to find the optimal deal discount. However, this would of course only
be a coincidence. Therefore, it is interesting to examine the sign of the difference in order
to determine whether the myopically detern1ined deal discount is too large (negative
difference) or too small (positive difference). We will assume positive average margins for
items in the focal category and in other categories.
When there is some degree of intra-category cannibalization as well as inter-category
cannibalization, i.e., both l/J12 and l/J23 are smaller than one, the difference is always
negative, indicating that the myopic deal is too large. The situation is more complex when
there is some intra-category cannibalization, but positive effects on sales in other categories
than the focal category. Large store sales change multiplier values may offset the effect of
the intra-category cannibalization, making the myopic deal too small compared to the
optimal deal discount. The margins are here of considerable importance for how large l/J23
has to be before the myopically determined deal discount becomes too small. The higher
the average margin of other categories, the faster the myopic deal turns out to be too small.
Another factor affecting the difference between optimal and myopic deal discount is the
deal elasticity. The difference between the optimal deal discount and the myopic deal
- 108 -
discount approaches zero as the deal elasticity approaches zero. High elasticities, on the
other hand, tend to inflate the difference, making the sub-optimization more severe.
5.6. Summary
This chapter presented a framework model for decomposing retailer iten1 sales as the
product of four factors. Item sales are modeled as the product of store traffic, average
basket value, the category's share of store sales, and the item's share of category sales. That
is,
Sl ~
B X X X (5.75)
~
sales of focal item store traffic B
~
S3
~
S2
~
average basket value focal category's share of store sales focal item's share of category sales
where B is store traffic measured as the number of shopping baskets checked out, and
Sl' S2' and S3 are sales of the focal item, sales of the focal category, and store sales. It was
noted that the three-level framework model can be extended to any number of levels. The
framework model was filled with response models for each of the four factors, i.e.,
response models for (1) the store traffic, (2) the average basket value, (3) the focal
category's share of store sales, and (4) the focal item's share of category sales.
Two types of cannibalization were formally defined. It was shown that there may be
intra-category cannibalization, ep 12' and inter-category cannibalization, ep 23' in the three-
level framework. These were defined such that:
(5.76)
and
(5.77)
where Llsp ils2 , Lls3 are the changes in item-, category-, and store-sales respectively
caused by a promotion on the focal item.
The retailer's promotion profit from a promotion of a single item were defined as the
change in store-level gross profit caused by the promotion of that item. The promotion
profit of a promotion of the focal item is:
- 109-
f11C 3 = ~t~ +f1st (t 8 -DEAL)+s;(t8 -t* -DEAL)+~(f1sz-f1st)+~(f1s3-f1sz )
item level pro~otional profit
promotional profit or stor~ level promotional profit
(5.78)
It was shown that promotion profits could falsely be measured at the item or category-
level, and that such measurement would lead to different estimates of the promotion
profits.
The deal discount that maximizes the store-level promotion profit was defined as the
optimal deal discount and it was shown how the optimal deal discount can be recognized
and found. Myopic deal optimization was defined as optimization of the item-level or the
category-level promotion profits. The cost of myopic optimization was defined as the
difference between maximum store-level promotion profits and the store-level promotion
profits resulting from the myopic optimal deal.
We examined how cannibalization affects promotion profits measured at different-levels
and how cannibalization affects the optimal deal discount as well as its impact on the
myopic deal.
- 110-
Appendix
S.A.I. Relationship between Cannibalization and Model Parameters

This appendix shows how the measures of cannibalization and the parameters of the
response models are related. It shows (1) how to calculate the rate of cannibalization when
the response model parameters are known, and (2) how the model parameters are related
when the cannibalization rates are known. The first is useful when the model has been
estimated on empirical data, while the second is more appropriate when specifying the
model parameters for a simulation because the degree of cannibalization should be more
intuitive measures than the response parameters.
(5.79)
(5.80)
(5.81)
(5.82)
(5.83)
(5.84)
(5.85)
- 111 -
(5.86)
(5.87)
(5.88)
(5.89)
- 112 -
6. SIMULATING PROMOTION PROFITS
6.1. Introduction
In this chapter, the model developed in Chapter 5 is used to measure the profit impact of
sales promotion for hypothetical model parameters. The chapter starts with a description of
the sales model parameters. Then, the promotional sales-responses of the item, the
category, and the store are described. The profit impact and the profit-maximizing deals are
described next. It is shown that measurement of the promotion profit at the item-level, the
category-level, and the store-level gives different estimates of the promotion profit.
Thereafter follows an investigation of the promotion profit's dependence on the sales-
response model paranleters. The sensitivity to changes in margins, size of trade deal, and
deal requirement are examined next. The chapter ends with a summariy of the conclusions
of the simulation study and a discussion of the implications for retailers and manufacturers.
Note that the simulation here is deterministic and what Kotler (1991) denotes a
marketing plan simulator and Cooper and Nakanishi (1988) refer to as "[to] simulate the
consequences of [plans]". This should not be confused with stochastic simulations like the
Monte Carlo method.
6.2. Definition of the Promotional Response Models

This section specifies the model parameters used in the simulation. Further, it shows the
sales-response and the basket-response to various sizes of the price deal discount.
6.2.1. Model Structure and Parameters
To illustrate and investigate how the deal discount and the special display for an item
affect retailer gross profits at different levels, a set of response models was specified. It is
assumed that the retailer decides on the promotional activities for one item (the focal item)
in a product category (the focal category). The specific promotional instruments the retailer
has to choose are the size of the price deal discount and whether or not to allocate special
display space to the item. The manufacturer offers the retailer a trade deal in the form of a
price discount on the units sold by the retailer during the promotion. Gross margins are
- 113 -
treated as homogeneous across items in the focal category (except for the focal item), and
as homogeneous across categories for other categories than the focal category. The
promotional status of other categories is assumed to be unaffected by the promotion of the
focal item.
The response models and promotional-profit model developed in Chapter 5 are used.
After consultations with industry experts, the model parameters were set to reasonable
values. The response model constants were selected to give the following baseline figures:
• When the focal item is sold at regular price and not on special display, its share
of the category sales is 30 percent and its share of the number of category
baskets is 25 percent.
• The category's share of store sales is then 1 percent while its share of the total
number of baskets is 5 percent.
• Store traffic is 3,000 baskets per day, averaging SEK 100 per basket.
Table 6.1 presents all the response model parameters of the base case model.
Sales-response Models
Item-level constant a v1 = -0.847
Category-level constant a v2 = -4.595
Store-level constant a v3 = 12.61
Item-level deal response f3vl = 5
Category-level deal response f3v2 = 1
Store-level deal response f3v3 = 0.002
Item-level display response DV1 = 0.5
Category-level display response DV2 = 0.1
Store-level display response DV3 = 0.002
Basket Response Models

Item-level constant a u1 = -1.099
Category-level constant a u2 = -2.944
Store-level constant a u3 = 8.006
Item-level deal response f3Ul = 5
Category-level deal response f3u2 = 1.0
Store-level deal response f3u3 = 0.001
Item-level display response DU1 = 0.5
Category-level display response DU2 =0.1
Store-level display response DU3 = 0.001
Basket Value Model

Constant a w3 = 4.605
Deal response f3w3 = 0.001
Display response DW3 = 0.001 =
Table 6.1. Parameters of the response models (Scenario 1: base case).
- 114-
The response model structure and parameters have now been described. In the next
section, we use the model and its parameters to calculate how sales respond to sales
promotion.
6.2.2. Sales Response

This section investigates how the sales-responses at the item-, category-, and store-
levels depends on the size of the price deal discount and whether or not a special display is
allocated to the focal item. More specifically, we look at the sales of the focal item, other
items in the focal category, category sales, sales of other categories, and store sales. It
should be pointed out again that, at all levels, sales refer to sales valued at regular prices,
which means that sales can be interpreted as a type of quantity index (see section 5.2.2).
The deal discount is varied between 0 and 30 percent in two-percent steps. For each
step, the item sales, category sales, and store sales are estimated for the price deal only and
for the price deal combined with a special display. The sales-responses are reported in
Table 6.2.
Deal Item Sales Category Store Sales

Discount No Display Display No Display Display No Display Display
00/0 900 1374 3000 3319 300000 300601
20/0 984 1484 3060 3385 300012 300613
40/0 1073 1599 3121 3453 300024 300625
60/0 1167 1719 3184 3522 300036 300637
8% 1267 1843 3248 3592 300048 300649
10 % 1372 1972 3313 3664 300060 300661
12% 1482 2104 3379 3737 300072 300673
14% 1597 2239 3447 3812 300084 300685
16% 1716 2377 3516 3888 300096 300697
180/0 1840 2517 3586 3966 300108 300709
20% 1968 2660 3658 4045 300120 300721
220/0 2100 2805 3731 4126- 300132 300733
240/0 2235 2951 3805 4208 300144 300745
260/0 2373 3098 3881 4292 300156 300757
28% 2513 3245 3959 4378 300168 300769
30% 2655 3394 4038 4465 300180 300781
Table 6.2. Sales offocal item, focal category, and store sales by deal discount with and
without special display.
The absolute sales levels presented in Table 6.2 are important as they provide the
starting point for the further analysis of promotional effects. Sales of other items in the
focal category and sales of other categories are not reported explicitly in Table 6.1, but can
easily be calculated; sales of other items in the category are the difference between category
sales and item sales. Sales of other categories are the difference between store sales and
categry sales. For example, when there is no deal or special display for the focal item, then
- 115 -
sales of the focal itenl are 900, sales of other items in the focal category are 2,100, and
sales of other categories are 297,000.
As can be seen from the definition of promotion profit (equation 5.43), the changes in
sales caused by the promotion may be even more inlportant than the absolute levels. Figure
6.1 shows the change in sales for different sizes of the price deal when there is no special
display. The changes, calculated from Table 6.1, relate to the sales that would occur in the
absence of a price deal and special display. For example, the change in item sales for a 20
percent deal discount is thus 1,968 less 900 or 1,068 if the item is not on display. A deal
discount of 20 percent combined with a special display increase item sales by 2,660 less
900 or 1,760.
Sales of the focal item are more responsive to the price deal than category sales and
store sales. This implies that the promoted item cannibalizes on other item's sales. This can
be seen in Figure 6.1 in the decreasing sales of other itenls in the focal category.
The effect on store sales is a small increase. However, Figure 6.1 shows that sales of
other categories decrease as a consequence of the price deal on the focal item.
Figure 6.2 shows the changes in sales when the focal item is on display. Note that all
changes are calculated as changes from the zero deal discount and no special display case.
This causes the intercepts observed in Figure 6.2. The change in item sales caused by a
special display alone is 474, Le., 1,374 less 900.
When the focal item is on special display, this model predicts that store sales will
increase more than category sales when the price deal is smaller than ten percent. This
means that sales of other categories increase until the deal is ten percent. At larger price
deal discounts, sales of other categories decrease. The change in sales of other items in the
category is negative for all deal siscounts which indicates that there is intra-category
cannibalization.
Cannibalization was formally defined in Section 5.2.4 and we shall now use that
definition. Recall that intra-category cannibalization, <1>12 (Phi12), was defined as the ratio
of the change in category-sales to the change in item-sales (see equation 5.10). The inter-
category cannibalization, <1>23 (Phi23) was defined as the ratio of store-sales change to
category-sales change (see equation 5.11).
- 116 -
2500
2000
1500
~ 1000
.. .. -
. .... ...... ---0- Focal Item
.. .. .... ......
ca
en ----£r-- Other Items
.5
G)
C)
c
co
500
....... ...... .. .. • Category
- x - Other Categories
J:
o ---Store
-500
-1000
·1500
Deal Discount
Figure 6.1. Changes in sales at different levels by deal discount when the focal item is not
on special display.
2500
2000
.. .... ......
1500
~ 1000
...... • .. /II
ca
en
--0---- Focal Item
.5 ----tr-- Other Items

500
8,
c .... • Category
co
.c
o - x - Other Categories
---Store
-500
-1000
·1500
Deal Discount
Figure 6.2. Changes in sales at different levels by deal discount when the focal item is on
special display.
- 117 -
Table 6.3 presents the degree of intra- and inter-category cannibalization for different
deal discounts, with and without special display. An example will clarify how the Table
entries in Table 6.3 were calculated. Normal sales of the focal item are 900 and normal
category sales are 3,000 as we know from Table 6.1. A promotion consisting of a 12
percent price deal plus a special display increases item sales to 2,104 and category sales to
3,737. This means that the promotion increased item sales by 1,204 and category sales by
737. The ratio of category change to item change is 0.61, which can be seen in Table 6.3.
Deal .,;;";~;,,,;,;itr.;,,;h_=o,;;;;,:utIooo:D=oJ;.._·s~pl;,;:;;,a~y With Display

Discount phi12 phi23 phi12 phi23
0% 0.67 1.88
2% 0.72 0.20 0.66 1.59
40/0 0.70 0.20 0.65 1.38
60/0 0.69 0.20 0.64 1.22
8% 0.68 0.19 0.63 1.10
10 % 0.66 0.19 0.62 0.99
12% 0.65 0.19 0.61 0.91
14% 0.64 0.19 0.61 0.84
16% 0.63 0.19 0.60 0.78
18% 0.62 0.18 0.60 0.73
200/0 0.62 0.18 0.59 0.69
22% 0.61 0.18 0.59 0.65
24% 0.60 0.18 0.59 0.62
260/0 0.60 0.18 0.59 0.59
28% 0.59 0.18 0.59 0.56
30% 0.59 0.17 0.59 0.53
Table 6.3. Cannibalization (phi) by deal discount, with and without special display.
The interpretation of the table entry (0.61) is that 61 percent of the sales increase of the
focal item translates into a sales increase of the category. The remaining 39 percent of the
increase in the focal item's sales come from the other items in the category. Small values
indicate severe cannibalization while 1.0 indicates no cannibalization. If the ratio is greater
than unity, it is interpreted as a multiplier. Ratios greater than 1.0 can be found in the upper
right comer of Table 6.3 for the store-to-category sales changes. The combination of
special display and a deal of eight percent increases category sales by SEK 0.63 for each
SEK 1.00 increase in item sales. Each SEK 1.00 increase in category sales in turn increases
store sales by SEK 1.10.
6.2.3. Basket Response
The previous section examined the effect of price deal and special display for the focal
item on sales of that item, sales of other items in the focal category, and sales of other
categories. This section shows how the retailer's promotional effol1s for the focal item
- 118 -
affect the number of buyers of the focal item, the number of buyers of the focal category,
and the total number of baskets (i.e., store traffic). The aim is to provide some intuition for
the mechanism by which sales promotions affects sales, i.e., by having an impact on the
number of buyers.
Table 6.4 shows the number of buyers of the item, the focal category, and store traffic
(i.e., the total number of baskets) for different magnitudes of the price deal, with and
without special display. It can be seen that sales promotion has little impact on store traffic
for the hypothetical item in the simulation model. The number of category baskets is
somewhat more affected by sales promotion and the number of item baskets is fairly
responsive to sales promotion.
Deal Item CategorY Store

Discount No Displav Disp/av No Displav Display No Display Display
0% 38 59 150 165 3000 3003
2% 41 64 153 168 3000 3003
4% 45 69 156 171 3000 3003
60/0 49 74 159 175 3000 3003
8% 54 80 162 178 3000 3003
100/0 58 86 165 181 3000 3003
12°!c> 64 92 168 185 3000 3003
14% 69 99 171 188 3000 3003
16°!c> 74 106 175 192 3000 3003
18% 80 112 178 196 3001 3004
20% 86 119 181 199 3001 3004
22% 92 126 185 203 3001 3004
24% 99 134 188 207 3001 3004
26°!c> 105 141 192 211 3001 3004
28% 112 148 195 215 3001 3004
30% 119 156 199 219 3001 3004
Table 6.4. Number of baskets containing the focal item, the focal category, and store traffic
by deal discount with and without special display.
Note that in contrast to the sales models, where category sales are the sum of sales of the
focal item and sales of other items in the category, the number of category baskets is not
the sum of the number of item baskets and the number of baskets containing other items.
This is because category baskets may contain more than one item from the focal category.
Therefore, it is not necessarily so that the number of other item baskets must decrease even
if the number of category baskets remains constant while the number of baskets containing
the focal item increases. Figures 6.3 and 6.4 illustrate the increases in the number of item
baskets, the number of category baskets, and store traffic for different sizes of the price
deal discount. In Figure 6.3, there is no special display allocated to the focal item, while in
Figure 6.4, the price deal is accompanied by a special display.
- 119 -
Figures 6.3 and 6.4 clearly show that the change in the number of item baskets is larger
than the change in the number of category baskets. The change in the number of category
baskets is in tum larger than the change in store traffic. A way to quantify the relation
between the increases in item baskets, category baskets, and store traffic is to calculate the
ratio of the increase in category baskets to the increase in item baskets '1112 (psi12), and the
ratio of increase in store traffic to the increase in category baskets '1'23 (psi23). The first of
these ratios is the share of the increase in item buyers that are also new buyers of the
category. The second ratio is the share of new buyers of the category who represent extra
store traffic. Table 6.5 shows these ratios for different sizes of deal discounts, with and
without special display, for the focal item.
Deal No Display Deal and Display

Discount psi12 psi23 psi12 psi23
0% 0.72 0.20
2% 0.79 0.02 0.70 0.17
4% 0.77 0.02 0.68 0.15
60/0 0.75 0.02 0.67 0.13
80/0 0.73 0.02 0.66 0.12
100/0 0.71 0.02 0.64 0.11
12% 0.69 0.02 0.63 0.10
14% 0.68 0.02 0.62 0.09
16°tlc> 0.67 0.02 0.62 0.08
18% 0.65 0.02 0.61 0.08
20% 0.64 0.02 0.60 0.07
22% 0.63 0.02 0.60 0.07
24% 0.62 0.02 0.59 0.07
26% 0.61 0.02 0.59 0.06
280/0 0.61 0.02 0.58 0.06
300/0 0.60 0.02 0.58 0.06
Table 6.5. Psi by deal discount, with and without special display.
Table 6.5 should be read as explained by the following example: If the retailer allocates
a special display to the focal item and sets the deal discount at 18 percent, then 61 percent
of the extra buyers of the item are also extra buyers of the category, i.e., they would not
have been buyers of the category had the promotion not been executed. Eight percent of the
extra buyers of the category represent an increase in store traffic. These are consumers
attracted to the store by the sales promotion for the focal item.
For this hypothetical item, sales promotion mainly serves to increase the number of
buyers of the item while its effect on store traffic is assumed to be small.
- 120-
120
100
VI
j
=
m 80
..
'0
CD
.a
- - D - - Focal Item
E 60 • • • Category
~
z - - Store Traffic
.5
.. .
. . .. ..
".
8, ".
c 40
ca
.s::.
o
.. . ".
",.
.. . . .
".
".
20
.. ... .. . ".
.. .
".
".
~ ~ g; ~ ~ ~ ~ ~ ~ ~
cfl. &;
~
ft.
~
~ § ft.
~
Deal Discount
Figure 6.3. Changes in the number of item baskets, category baskets, and store traffic by
deal discount when the focal item is not on special display.
120
100
VI
j
m= 80
..
'0
.. ,. .. ......
- - D - - Focal Item
CD
.a
E 60
. ....
• • • Category
~
z
.5 . .. ". - - Store Traffic
...... ....
".
8,
o
c
ca
.s::.
40
...... ......
20 .... .. . .. .
Deal Discount
Figure 6.4. Changes in the number of item baskets, category baskets, and store traffic by
deal discount when the focal item is on special display.
- 121 -
6.3. Promotion Profits Measured at Three Levels
Promotional sales-response may be interesting in its own right, but our main interest in
this study is to define and investigate the retailer's profit of sales promotion. The aim of
this section is thus to demonstrate the profit impact of the sales-responses reported above.
In order to do that, we need to specify the parameters of the promotion-profit model
defined in Chapter 5. The following parameter values were used:
Gross margin of item (percent of regular retail price) ~ =0.20
Average effective gross margin of other items in the focal ~ =0.20

category
Average effective gross margin of items in other categories ~ =0.20
Trade deal offered on focal item (percent of regular retail price) TROFFER =0.20
Minimum discount required to obtain the trade deal DEALREQ =0 or 0.10
Indicator of display requirement to obtain the trade deal DISPREQ =0
This section will show that, given the models and the parameters specified above, the
retailer will come to different conclusions concerning the profitability of the promotion and
set different price deal discounts depending on the level at which the promotion profit is
evaluated.
As the parameters indicate, we assume that the retailer is offered a per unit trade deal of
20 percent of the retail price (i.e., 25 percent of the retailer's product cost when the retailer
cost of the product is 80 percent of the retail price). We evaluate the promotion profits for
two cases: (1) the retailer receives the trade money irrespective of the deal discount; and (2)
the retailer receives the trade money only if the deal discount is set at 10 percent or more.
In both cases, we investigate the promotion profits at the item-, category-, and store-levels.
6.3.1. Level of Analysis, Deal Discount, and Special Display
We start with the profit impact of sales promotion when there is no deal requirement,
i.e., when the retailer can absorb any part of the trade money. Figure 6.5 shows the
promotion profit measured at the item-, category-, and store-level at different deal
discounts when the focal item is not on special display. The retailer's promotion profit is
evaluated at two-percent intervals.
Note that the promotion profit measured at the item-level is higher than the promotion
profit measured at the category- and store-level for all deal discounts evaluated. This means
- 122-
that the item-level measurement of promotion profit overestimates the retailer's profit of
the promotion. Figure 6.5 shows that the promotion profit is negative for all deal discounts.
The point here is that the item-, category-, and store-level measurement come to different
estimates of the promotion profit. In other words, the estimated promotion profit is
dependent on the level of analysis used.
Even more important is that the three measures reach their highest value at different deal
discounts. This implies that the measurement level determines the deal discount that the
profit-maximizing retailer will set.
The discount that leads to highest gross margin from sales of the promoted product is a
discount of 14 percent. The category-level promotion profit peaks at a deal discount of 8
percent while the store-level promotion profit is highest when the deal discount is 0
percent.
Figure 6.6 shows the promotion profits of a special display allocated to the focal item
combined with different deal discounts. This time, the store-level promotion profit is
higher than the item-level and the category-level promotion profits for small deal
discounts. For larger discounts, the item-level and the category-level promotion profit
measurement again overestimate the promotion profit.
The deal discount which the profit-maximizing retailer will use depends, again, on the
level of analysis. Item- and category-level analyses lead to larger deal discounts than store-
level evaluation.
Note that the figures show the change in the retailer's gross profit from the item, from
the category, and from all categories. This is calculated as the gross margin from sales of
the item at each discount level and display status less the gross margin when the discount is
zero and the item is not on special display. In this case, the baseline implies that the retailer
absorbs the whole trade deal offered by the manufacturer.
The effect on category-level gross margin is small when the item is not on special
display and the discount is smaller than 20 percent. At larger discounts, the promoted item
has a lower margin than the other items in the category. This means that cannibalization
hurts gross profits. Special display also leads to cannibalization of other items' sales, but as
long as the effective margin of the promoted item is higher than that of the cannibalized
items, the retailer's gross profit from the category increases.
- 123 -
100.00
50.00
-50.00
:a -100.00
0:. --Focal Item
cac -150.00
.2 • - • Category
~ -200.00 --Store
o
et -250.00
-300.00
-350.00
-400.00
-450.00
Deal Discount
Figure 6.5. Promotion profits measured at item-, category-, and store-level when the focal
item is promoted without special display.
300.00
200.00
100.00
~
et 0.00 +--+--+---+---+---+----r------\---l---f~___+_----:...__.d_-_+_-__+_-___I_~_j --Focalltem
'ii
c
o - - • Category
:g
§ -100.00 --Store
et
-200.00
-300.00
-400.00
Deal Discount
item is on special display.
- 124-
The store-level promotion profit is the real concern of the retail manager (see Chapter
5). Figure 6.6 shows that the retailer would maximize promotion profit by absorbing all of
the trade deal, not . passing anything on to the consumers. Allocating a special display to the
item would increase gross profits by SEK 215 (see Table 6.5). If the display space would
not be worth more than SEK 215 when used for another item or category, the retailer
would maximize profits by displaying the item without a price deal.
The retailer's profits from different combinations of deal discount and special display for
the focal itenl are shown in Table 6.5. The profit-maximizing deal discount depends on the
level of analysis. If item-level gross profit is maximized when deciding on the deal
discount, the deal discount would be set to 14 percent when not on display while the
discount would be set to 10 percent when on special display. At the category-level, the
profit-maximizing deal would be 8 percent deal discount without display and 4 percent
with special display. The store-level analysis suggests a deal discount of 0 regardless of the
special display.
The cost (or opportunity loss) incurred by the retailer when using item-level analysis
rather than store-level analysis can be calculated as follows: Assume that the focal item is
not on display. The item-level analysis would lead the retailer to use a deal discount of 14
percent. The store-level result of that deal is SEK 67.40 lower than when the retailer
absorbs the whole trade deal, which is the optimal decision according to the store-level
analysis. If the item were displayed, the loss due to sub-optimization would be 214.93 less
149.31, or SEK 65.62.
Deal
Discount No Di Is
00/0 0.00 189.62 0.00 158.54
2% 13.75 204.04 9.06 164.19 -0.56
4% 26.10 215.82 15.88 166.481 -3.60
6% 36.74 224.57 20.15 165.08 -9.44
8% 45.31 229.92 1 21.531 159.69 -18.41
10 % 51.47 231 .51 1 -----19.......---70.... 150.00 -30.84
120/0 54.86 229.00 14.33 135.77 -47.07
14% 55.121 222.07 5.12 116.75 -67.40
16% 51.89 210.44 -8.24 92.73 -92.15
18% 44.84 193.84 -26.02 63.54 -121.59
20% 33.63 172.03 -48.48 29.03 -156.00
22% 17.96 144.84 -75.86 -10.91 -195.59
240/0 -2.44 112.09 -108.34 -56.37 -240.58
26% -27.84 73.65 -146.10 -107.40 -291.15
28% -58.45 29.43 -189.27 -164.05 -347.42
300~ -94.46 -20.65 -237.96 -226.32 -409.53
Table 6.5. Retailer's profits from a sales promotion evaluated at the item-level, category-
level, and store-level.
- 125 -
6.3.2. The Effect of a Deal Requirement
In the analysis above, it was assumed that the retailer could absorb the whole trade deal
offered by the manufacturer (although only for the products actually sold). We shall now
see how the results change when the manufacturer imposes the condition that the retailer
must discount the item by at least 10 percent to receive the trade deal worth 20 percent of
the retail price. Figures 6.7 and 6.8 show the retailer's promotion profits when the
manufacturer imposes a 10 percent deal discount requirement.
Figure 6.7 shows the promotion profit measured at the item-, category-, and store-level
when the focal item is on deal without display. Item- and category-level measurement again
overestimate the promotion profit as compared with the store-level measurement.
All three measures are negative for small deal discounts. This is because the retailer
does not receive the trade money unless the focal item is awarded a 10 percent deal
discount or more and the deal price elasticity is not large enough to increase the sales
volume enough to compensate for the lower per unit (effective) margin. The retailer's
effective nlargin on the item decreased because the deal discount is entirely financed by the
retailer at discounts below ten percent.
When the deal discount reaches 10 percent, all three promotion profit measures jump to
positive values because the retailer receives the trade money. (The jumps are actually more
direct than it appears in the figures).
The category-level and the store-level profit measures both peak at a deal discount of 10
percent, Le., the minimum deal discount required by the manufacturer. The maximum item-
level profit is obtained at a deal discount of 14 percent.
Figure 6.8 presents the promotion profits from the combination of a special display and
a price deal. The item-level and the category-level promotion profit measures overestimate
the profit at deal discounts greater than (or equal to) 10 percent. All three measures reach
their maxima when the deal discount is 10 percent. This means that the profit-maximizing
retailer would set the same deal discount irrespectively of the measurement-level if the
focal item is allocated a special display.
- 126-
250.00
200.00 . . . . ..
150.00
...
100.00
~
...
50.00 ...
~ --Focalltem
0.00 ........,'-=~---t---+----tf5-1--t---+-----t---+--+---+--~+-----t---+-IIIH-------l
,, .- • Category
~
r!- r!- r!- r!- r!- r!- --Store
-50.00
:! ~ ~ re ex)
(\J ,g
-100.00
-150.00
-200.00
-250.00
Deal Discount
item is promoted without special display and the manufacturer requires the retailer to
discount the item by 10 percent.
500.00
400.00
300.00
200.00 --Focalltem
• - • Category
100.00 --Store
0.00 _+_-+--__+-~~~--+---+--__+-__+_-_t_-_+_-t____t~__t_-___/_li...._____J
-100.00
-200.00
Deal Discount
item is on special display and the manufacturer requires the retailer to discount the item by
10 percent.
- 127 -
The numerical values underlying the figures 6.7 and 6.8 are supplied in Table 6.6. The
retailer profit maximizes when the focal item is discounted at ten percent. The optimal
discount is the same when the item is on special display and when it is not allocated any
extra display space. Store-level promotion profit is SEK 180 higher when the focal item is
allocated a special display. If the opportunity cost of display space is lower than SEK 180,
the retailer would maximize profits by displaying the item.
Again, the retailer would choose a too large price discount if promotion profits were
evaluated at the item-level rather than at the store-level. The manufacturer's deal
°
requirement, Le., that the item must be discounted by at least 1 percent, leads to a store-
level optimal deal discount equal to the manufacturer's requirement. Thus, the profit-
maximizing retailer would pass through half of the trade deal to the consumers.
Since the optimal deal discount is 10 percent at all levels of analysis when the focal item
is displayed, there is no loss incurred by the myopic optimization. Under tha assumption
that the focal item is not displayed, the retailer would set the deal discount to 14 percent if
item-level analysis were used. Store-level analysis would lead to a 10 percent deal. The
loss due to using the myopic approach is SEK 36.56 (Le., 149.16 - 112.60).
Deal Item level Profit Category Level Profit Store Level Profit
Discount No Displav Display No Display Display No Display Display
0% 0.00 94.81 0.00 63.73 0.00 120.12
2%) -2.96 87.18 -7.65 47.33 -17.27 92.84
40/0 -8.40 75.92 -18.62 26.58 -38.10 60.95
60/0 -16.64 60.70 -33.23 1.22 -62.81 24.18
8% -28.01 41.22 -51.79 -29.01 -91.73 -17.74
10% 231.47 411.5111 199.7011 330.001 I 149.1611 329.311
12% 234.86 409.00 194.33 315.77 132.93 302.84
14% [ 235.121 402.07 185.12 296.75 112.60 271.28
16%) 231.89 390.44 171.76 272.73 87.85 234.43
180/0 224.84 373.84 153.98 243.54 58.41 192.12
20% 213.63 352.03 131.52 209.03 24.00 144.17
22% 197.96 324.84 104.14 169.09 -15.59 90.49
240/0 177.56 292.09 71.66 123.63 -60.58 30.96
26% 152.16 253.65 33.90 72.60 -111.15 -34.46
28% 121.55 209.43 -9.27 15.95 -167.42 -105.83
30% 85.54 159.35 -57.96 -46.32 -229.53 -183.15
Table 6.6. Retailer's promotion profits at the item-, category-, and store-level for different
levels ofdiscount and special display status.
These analyses show that the manufacturer has reason to worry about the retailer's pass-
through of the trade deal to the consumers. The analysis of store-level profit of the
promotion shows that the retailer may be better off absorbing the whole trade deal.
However, if the manufacturer can design the trade deal in such a way that the retailer must
- 128 -
pass through at least a part of the trade deal in order to receive the discount, it may be
optimal for the retailer to offer the consumers a discount.
This is very important for the manufacturer because the retailer's pass-through has a
considerable impact on the nlanufacturer's sales volume. Recall that normal sales of the
item (in this store) are SEK 900. When the item is on a ten percent deal, its sales rise to
SEK 1,372 valued at regular price, while sales increase to SEK 1,972 if the price deal is
combined with a special display. The sales volume increase of the focal item is thus 52
percent when only on price deal and 119 percent when the deal is accompanied by special
display.
6.4. Pronlotion Profits of Four other Response Patterns

The previous sections have examined the impact of sales promotion on sales and profits
for a hypothetical item in a hypothetical store. The results are obviously influenced by the
choice of model parameters and the assumptions implied by the paranleters. Therefore, this
section examines four additional sets of model parameters, or scenarios, each implying a
different type of sales-response pattern. In these scenarios, the price deal response
parameters were varied as shown in Table 6.7. All other parameters were held constant
across the scenarios. Note that scenario 1 was described above.
Scenario
Model 1 2 3 4 5
Focal item's share of category sales 5 5 5 10 10
Focal category's share of store sales 1 1.5 1.5 1.5 1
Basket value 0.001 0.015 0.015 0.015 0.001
Store traffic 0.001 0.001 0.015 0.015 0.001
Category baskets' share of all baskets 1 1.5 1.5 1.5 1
Item baskets' share of category baskets 5 5 5 10 10
Table 6.7. Response model's price deal response parameters for scenario by model. Other
parameters remain constant across the scenarios.
In scenario 2, the price deal response of the category is stronger than in the base
scenario. The average basket size is also more responsive to the price deal in this scenario.
The other parameters are the same as in scenario 1. Scenario 3, differs from Scenario 2 by
assuming that store traffic is also responsive to the price deal. The item-level response is
assumed to be stronger in Scenario 4 while the other parameters are at the same level as in
Scenario 3. The last scenario keeps the strong item-level response from Scenario 4 while
all other parameters are set at the levels used in the base scenario.
- 129-
The analysis in this section focuses on the effects of price deal on sales and profits. The
effects of special display are not analyzed in this section to save space and enhance clarity.
6.4.1. Scenario 2: Moderate Cannibalization

This scenario assumes that the category's response to the deal discount is stronger than
in the base scenario. The average basket size is also more responsive in this scenario. The
combination of stronger category- and store-level responsiveness yields a scenario with
modest intra- and inter-category cannibalization. This can be seen in Figure 6.9 depicting
the changes in sales for various sizes of the deal discount.
The degree of cannibalization is very low for deal discounts below ten percent, but
increases progressively for larger deal discounts. The promotion profits measured at the
item-level, the category-level, and store-level are presented in Figure 6.10. In this scenario
(in contrast to Scenario 1), deal discounts below 20 percent are profitable at all
measurement levels.
Figure 6.10 suggests that the item-level measurement as well as the category-level
measurement gives upwardly biased estimates of the promotion profits. Table 6.8 shows
that using item-level analysis to determine the optimal deal discount leads to a 16 percent
deal discount while category- and store-level analyses both set deal discount at 12 percent. 8
Deal Sales Promotional Profits

Discount Item CategorY Store Item Category Store
00/0 900 3000 300000 0.00 0.00 0.00
2% 994 3091 300096 17.57 17.13 18.05
40/0 1095 3186 300192 34.04 32.25 33.54
60/0 1203 3283 300288 49.03 44.94 46.05
80/0 1319 3383 300384 62.14 54.80 55.15
10 % 1443 3485 300480 72.93 61.40 60.39
12% 1575 3591 300577 80.95 64.281 61.301
14% 1714 3701 300673 85.72 63.00 57.41
160/0 1862 3813 300769 86.761 57.10 48.26
18% 2016 3929 300865 83.58 46.14 33.37
20% 2178 4048 300962 75.69 29.68 12.31
220/0 2348 4171 301058 62.60 7.31 -15.38
240/0 2524 4298 301154 43.85 -21.38 -50.12
26% 2707 4428 301251 18.97 -56.76 -92.30
28% 2896 4563 301347 -12.46 -99.18 -142.29
300/0 3091 4701 301443 -50.86 -148.97 -200.45
Table 6.8. Scenario 2: Sales and promotion profits measured at the item-, category-, and
store-level for different sizes of the deal discount.
8Category- and store-level analyses give the same "optimal" price deal here because the promotion profits
are evaluated for every 2-percent (as opposed to every I-percent as used later) increase in deal size.
- 130-
In comparison with the store-level promotion profit of a 12 percent deal discount, the
store-level promotion profit is SEK 13.04 lower for the 16 percent deal discount. This
opportunity cost may not seem large. However, this implies that a retailer determining the
deal discount by ising the myopic-level analysis improve the promotion profit by 27
percent by switching to the store-level model. Another way to appreciate the value of the
store-level (holistic) analysis approach is to multiply the SEK 13.04 loss by seven to get the
profits forgone for this promotion during the week the promotion runs. Multiply again by
the number of promoted products during one week, say 300, and by the 52 weeks of the
year. This is, of course, an overly simplistic way to measure the value of the holistic
approach, but it gives at least an indication of the potential of a less myopic analysis.
- 131 -
2500
2000
1500
Focal Item
~ -0--
ii
en --tr-- Other Items
c
'Q; 1000 • • • Category
C)
c
as - x - Other Categories
.c
o
- - Store Sales
500
-500
Deal Discount
Figure 6.9. Scenario 2: Changes in sales by deal discount.
100.00
50.00
0.00 -G""'---+-----+----+----+----+----+----+---+---+---+----3OllI.---+-'-----I----.,f--~I______j
'0
a:
ii
-50.00 --0--- Focal Item
c
o • • • Category
:g
~ -100.00 --Store
Q.
-150.00
-200.00
-250.00
Deal Discount
Figure 6.10. Scenario 2: Promotion profit measured at the item-level, category-level, and
store-level for different deal discounts.
- 132-
6.4.2. Scenario 3: Store Traffic Generation
In this scenario, the price deal has a stronger impact on store traffic than in the previous
scenarios. In combination with the effect on average basket size, the impact on store traffic
leads to more responsive store sales. Figure 6.11 shows that the increase in store sales is
larger than the increase in sales of the focal item. There is some intra-category
cannibalization in this scenario, too, but category sales are more responsive to the deal
discount. Between categories, on the other hand, there is no cannibalization. Sales of other
categories than the focal category also respond positively to the deal discount. This is a
consequence of the price deal's impact on store traffic.
The profit impact of the price deal shows a different pattern in this scenario (Figure
6.12). The myopic (Le., the item-level) profit measure understates the true promotion
profit. What is interesting in this scenario is that the category-level analysis understates the
promotion profits even more than the item-level analysis.
The optimal deal discount is 18 percent when evaluated at the store-level. This is to be
conlpared with the 16 percent deal discount suggested by the item-level model and the 12
percent deal discount suggested by the category-level model. The intriguing point here is
that the sub-optimization is more severe when using a category-level model than when
using the most myopic model. Using a category management model can, thus, be counter
productive when intra-category cannibalization is combined with positive effects on other
categories. Table 6.9 gives the promotion profits needed to calculate the cost of myopia.
The item-level model leads to a loss of SEK 1.98 while the loss caused by the category-
level model is SEK 22.71.
Deal Sales Promotional Profit
Discount Item CategorY Store Item Category Store
0% 900 3000 300000 0.00 0.00 0.00
2% 994 3092 300180 17.68 17.36 34.91
4% 1095 3187 300360 34.26 32.70 67.27
6% 1204 3285 300540 49.38 45.64 96.66
80/0 1321 3386 300721 62.61 55.74 122.65
10 % 1445 3490 300901 73.54 62.58 144.78
12% 1577 3598 301082 81.69 65.701 162.59
14% 1718 3708 301263 86.60 64.65 175.59
16% 1866 3822 301443 87.761 58.97 183.32
18% 2021 3939 301624 84.70 48.22 185.301
20% 2185 4060 301805 76.91 31.95 181.08
220/0 2355 4184 301987 63.91 9.73 170.21
24 % 2533 4312 302168 45.21 -18.83 152.26
26% 2717 4444 302349 20.35 -54.12 126.82
28% 2908 4580 302531 -11 .10 -96.50 93.52
30% 3104 4721 302712 -49.56 -146.31 52.00
- 133 -
3000
2500
2000
VI
G)
---0-- Focal Item
~ 1500 ----tr-- Other Items

.5
G) - - • Category
C)
c
.! 1000 - x - Other Categories
o - - Store Sales
500
·500
Deal Discount
200.00
150.00
100.00
:aD:
50.00 --0-- Focal Item
cac
0 • • • Category
:g
E 0.00 --Store
0
D:
-50.00
,,
-100.00
-150.00
Deal Discount
Figure 6.12. Scenario 3: Promotion profit measured at the item-level, category-level, and
- 134-
6.4.3. Scenario 4: Intra-Category Cannibalization and Store Traffic Generation
Intra-category cannibalization is more pronounced in scenario 4 than in scenario 3. The

responsiveness of store sales and category sales are equal to the previous scenario. Figure
6.13 shows the change in sales at different measurement levels for various deal discounts.
Item-level measurement exaggerates the promotion profits and leads to larger than
optimal deal discounts in this scenario. The effect of using category-level measurement is
the opposite, leading to smaller than optimal deal discounts (see Figure 6.14).
The optimal deal discount according to item-level analysis is 18 percent, while the
category-level analysis sets the deal discount at 12 percent. According to the store-level
evaluation of deal profits, the optimal deal discount is 14 percent. The sub-optimization
cost of the item-level model is SEK 12.02 while the loss incurred by using the category-
level model is SEK 8.34 calculated from the promotion-profit figures in Table 6.10.
Deal Sales Promotional Profit

Discount Item CategolJ' Store Item CategolJ' Store
00/0 900 3000 300000 0.00 0.00 0.00
2% 1063 3092 300180 43.75 29.71 47.26
4% 1243 3187 300360 87.51 56.37 90.94
6% 1441 3285 300540 129.80 78.75 129.78
80/0 1653 3386 300721 169.00 95.63 162.55
10 % 1878 3490 300901 203.44 105.88 188.09
12% 2113 3598 301082 231.56 108.521 205.41
14% 2354 3708 301263 251.95 102.81 213.751
16% 2598 3822 301443 263.49 88.26 212.61
18% 2843 3939 301624 265.371 64.64 201.73
20% 3085 4060 301805 257.08 31.95 181.08
22°k 3325 4184 301987 238.43 -9.66 150.82
24% 3559 4312 302168 209.44 -59.89 111.20
26% 3788 4444 302349 170.32 -118.39 62.55
28% 4011 4580 302531 121.36 -184.81 5.22
30% 4229 4721 302712 62.93 -258.80 -60.50
store-level for different sizes of the deal discount.
Note that the category:;)evel model performs better than the item-level model in this
scenario in spite of the positive effects on store sales. The reason is that the intra-category
cannibalization is too strong compared with the increase in store sales.
- 135 -
3500
3000
2500
2000
fIJ 1500 --0--- Focal Item

Q)
ii
en 1000
---l:r-- Other Items
.5
Q) - - • Category
0)
c 500
ca -x- Other Categories
.c
0
- - Store Sales
-500
-1000
-1500
-2000
Deal Discount
300.00
200.00
100.00
:aIi:
--0-- Focal Item
ii
c
o 0.00 - - • Category
:g
E --Store
o
Ii:
-100.00 ,
"
"
-200.00 "
-300.00
Deal Discount
Figure 6.14. Scenario 4: Promotion profits measured at the item-level, category-level, and
- 136-
6.4.4. Scenario 5: More Cannibalization
The last scenario is characterized by responsive item-sales and fairly inelastic category-
and store-sales. The implication is considerable intra-category and inter-category
cannibalization. Figure 6.15 shows how the sales-responses depend on the deal discount.
As expected, the item-level analysis grossly overstates the promotion profits. Also the
category-level analysis leads to an overly optimistic view of promotion profits. The store-
level profit (loss) of the promotion is almost unaffected by deal discounts on the focal item
up to deals of about 12 percent. Above this point, the store-level profit of the deal rapidly
becomes negative.
According to Table 6.11, the optimal deal discount is 6 percent. Retailers using
category-level analysis would set the deal at 10 percent, while those planning deal
discounts using the item-level model would reduce the price by 16 percent.
Deal Sales Promotional profit

Discount Item Category Store Item CategorY Store
0% 900 3000 300000 0.00 0.00 0.00
20/0 1051 3060 300012 39.55 21.28 11.66
4% 1217 3121 300024 78.25 39.06 19.58
6% 1396 3184 300036 114.68 52.24 22.661
8% 1585 3248 300048 147.34 59.79 19.85
10 % 1783 3313 300060 174.77 60.791 10.26
12% 1984 3379 300072 195.63 54.55 -6.85
14% 2188 3447 300084 208.82 40.59 -31.93
16% 2390 3516 300096 213.541 18.70 -65.21
180/0 2588 3586 300108 209.31 -11.07 -106.64
20% 2780 3658 300120 195.95 -48.48 -156.00
22% 2964 3731 300132 173.57 -93.15 -212.88
24% 3140 3805 300144 142.48 -144.57 -276.81
26% 3308 3881 300156 103.12 -202.22 -347.27
280/0 3467 3959 300168 56.03 -265.59 -423.74
300/0 3618 4038 300180 1.76 -334.19 -505.75
The 12 percent deal discount is SEK 2.81 worse than the optimal deal while the 16
percent deal costs the retailer SEK 87.87 in foregone gross profits. This indicates that using
the myopic analysis for setting the size of the deal discount can be costly when (1) item
sales are responsive to deal discounts and (2) cannibalization is an important source of that
incremental sales volume.
- 137 -
3000
2500
2000
1500
...... ......
UJ ---0-- Focal Item
.......
CD
ii 1000
(J)
.5
CD 500 ....... .... .... ---f:r--
..
Other Items
.. • Category
C)
c -x-
ca Other Categories
s=
0
- - Store Sales
-500
-1000
-1500
·2000
Deal Discount
Figure 6.15. Scenario 5: Changes in sales at by deal discount.
300.00
200.00
.. .. .. .. . . . . ..
100.00
... III
0.00
is ~ rfl.
~ ~ g; rfl.
D:. ~
ii -100.00 ----0-- Focal Item
c
o .... • Category
~ -200.00 --Store
o
D:.
-300.00
-400.00
-500.00
-600.00
Deal Discount
Figure 6.16. Scenario 5: Promotion profits measured at the item-level, category-level, and
- 138 -
6.5. Determinants of the Size of the Deal Discount
In this section, we examine some determinants of the size of the deal discount under the
assumption that the retailer maximizes current promotion profits. The determiants we
consider are: (1) the level of analysis, (2) the size of the trade deal discount, (3) deal
requirements imposed by the manufacturer, and (4) the gross margins of the focal category.
The size of the trade deal discount as a determinant of the retailer deal discount is
directly related to the question of retailer pass-through of trade deals. It will be seen that
three factors moderate the effect of the trade deal on deal discount. First, the level of
analysis that the retailer uses for determining the deal discount influences the deal discount.
It is assumed that the retailer will set the deal discount that maximizes promotion profits,
given the level of analysis used. Second, the gross margins of the focal category is
positively related to the deal discount. Third, the sales-response parameters (the scenarios
described above) are also important determinants of the deal discount.
Three additional issues are addressed in this section. The first is the promotion profit's
dependence on the trade deal. The second issue is how the trade deal influences the
opportunity loss due to myopic deal optimization. The third issue is the gross margin's
impact on the deal discount and the pass-through.
6.5.1. Level of Analysis, Trade Deal Discount, and Response Pattern

In order to investigate the impact of the level of analysis and the size of the trade deal
discount on the size of the retailer deal discount we need to define three variables; the
retailer determining the deal discount by using the item-level analysis would set the deal
discount to Deall. If the category-level analysis were used, the retailer would set the deal
to Dea12 while the store-level analysis would lead to Deal3.
These three variables were determined by evaluating the profit (measured at the item-
level, the category-level, and the store-level) for each 0.1 percent step of the deal discount
between zero and 100 percent. While this procedure does not find the exact optimal deal
discount, it provides reasonable precision and can be seen as an approximation of the
reality where the deal discount is not a perfectly continuous variable.
We start with the Scenario 1. Figure 6.17 shows how the deal discount depends on the
trade deal offered by the manufacturer and the level of analysis used by the retailer for
setting the deal discount. In general, the larger the trade deal, the larger the deal offered to
the consumers.
In Scenario 1, the item-level analysis leads to the largest deals and the largest pass-
through of trade deal money to the consumers, while store-level analysis leads to the lowest
pass-through.
- 139-
Indeed, small trade deals will be totally absorbed by the profit-maximizing retailer using
store-level or category-level analysis. Trade deals larger than 19 percent (of the retail price)
lead to small deal discounts in the store. The deal discount then increases linearly with the
size of the trade deal. This means that the relative pass-through (deal discount divided by
the trade deal) is higher the larger the trade deal, assuming that the retailer seeks to
maximize store-level promotion profits.
30%
25%
§ 20%
oCo)
til
C --Oeal1
-=
C
15% - - • Oeal2
.!! --Oeal3
~ 10%
5%
Trade Deal Discount
Figure 6.17. Scenario 1: Deal discount based on item (Deal1), category (Deal2), and store
(Deal3) level deal optimization asfunctions of the size of the trade deal offered.
Figure 6.18 presents the results for Scenario 2. Item-level analysis again leads to larger
deal discounts than the category-level and the store-level analyses. The difference between
the deal discounts determined by the three methods is smaller than in Scenario 1, but
increases with the size of the trade deal.
The deal discounts are approximately proportional to the trade deal discount. This
implies that the relative pass-through is constant. Retailers using the item-level deal
optimization pass through three quarters of the trade money while the relative pass-through
of retailers who use the store-level deal optimization is about 50 percent.
The results of Scenario 3 (Figure 6.19) differ from those of the first two scenarios; the
store-level deal optimization leads to larger deals than category- and store-level deal
optimization for most trade deals. Substantial trade deals, however, lead to larger deal
discounts when the retailer optimizes the deal discount at the item-level rather than at the
store-level.
- 140-
The relative pass-through is constant for category- and store-level deal optimization but
decreases with the size of the trade deal when using store-level optimization. The retailers
using store-level 9ptimization would add their own resources to the trade deal on those
smaller than 16 percent. This is due to the rather strong store sales effect of the price deal
in scenario 3.
In Scenario 4, store-level deal optimization leads to smaller deal discounts than item-
level optimization but larger than the deals determined by category-level optimization (see
Figure 6.20). Relative pass-through is independent of the trade deal when the retailer uses
category-level optimization. When using item or store-level deal optimization, the relative
pass-through decreases with the size of the trade deal.
Figure 6.21 presents the results for Scenario 5. Item-level optimization leads to larger
deals than category-level optimization, while deals set to optimize category-level
promotion profits are larger than deals determined by store-level deal optimization. This
scenario is similar to the Scenario 1. The difference is that the item-level sales-response is
stronger than in Scenario 1, which leads to a larger difference between Deal] and Deal3
than in Scenario 1.
In summary, these analyses show that the deal discount is positively related to the size
of the trade deal. Whether the relative pass-through increases, decreases, or is constant with
the trade deal depends on the sales-response (i.e., the scenario) and the level of deal
optimization applied by the retailer. The retailer's level of optimization (item, category, or
store) also influences the absolute pass-through, or the deal discount. Obviously, it is
important for the manufacturer to understand these mechanisms and have knowledge about
the sales-response and the level of analysis used by the retailer, in order to manage trade
promotions effectively.
- 141 -
30%
25%
20%
--OeaJ1
15% - - -OeaJ2
--OeaJ3
10%
5%
0% ~-+--+-+-+-+-+-+-+-+-+-+-+-+-+--+--+-I-+----+---1-+-+-+-+-+-+-+--+-+-+-+--+---+-t-+--+-f-+---j
Trade Deal Discount
(Deal3) level deal optimization asfunctions ofthe trade deal offered.
30%
25%
§ 20%
ou
(I)
c --OeaJ1
'IC 15% - • - OeaJ2
.! --OeaJ3
~a: 10%
5%
0% ~-+--+-+-l-+-+-+-+-+-\-+--+-+-+-+--t--I-+---t--l-+-+-+---t-+-+-+-+-+-\-t--t--+--+-+---+-f---I---1
Trade Deal Discount
(Deal3) level deal optimization asfunctions of the trade deal offered.
- 142-
30%
25%
20%
--Deal1
15% - - • Deal2
--Deal3
10%
5%
0% ...f!-+-+--+-+-t-+-+-+-+-+-+-+--+--+-t--+---+---1f--+--+--t-f--+-l--+-+-+-+-+-+-+-+--+--+-+----+-+-1--+---l
Trade Deal Discount
Figure 6.20. Scenario 4: Deal discount based on item (DeaI1), category (DeaI2), and store
(DeaI3) level deal optimization asfunctions ofthe trade deal offered.
25%
20%
wi •
'E
:::t
o
~ 15%
o --Deal1
m
CD -- • Deal2
Q
_ _ Deal3
~
! 10%
a::
5%
Trade Deal Discount
Figure 6.21. Scenario 5: Deal discount based on item-, category-, and store-level deal
optimization as functions of the trade deal offered.
- 143 -
6.5.2. The Cost of Myopic Optimization
We have seen how the deal discount depends on the size of the trade deal and the level
of optimization that the retailer uses. Now, we shall see how these two factors influence the
promotion profits measured at the store-level. In the following, Profit3 is the store-level
promotion profit of Deal3. Similarly, Profitl and Profit2 are the store-level promotion
profits resulting from Deall and DeaI2,respectively. The opportunity losses caused by
myopic deal optimization are Lossl = Profit3 - Profitl and Loss2 = Profit3 - Profit2.
Figure 6.22 presents the store-level promotion profits of Deall, Deal2, and Deal3 in
Scenario 1. It should be noted that the irregularities of the curves presented in this section
were introduced by the search routine used; the curves should really be smooth and regular.
Starting with Profit3, i.e., the store-level promotion profit of the deal optimized at the
store-level, Figure 6.22 shows that the promotion profits are zero for small trade deals.
This is because small trade deals do not induce the retailer to pass anything of the trade
deal through to the consumers. Recall that the trade money pocketed by the retailer is not
included in the promotion profit because this money has no effect on the retailer's decision
when the trade deal is not made conditional on a retail promotion. The retailer's promotion
profit increases rapidly with the size of the trade deal.
The store-level profit of category-level deal optimization, denoted Profit2, is zero as
long as the trade deal is small enough not to induce a price deal. As Deal2 and Deal3 are
equal when the trade deal is small, the sub-optimization cost is also zero. Moderate-sized
trade deal discounts lead to larger than optimal retailer deal discounts, causing opportunity
losses for the retailer. Large trade deals lead to positive promotion profits (Profit2). The the
opportunity loss is approximately constant for moderate and large trade deals (see Figure
6.23).
A sin1ilar pattern can be observed for the store-level profit of item-level deal
optimization, i.e., Profitl. Large trade deals induce the retailer to use deal discounts that
generate positive store-level profits. The opportunity cost due to sub-optimization is
constant for moderate and large trade deals.
- 144-
80
60
I
I
40
20 --Profit1
- • • Profit2
--Profit3
....
-20
-NM
~~~o~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
~~~OO~~~~~~~~~:::~~~~~N~~regM~~~~~~M~~
~~
-40
-60
Trade Deal Discount
Figure 6.22. Scenario 1: Store-level promotion profit of deal discount optimized at item-,
category-, and store-level asfunctions of the trade deal offered.
70 80
70
60
60
50
as 50
'is.
o 40 --Loss1
>-
:E
'5
i0 30
40 -- • Loss2
--Profit3
30
20
20
10 ~
~
10
~
I
Trade Deal Discount
Figure 6.23. Scenario 1: Sub-optimization costs and maximum promotion profit as

functions of the trade deal offered.
- 145 -
Figure 6.24 shows that the promotion profits rise sharply with the size of the trade deal
in Scenario 2. As category- and store-level deal optimization lead to almost the same deal
discount in this scenario, the opportunity loss due to category-level optimization is small.
The item-level deal optimization causes somewhat larger opportunity losses than the
category-level optimization. Figure 6.24 shows that the opportunity losses increase slowly
with the size of the trade deal.
The retailer's promotion profit, again, rises progressively with the size of the trade deal
in Scenario 3 (Figure 6.25), . Larger trade deals reduce the opportunity losses due to
myopic deal optimization. In this scenario, the opportunity loss due to category-level deal
optimization is larger than the opportunity losses caused by item-level optimization.
Figure 6.26 depicts the profit outcome and opportunity losses in Scenario 4 as functions
of the trade deal discount. Large trade deal discounts are much more profitable for the
retailer than small trade deals. The use of myopic deal optimization leads to small
opportunity losses for the retailer in this scenario.
The picture is different for Scenario 5 (see Figure 6.27). Small trade deals lead to no
incremental profits for the retailer (except for the absorbed trade money). For larger trade
deals, the retailer's profits rise fairly rapidly.
Category-level deal optimization costs the retailer little in terms of foregone promotion
profits. Item-level optimization, on the other hand, is more costly. Item-level optimization
leads to store-level promotional losses for small trade deals and to 20 percent lower than
maximum profits for large trade deals.
To summarize, the retailer's promotion profit is positively related to the size of the trade
deal. The severity of myopic deal optimization depends on the sales-response model. The
promotional sales-response also influences the relation between the trade deal discount and
the cost of sub-optimization. In addition, it determines whether the item-level deal
optimization is better or worse than the category-level deal optimization.
- 146-
30 300
25 250
20 200
lU
'6.
0 --Loss1
0
>-
==
_
U;
15 150 -. • Loss2
--Profit3
0
0
10 100
50
Trade Deal Discount

40 500
450
35
400
30
350
lU 25
'6. 300
0 --Loss1
>-
==
_
0
u;
20 250 -. • Loss2
--Profit3
0 200
o 15
150
10
100
50
Trade Deal Discount

- 147 -
26 700
2Q
... 600
500
ca
.0. 15
o ....... 400 --Loss1
~
'0
-;; .... 300
- • • Loss2
--Profit3
8 10
..... -- .. .. ..
...... 200
100
Trade Deal Discount

functions ofthe trade deal offered.
100 250
90
80 200
70
:=;
ca '5
.6. 60 150 0:. --Loss1
0
>-
== 50
"i
c
0 .. • Loss2
'5
1;)
:sE --Profit3
0 40 100 0
0
0:.
30
20 50
10
Trade Deal Discount

- 148-
6.5.3. The Effects of a Deal Requirement
We shall here investigate the effects of a deal requirement on (1) the optimal deal
discount, (2) the maximum promotion profit, and (3) the cost of myopic optimization of the
deal discount. It is assumed that the profit margins of the focal item, other items in the
focal catgeory, as well as other categories are 20 percent and the trade deal offered is 20
percent. Changing only the deal requirement and the scenario, we get the results presented
in Table 6.12.
Item DealReg Deal1 Deal2 Deal3 Loss1 Loss2 Ptofit3.

Scenario 1 00/0 130/0 8% 1% 56.71 18.42 0.01
Scenario 2 0% 16% 12% 11% 13.15 0.12 61.41
Scenario 3 00/0 16 % 13% 18% 1.98 15.59 185.30
Scenario 4 00/0 17 % 120/0 15% 5.99 8.98 214.39
Scenario 5 00/0 16 % 9% 6% 87.86 6.70 22.66
Scenario 1 1% 13% 8% 1% 56.71 18.42 180.01

Scenario 2 1% 16% 12% 11 % 13.15 0.12 241.41
Scenario 3 1% 16% 130/0 18% 1.98 15.59 305.30
Scenario 4 1% 170/0 120/0 15% 5.99 8.98 394.39
Scenario 5 1% 160/0 90/0 60/0 87.86 6.70 202.66
Scenario 1 10% 13% 10 % 10% 25.86 0.00 149.16

Scenario 2 10% 160/0 120/0 11 % 13.15 0.12 241.41
Scenario 3 10% 16% 13% 18% 1.98 15.59 365.30
Scenario 4 100/0 170/0 120/0 15% 5.99 8.98 394.39
Scenario 5 100/0 16% 10% 10% 75.46 0.00 190.26
Table 6.12.. Optimal deal discount, cost ofmyopia, and optimal deal profit.
The one-percent deal requirement has no effect on the optimal deal discount,
irrespective of the level of analysis. The ten percent deal requirement affects the category-
and store-level optimal deal discount for scenarios 1 and 5 while the item-level optimal
deal discount is not affected.
Although the store-level optimal deal discount is not affected by the one-percent deal
requirement, a comparison of the promotion profits shows that the profits increase due to
the deal requirement.
A comment about the effect of DEALREQ on Profit3 is necessary because it may seem
contra-intuitive that the deal requirement has a positive impact on Profit3. The reason is
that the promotion profit, i.e., Profit3, is defined as the change in retailer gross profits
caused by the deal. If the trade deal is not conditional on a price deal in the store, the effect
is merely to increase the gross margin on the focal item. The margin increase leads to larger
gross profits on the normal sales of the focal item. However, this is not included in the
promotion profit, because the price deal does not cause this change. On the other hand, if
- 149-
the trade deal is made conditional on a price deal in the store, the retailer would not benefit
from the higher gross margin on the normal sales of the focal item until the deal discount
reaches the threshold. At the threshold, the retailer's gross margin suddenly increases by the
size of the trade deal. The conclusion of this discussion is that although a deal requirement
leads to higher promotion Profit3, it can, as any restriction in an optimization problem (see
e.g., Eppen and Gould 1985), never improve the retailer's total gross profits. On the
contrary, the deal requirement can only hurt, or leave unaffected, the retailer's gross profits.
In the absence of the deal requirement, the retailer could absorb all of the trade deal and
thereby increase the store profits. However, this profit increment would not be regarded as
a pron10tion profit because it would not be caused by the retailer's deal discount. The effect
of the one-percent deal requirement is to make this incremental profit relevant for the
promotion decision and, thus, part of the promotion profit.
For example, the increase in promotion profits for scenario 1 from 0.01 to 180.01 means
that the retailer receives SEK 180 from the manufacturer without doing anything even if
there were no deal requirement.
A comparison of the promotion profits for a deal requirement of one percent and ten
percent shows how the deal requirement may hurt the retailer's profits. For scenario 1, the
ten-percent deal-requirement leads the retailer to increase the deal discount to ten percent
from the otherwise optimal 1 percent deal discount. This reduces store-level promotion
profits from SEK 180.01 to 149.16. The deal requirement thus costs the retailer the
difference of SEK 30.85.
6.5.4. Margins and Optimal Deal Discount
How does the gross margin of the focal category affect the pass-through of trade deals to
the consumers? To answer this question, we will assun1e that the retailer uses store-level
deal optimization. We also assume that the gross margin of the focal item is equal to that of
all other items in the focal category. The average gross margin of other categories is, as in
all the previous analyses, assumed to be 20 percent.
Figure 6.28 shows the results for scenario 1. The 20 percent category gross margin case,
used in the analyses in the previous section, is marked with a thick line. In this case, the
focal category and other categories have equal margins. Figure 6.28 shows that higher
margins clearly lead to greater pass-through of trade money.
Results for the other four scenarios can be found in the appendix to this chapter. The
pattern is the same for all scenarios; the higher the margin, the larger the deal discount.
- 150-
25%
20%
C Margin
::::s
o(,)
U)
C
iii 15% - • • 10%
G)
C --20%
.!i - - - 30%
~a: 10% --40%
iii
E
:s.o
5%
Trade Deal Discount
Figure 6.28. Scenario 1: Store-level optimal deal discount as a function of the trade deal
for different category gross margins.
- 151 -
6.6. Meta Analysis of the Deal Discount and the Cost of Myopia
This section attempts to produce an overview of the impact of the level of analysis, the
response patterns, the trade deal offered, the deal requirements, the margin of the focal
item, and the margin of other items in the focal category on the size of the deal discount
and the cost of myopic deal discount optimization. In contrast to the previous sections, we
will here study average outcomes.
6.6.1. Procedure
The following procedure was used to produce the simulation results used as input in the
meta-level analysis. For each of the five sales-response scenarios described in Section 6.4,
we examine two levels for each of the four following parameters:
(1) the deal requirement (DEALREQ) [0, 10],
(2) the size of the trade deal offered (TROFFER) [10,20]
(3) the retailer's gross margin on the focal item (~ or ml) [20, 30]
(4) the retailer's gross margin on other items in the focal category (~ or m2) [20, 30].
All four parameters are expressed in percent of the regular retail price. The symbols
used for these parameters appear within parenthesis () and the levels are presented within
brackets [].
The total number of combinations (or cases) examined is thus 80 (=5 x 2 x 2 x 2 x 2).
Table 6.16 in the appendix to this chapter presents the complete output of these analyses.
The results of these simulation runs will be presented and analyzed in a more compressed
format than in the preceding sections.
The six output variables examined were: Deall the myopic optimal deal discount
determined by using the item-level analysis and Deal2 is the deal discount determined by
category-level analysis. Deal3 is the optimal deal discount, which is determined by using
store-level analysis. These three variables were determined by evaluating the profit
(measured at the item-level, the category-level, and the store-level) for each 1 percent step
between zero and 100 percent deal discount and selecting the deals with the highest profit.
Profit3 is the store-level profit of the deal of size Deal3, i.e., the profit of the profit-
maximizing deal. The variables Lossl and Loss2, are the losses due to sub-optimization by
setting deal to Deall or Deal2. These are calculated as described in Section 6.5.2.
6.6.2. The IOlpact of the Level of Analysis
The impact of the level of analysis on the deal discounts and the losses due to myopic
optimization is examined by examination of the summary statistics of the simulation
output. The summary statistics are presented in Table 6.13. In these 80 simulations, the
- 152-
deal discount set by item-level profit maximization is in general larger than the optimal
deal. The category-level analysis, on the other hand, leads to slightly smaller deal discounts
than the average optimal deal discount.
While the smallest deal discount set by the store-level and the category-level analyses
was zero (i.e., no price deal), the minimum deal set by the item-level analysis was six
percent. The largest optimal deal in these simulations was 23 percent, which was also the
largest deal suggested by the item-level optimization. The largest deal suggested by the
category-level optimization was 19 percent.
Deal1 Deal2 Deal3 Losst Loss2 Profit3

Min 6.0 0.0 0.0 165.15 44.99 0.00
Max 23.0 19.0 23.0 0.00 0.00 579.62
Avg. 15.7 10.7 10.8 37.65 8.50 154.49
Stdv. 4.5 4.1 5.9 42.95 8.82 138.30
Table 6.13. Summary of simulation results (all simulated cases).
On average, the optimal deal discount returns a profit of SEK 154.49. The range and
standard deviation of the maximum profit were large in this set of simulation runs. The
category-level and the item-level analyses sometimes led to the same deal discount as the
store-level analysis; In those cases, the losses due to myopic optimization were zero.
The item-level analysis on average led to larger losses than the category-level analysis.
The largest sub-optimization cost observed in the simulations, SEK 165.~5, was caused by
item-level optimization of the deal discount.
Another way to analyze the outcomes of the simulation is to examine the frequency of
myopic deals larger than, equal to, and smaller than the optimal deal discount. This is
shown in Table 6.14, which reveals that item-level analysis is very likely to lead to larger
than optimal deal discounts, but leads to smaller than optimal deal discounts in eight of the
80 cases investigated here. The item-level deal was equal to the optimal deal in nine cases.
Item Level Optimization Category Level Optimization

# % # %
Myopic deal too small 8 10 % 30 38%
Myopic deal optimal 9 11% 18 23%
Myopic deal too Ia.rge 63 79% 32 40%
Total 80 1000/0 80 100 %
Table 6.14. The effect ofmyopic determination of deal discount. Absolute and relative
frequencies.
In 18 out of 80 cases the deal discount maximizing category profits coincided with the
deal discount maximizing store profits. In the remaining 62 cases, the category-level
- 153 -
optimization led to larger than optimal deal discounts as often as it led to smaller than
optimal deal discounts.
6.6.3. Scenarios, Trade Deals, Deal Requirements, and Margins

So far, we have seen the overall impact of the level of analysis used when the deal
discount id determined. We shall now examine the effects of some intervening factors.
Table 6.15 presents averages of the output variables broken down by scenario, margins,
trade deal, and deal requiren1ent.
The average deal discount suggested by the most myopic analysis (i.e., the item-level
analysis) is larger than the average optimal deal discount in four out of five scenarios.
Scenario 3 is the exception and we recall from the scenario descriptions in Section 6.4 that
the increase in store sales is much larger than the increase in the focal item's sales. Scenario
3 is the only scenario where the average loss due to sub-optimization is larger when using
category-level analysis than when item-level analysis is used. The category-level analysis
leads to an average deal discount that is larger than the average optimal deal discount in
three out of five scenarios.
Variable Variable
Name Level Deal1 Deal2 Deal3 Loss1 Loss2 Profit3 n
Scenario 1 13.5 8.9 5.3 49.76 8.94 50.06 16
Scenario 2 16.0 12.2 11.3 16.74 0.46 133.13 16
Scenario 3 16.0 12.4 17.1 2.65 16.86 254.43 16
Scenario 4 17.0 11 .1 13.7 16.82 8.77 261.17 16
Scenario 5 15.8 9.1 6.8 102.29 7.49 73.67 16
TROFFER 10 12.7 8.6 8.8 32.67 8.13 92.64 40
TROFFER 20 18.6 12.9 12.9 42.63 8.88 216.34 40
mt 20 12.7 8.6 8.8 32.67 8.13 115.14 40
mt 30 18.6 12.9 12.9 42.63 8.88 193.84 40
m2 20 15.65 11.575 11.85 27.298 7.767 168.64 40
m2 30 15.65 9.875 9.825 48.004 9.2343 140.34 40
DEALREQ 0 15.5 9.6 9.6 44.79 10.82 95.26 40
DEALREQ 10 15.9 11.8 12.1 30.51 6.19 213.72 40
Table 6.15. Average deal discounts, cost ofmyopia, and optimal profit (bivariate
analyses).
Not surprisingly, increasing the offered trade deal (TROFFER) from 20 to 30 percent
increases (1) the size of the optimal deal, (2) the size of the myopic deal discounts, and (3)
the promotion profit. Less obvious is that the larger trade deal leads to somewhat larger
costs of sub-optimization.
The effects of changing the margin on the focal item from 20 to 30 percent are the same
as when changing the size of the trade deal.
- 154-
Increasing the margin of other items in the focal category has no effect on the size of the
deal discount determined by using the item-level analysis. The size of the optimal deal
discount as well as the deal discount determined by using the category-level analysis, on
the other hand, were smaller when the margin on other items in the focal category was 30
percent than when the margin was 20 percent. In other words, the higher the margin of
other items, the smaller the optimal deal discount.
A requirement from the manufacturer that the retailer must pass through at least 10
percentage points of the trade deal to the consumers in order to receive the trade money
leads to an increase in the (average) optimal deal. Recall that the reason for this contra-
intuitive result is that the promotion profit is measured on an incremental basis, where the
increment is measured in relation to the gross profit the retailer would obtain if there would
be no retailer promotion for the focal item (see Section 6.5.3).
6.7. Conclusions
This chapter has shown that the retailer will set different deal discounts depending on
whether item-level, category-level, or store-level evaluation is used to find the optimal deal
discount. It is proposed that store-level deal optimization is the relevant level of analysis
for the retailer.
In general, item-level deal optimization leads to larger than optimal deal discounts. The
category-level optimization often leads to larger than optimal deal discounts. However. the
deal discount is likely to be closer to the optimal deal discount. The item-level and the
category-level deal optimization result in larger than optimal deal discounts because the
intra- and inter-category cannibalization are disregarded. However, if store traffic and store
sales are responsive to the sales promotion (i.e., when there is no cannibalization), the
item-level and the category-level optimization may lead to smaller than optimal deal
discounts.
It was found that the size of the trade deal affects the deal discount positively. The
relative pass-through is also dependent on the trade deal. However, the sign of the relation
depends on the sales-response and the degree of cannibalization. Deal requirements
imposed on the retailer by the manufacturer increase the size of the optimal deal discount.
The implications for retailers and manufacturers are important. For the retailer, the most
important observation is that item-level evaluation of promotion profits is myopic and
leads to opportunity losses. Also category-level analysis is insufficient for the retailer, who
should rather adopt the store-level perspective on promotion profits.
Manufacturers should also find the results interesting as the analysis shows that the size
of the trade promotion will have effects on the deal discount in the store. In order to
- 155 -
manage promotions effectively, manufacturers need to know how sales respond to retail
promotions as well as to the level of deal optimization used by retail managers.
Imposing deal requirements seems to be an effective way of increasing the pass-through
of trade deals to the consumers, because deal requirements alter the shape of the retailer's
profit functiun and generally increase the size of the optimal deal. This is important for the
manufacturer because the deal discount influences sales to the consumers.
Thus, retail managers as well as manufacturers should find the model described in this
chapter useful in promotion management. Point-of-sales scanner data would be useful for
parameterizing the models for such usage. Alternatively, the model could be regarded as a
decision calculus model (Little 1975a, 1975b; Lodish 1982) that can help retailers and
manufacturers to calculate the implications of the assumptions they make.
- 156-
Appendix
6.A.l. Gross margins, Trade Deals, and Optimal Deal discount
35%
30%
c Margin
S 25%
~
C
- • • 10%
.
120%
C
.!!
--20%
- - - 30%
ta:: 15%
--40%
'ii
E
8- 10%
5%
Trade Deal Discount
Figure 6.29. Scenario 2: Store-level optimal deal discount for different category gross
margins as functions of the trade deal offered.
40%
35%
§ 30% Margin
o
u
tn
C 25%
1C -
--20%
• • 10%
.!! 20%
- - 30%
ta:: --40%
'iii 15%
E
ao 10%
5%
0% +-+-+-+-+-+-+-1-+I-llr-----+-I-1-1-11-+-1-1-1-11--+--1+1-+1-+-1+1-+1-+-1+-1-+I-llr-----+-l-1-1-11-+-1-+1-11-+-1+1-+1--+--1+1-+1-+-1+1-+I---il
Trade Deal Discount
- 157 -
35%
30%
c Margin
g 25%
(,)
o
is
~
- • • 10%
.
.S!
20%
--20%
- - - 30%
~a: 15%
--40%
'i
E
g 10%
5%
0% +-+-+--+-+-+--+-+-+--+-+-+-+-+-+--+--t---+--I--t---+-t-t--+-t-+----+--,I---+--+---i-t--+-t---+--+--t-+--+-+---j
Trade Deal Discount
20%
18%
16%
Margin
§ 14%
o(,)
o
C 12% - • • 10%
.S! --20%
!
G)
10%
- - - 30%
a:
~ 8% --40%
:s.
o 6%
4%
2% tI
I
0% -1-+----I-.-+...-+-+-+-f-""""""'~..pt___Io-t-+-I+-l-j-+-t-+-+-+-+-+-+--+-+-t--+-+-+-+--t-+--+-t-+___t__1
Trade Deal Discount
margins asfunctions of the trade deal offered.
- 158-
6.A.2. Simulation,Results
Case Seen· DEALBEQ TBQFFER rot ro2 Dealt Deal2 Dea/3 Losst Loss? Profit3
1 1 0 10 20 20 6 2 0 36.12 8.92 0.00
2 1 0 10 20 30 6 0 0 44.42 0.00 0.00
3 1 0 10 30 20 13 8 1 56.71 18.42 0.01
4 1 0 10 30 30 13 6 0 79.29 17.73 0.00
5 1 0 20 20 20 13 8 1 56.71 18.42 0.01
6 1 0 20 20 30 13 6 0 79.29 17.73 0.00
7 1 0 20 30 20 20 14 8 67.43 15.99 18.25
8 1 0 20 30 30 20 12 5 99.34 18.27 9.10
9 1 10 10 20 20 10 10 10 0.00 0.00 12.00
10 1 10 10 20 30 10 10 0 3.89 3.89 0.00
11 1 10 10 30 20 13 10 10 25.86 0.00 59.16
12 1 10 10 30 30 13 10 10 32.56 0.00 43.27
13 1 10 20 20 20 13 10 10 25.86 0.00 149.16
14 1 10 20 20 30 13 10 10 32.56 0.00 133.27
15 1 10 20 30 20 20 14 10 65.50 14.06 196.32
16 1 10 20 30 30 20 12 10 90.67 9.60 180.43
17 2 0 10 20 20 8 6 6 2.52 0.00 15.75
18 2 0 10 20 30 8 5 5 4.43 0.00 14.00
19 2 0 10 30 20 16 12 11 13.15 0.12 61.41
20 2 0 10 30 30 16 11 10 21.20 0.20 54.62
21 2 0 20 20 20 16 12 11 13.15 0.12 61.41
22 2 0 20 20 30 16 11 10 21.20 0.20 54.62
23 2 0 20 30 20 23 19 17 23.70 2.07 145.36
24 2 0 20 30 30 23 17 16 38.01 0.95 129.58
25 2 10 10 20 20 10 10 10 0.00 0.00 96.08
26 2 10 10 20 30 10 10 10 0.00 0.00 90.31
27 2 10 10 30 20 16 12 11 13.15 0.12 151.41
28 2 10 10 30 30 16 11 10 21.20 0.20 144.62
29 2 10 20 20 20 16 12 11 13.15 0.12 241.41
30 2 10 20 20 30 16 11 10 21.20 0.20 234.62
31 2 10 20 30 20 23 19 17 23.70 2.07 325.36
32 2 10 20 30 30 23 17 16 38.01 0.95 309.58
33 3 0 10 20 20 8 6 13 14.47 28.80 95.06
34 3 0 10 20 30 8 5 12 9.69 30.61 86.84
35 3 0 10 30 20 16 13 18 1.98 15.59 185.30
36 3 0 10 30 30 16 11 16 0.00 21.35 168.92
37 3 0 20 20 20 16 13 18 1.98 15.59 185.30
38 3 0 20 20 30 16 11 16 0.00 21.35 168.92
39 3 0 20 30 20 23 19 23 0.00 12.24 316.44
40 3 0 20 30 30 23 17 21 1.73 15.60 288.65
Table 6. 16a. Simulation results.
- 159-
Case Seen. DEALREQ TROFFER m1 m2 Dea/1 Dea/2 Dea/3 Loss 1 Loss2 Profit3
41 3 10 10 20 20 10 10 13 4.79 4.79 185.06
42 3 10 10 20 30 10 10 12 2.05 2.05 176.84
43 3 10 10 30 20 16 13 18 1.98 15.59 275.30
44 3 10 10 30 30 16 11 16 0.00 21.35 258.92
45 3 10 20 20 20 16 13 18 1.98 15.59 365.30
46 3 10 20 20 30 16 11 16 0.00 21.35 348.92
47 3 10 20 30 20 23 19 23 0.00 12.24 496.44
48 3 10 20 30 30 23 17 21 1.73 15.60 468.65
49 4 0 10 20 20 12 6 10 6.14 14.55 90.27
50 4 0 10 20 30 12 3 7 28.90 18.07 51.52
51 4 0 10 30 20 17 12 15 5.99 8.98 214.39
52 4 0 10 30 30 17 9 12 29.54 10.23 143.89
53 4 0 20 20 20 17 12 15 5.99 8.98 214.39
54 4 0 20 20 30 17 9 12 29.54 10.23 143.89
55 4 0 20 30 20 22 17 20 6.34 9.20 399.62
56 4 0 20 30 30 22 14 17 27.15 11.84 296.38
57 4 10 10 20 20 12 10 10 6.14 0.00 180.27
58 4 10 10 20 30 12 10 10 18.87 0.00 131.49
59 4 10 10 30 20 17 12 15 5.99 8.98 304.39
60 4 10 10 30 30 17 10 12 29.54 4.58 233.89
61 4 10 20 20 20 17 12 15 5.99 8.98 394.39
62 4 10 20 20 30 17 10 12 29.54 4.58 323.89
63 4 10 20 30 20 22 17 20 6.34 9.20 579.62
64 4 10 20 30 30 22 14 17 27.15 11.84 476.38
65 5 0 10 20 20 11 3 0 95.63 7.11 0.00
66 5 0 10 20 30 11 0 0 159.37 0.00 0.00
67 5 0 10 30 20 16 9 6 87.86 6.70 22.66
68 5 0 10 30 30 16 6 2 165.15 11.09 2 ..53
69 5 0 20 20 20 16 9 6 87.86 6.70 22.66
70 5 0 20 20 30 16 6 2 165.15 11.09 2.53
71 5 0 20 30 20 20 15 12 69.61 10.22 101.59
72 5 0 20 30 30 20 11 8 134.86 7.37 44.62
73 5 10 10 20 20 11 10 10 17.63 0.00 12.00
74 5 10 10 20 30 11 10 0 69.37 44.99 0.00
75 5 10 10 30 20 16 10 10 75.46 0.00 100.26
76 5 10 10 30 30 16 10 10 115.90 0.00 43.27
77 5 10 20 20 20 16 10 10 75.46 0.00 190.26
78 5 10 20 20 30 16 10 10 115.90 0.00 133.27
79 5 10 20 30 20 20 15 12 69.61 10.22 281.59
80 5 10 20 30 30 20 11 10 131.76 4.28 221.53
Table 6.16b. Simulation results (continued).
- 160-
7. EMPIRICAL DATA AND ESTIMATION METHOD
7.1. Introduction
This chapter discusses the application of the promotion-profit model developed in
Chapter 5 to real data and can be seen as preparation for the three following chapters. In the
first section, the purpose of the empirical applications is discussed as- well as the criteria
used to select the cases. The second section discusses different types of data sources, with
an emphasis on scanner data. The third section presents the databases used for the present
research, Le., the databases collected at the Foundation for Distribution Research. Section
four shows the store-level data series, common to all three empirical illustrations. Last, the
estimation procedures used to parameterize the response models are described and
discussed.
7.1.1. Purpose of Empirical Illustrations

The purposes of the empirical illustrations are: (1) to show that the model can be applied
to real data, (2) to obtain estimates of how some different items' sales response to
promotion are composed, and (3) to illustrate the promotion profits for these items'
response structures.
7.1.2. Selection of Cases
Previous research has found that the promotional response varies across product
categories as well as across items within a product category (e.g., Bemmaor and Mouchoux
1991; Bolton 1989; Julander and Persson 1992; and Persson 1993). It is therefore important
to test the application of the model in more than one product category. It was decided that
applying the model to three product categories would suffice as empirical illustration of the
model's capabilities.
The following criteria were used for selection of the product categories for this study:
First, the category should have fairly large sales. Second, it had to be a category where
promotion had been used rather frequently. The categories selected are regular ground
coffee, breakfast cereals, and pasta.
- 161 -
Within each of these three product categories, the item selection followed criteria
similar to those used for category selection. The items should be fairly important in the
category and they must be promoted with some frequency.
The reasoning behind the case selection criteria is rather pragmatic and has to do with
the estimation of the response models. Clearly, it is a prerequisite for measurement of
promotional response that there have been some promotions. Promotion all the time, on the
other hand, would not have been useful either. The high volume criterion is important
because it reduces the problem with observations of zero sales. There are two problems of
observations of zero sales volume: (1) In general, zero sales volume is problematic because
it is a type of truncated data, which means that the dependent variable (sales) is cut off at
zero (since we cannot observe negative sales). Truncated data tend to bias the estimates of
the independent variables' effect. (2) In the binomiallogit models used here, observations
of zero sales are problematic because they are treated as missing data (because the
logarithm of zero is not defined).
Retail managers would also face these considerations when implementing systems based
on the model presented in this report. It is therefore important to note that the case-
selection criteria coincide with criteria for managerial relevance. The items are large-
volume products and consequently important for the store. The frequent promotion of these
items means that they consume much management time for designing the promotions.
This case selection procedure obviously does not result in a random sample.
Consequently, the results can not be used to make inferences about the population
parameters, i.e., the promotional response and profits of items in general in the store.
7.2. Data Sources

The data sources that can be used for parameterizing response models are discussed in
this section. First, historical data and experimental data are discussed. Then, information
technology is identified as a factor making huge amounts of data available to managers.
Third, different types of scanner data are presented.
7.2.1. Managerial Judgment, Historical Data and Experiments
There are three ways to parameterize a response function: managerial judgment,

statistical analysis of historical data, and experiments. Managerial judgment may be the
cheapest and quickest way to obtain and may in some situations be the only method
available. The obvious disadvantage of this type of data is that it is subjective rather than
objective data.
- 162-
Statistical analysis is a powerful collection of methods for making sense out of historical
data. It provides a systematic way of parameterizing response models. Using statistical
analysis of past empirical data for making forecasts is not to assume that the future will be
like the present. Then, we would not need the statistics. The assumption is rather that the
mechanisms at work are fairly stable. Experience has shown that this is a reasonable
assumption. Parsons et al. (1993) conclude that past data are useful for making many
practical marketing decisions and that when past data are available they should be utilized.
Parameters obtained from analyses of past data can be used as prior estimates that are
updated with judgmental inputs. Blattberg and Hoch (1990) found that statistical models
combined with managerial judgment outperformed either of the methods used on its own.
The past variance and covariance of the independent variables are important when using
past data for estimating the parameters in a response function. If the decision variables
have not varied in the past it is impossible to know the response to a change in an
independent variable. Also, even if the variable has varied enough to obtain an estimate of
its effect, the range within which forecasts are valid is restricted to the range the variable
has varied in the past. The response to a change in a decision variable may also be difficult
to estimate if the independent variables have been highly correlated. This is known as
multicollinearity and will be discussed later.
If past data do not provide enough variation or suffer from multicollinearity, an
experiment may be the best way to obtain estimates of the parameters of the response
model. Experiments are also useful when causality is an issue, because past data cannot
really be used to make inferences about causality. In an experiment, the treatment variables
are varied systematically and the response to these variations is observed.
Experiments can be conducted in a laboratory or on the field. Laboratory experiments
are the experiments that allow the researcher the greatest control over the experimental
stimuli and covariates. The external validity of laboratory experiments may be low. Field
experiments, on the other hand, do not suffer the from low external validity but cannot be
as thoroughly controlled as laboratory experiments. In-store experiments (a type of field
experiment) were reviewed by Doyle and Gidengil (1977). They found that in-store
experiments had not contributed much to marketing practice or marketing knowledge.
However, Chapman (1987) used a field experiment to investigate the profitability of a
retailer couponing. He found that an inexpensive experiment could help the retailer to
increase profits.
In comparison with past data, experiments have the drawback that the data are not
immediately available when a question arises. An experiment must then be designed and
conducted before the data become available to the researcher. Past data, on the other hand,
are available at any time the researcher needs them (provided that they have been collected
- 163 -
and stored). Other drawbacks of experiments are that they may be expensive and usually do
not give as many observations as past data.
Past data may be regarded as observations from naturally occurring experiments. A
manager who recognizes the importance of variance in the decision variables in order to
measure the impact of those variables can devise a long-term series of quasi-experiments
by systematically varying the levels of the decision variables. This can be done within a
tracking and feedback framework.
7.2.2. Information Technology
Recent developments in information technology, especially the rapid growth in storage

and processing capacity of computers, increase the availability of business data
dramatically. Companies in all kinds of industries can now store and process transaction
data at a disaggregate level. This means unprecedented possibilities to base business
decisions on relevant and timely data, or information.
New information technology has also entered the modem supermarket. Scanner
equipped check-outs and bar-coded merchandise promise productivity improvements in the
operations of the store. The immediate benefits for the retailer are faster and more accurate
check-out and the merchandise need not be price-marked.
Although more productive operations are valuable for the retailers, the transaction
databases that the computerized cash registers, or Electronic Point-of-Sales systems
(EPOS), produce as a by-product are even more valuable. The huge databases created by
the computerized check-outs may transfer power from the manufacturer to the retailer. Data
may by itself be useful for the retailer, but the real potential is when statistical methods are
used to combine data and models. Application of statistical methods for combining models
and data can be seen as a form of refinement of raw data to decision relevant information.
If retail managers learn to refine their own databases to decision support, they can improve
store profits and shift power from the brand manager or manufacturer.
Sales promotion, the subject of this report, is just one of many areas where the retail
manager and retail organizations could benefit from decision-support based on data and
models. As Lilien notes:
"New data sources are having a major impact on developments in
modeling markets. One of the most influential developments in the 1980s
has been the impact ofscanner data" Lilien (1993, p.13).
- 164-
7.2.3. Scanner Data in General
Measurement usually disturbs the system being studied, and the more the system is
disturbed, the worse the validity of the measurement. This is especially true when human
behavior is studied. Broadly speaking, there are two approaches to the study of human
behavior. We can ask the individual questions concerning the past, present, or intended
behavior, or, we can observe the behavior. If we choose to make observations, we can
either observe the behavior directly by direct observation, or we can observe the behavior
indirectly through the traces the behavior may leave. When human behavior is observed,
the individual being observed may be aware or unaware of being observed. The
cooperation of the observed individual mayor may not be needed. The less the individual
knows and perceives from the observation, the better the validity. What we are interested in
is therefore an unobtrusive method of studying behavior. In retailing, such data are now
available in form of transaction data collected by the point-of-sales systems.
There are five nodes on a packaged consumer good's way from the production at the
manufacturer to its consumption or final disposition in the consumer's home. Data can be
collected between these nodes. Curry (1989 and 1993) described the nodes and the data
collection points with the scheme shown in Figure 7.1. We will in this study use data
collected at the point-of-sales.
Manufacturer
factory
shipment
data
Warehouse
warehouse
r------*- withdrawal
data
point-of-sales
scanner data
Consumer
in-home
scanner
panel data
Product
disposition
Figure 7.1. Product flow, data collection points, and type ofdata.
- 165 -
Julander (1987) suggested that receipt data should be a valuable complement to the
traditional weekly-aggregated item sales data collected in the stores and to the panel data.
Cooper and Nakanishi (1988) note that the shopping basket is the basic transaction unit
in optical scanner point-of-sale systems. However, competitive analyses or marketing-
effectiveness studies focusing on a single category only utilize a subset of the information
on the receipt. Traditionally this subset is recorded in two ways. The sales of each item in
the basket are added to an accumulating record for each item. If the customer is a member
of a scanner panel, the transaction is sent to a record keeping track of that customer's
purchases in different stores (Cooper and Nakanishi 1988).
Store data contain item-level information about the sales and prices of all items in a
store during a specified time period. The time period is usually one week. Panel data, on
the other hand, contain household- or consumer-level information about all the purchases
of that household or individual.
Julander (1992) showed how valuable information can be gained by studying the receipt
data. He therefore urged retailers to collect and analyze scanner data on the receipt-level.
This is in line with the principles that McCann and Gallagher (1991) outlined for
merchandising databases. In order to maintain flexibility for future needs, a merchandising
database should store data at the lo\¥est levels of aggregation (McCann and Gallagher
1991). Hemant and Persson (1994) created single-source data of consumers' purchases and
household characteristics by combining receipt data with data from a survey. The
procedure used enabled the establishment of the exact links between questionnaires and
receipts.
Much of the recent research concerns panel data, as noted by Roberts (1990):
"... the availability of household panel data has caused an exodus of
interest from aggregate store-level data. This is unfortunate, since for
many geographic areas in the U.S.A. for which scanner data is available,
there is no panel. Also, in many instances, except for ad hoc analyses, the
additional expense of a panel cannot be justified" (Roberts 1990).
Blattberg and Neslin (1990) note that both store data and panel data are influenced by
noncoverage: the stores or panelists used to generate the data may not be representative of
the total market. This position is clearly a manufacturer's one. From the retailer perspective,
store data are not influenced by noncoverage. The data reflect purchases by all the retailer's
customers. This is in contrast to panel data, which trace the purchases of some of the store's
customers in several stores.
- 166-
7.3. The Scanner Data Project
The data used in this research were collected for a large research project at the
Foundation for Distribution Research at the Stockholm School of Economics. The project
aims to develop methods to extract decision relevant information from Electronic-Point-of-
Sales systems (EPOS) in grocery retailing (Julander and Holmberg 1987). The largest
Swedish grocery retailer, ICA, has made one store available for data collection. The store is
a fairly large (annual turnover approximately MSEK 80) and modem supermarket in a
Stockholm suburb. It is open between 8:00 a.m. and 8:00 p.m., seven days a week all year,
except Christmas Day, Easter Friday and Midsummer's Day. The store carries about 10,000
different items.
7.3.1. Data Collection History and Procedures

The store has been visited weekly since the project started in the spring of 1987. Weekly
sales figures, prices, and deal prices on item-level for all items in the store have been
collected since the project started by copying the sales file from the EPOS. In-store special
displays have been registered during the weekly visits to the store. In four product
categories, in-store special displays were recorded from the beginning; from October 1987
(week 41), all in-store special displays have been registered. Since the summer of 1988, the
EAN bar codes of displayed items have been registered using an optical hand-scanner. We
distinguish between two types of in-store special display: (1) special display in connection
to the product's regular shelf location, and (2) free-standing special display.
Newspaper feature advertising data have been collected by cutting out the
advertisements appearing in Dagens Nyheter (thw major daily Swedish newspaper). The
EAN codes for all items in these advertisements were recovered and entered in an
advertising datafile. From the beginning, we searched for the EAN codes of advertised
items in our article database, but now receive bar coded lists with the advertised items. We
still use the advertisements for coding the relative size of the item in the advertisement. We
also identify which day the advertisement occurred in the paper. The retail chain's leaflets
have been collected and coded using a similar procedure.
A pilot study for testing the usefulness of receipt data was conducted in October 1987.
A random sample of ten percent of the receipts during one week was manually entered in a
database. The results of the analyses, reported by Julander (1989, 1992), were encouraging
and led to the development of a system for the automatic logging of receipts. The system
was first tested in October 1990 and during the spring of 1991. Since August 1991, the
system has been continuously collecting electronic copies of all receipts generated in the
- 167 -
store. Due to technical problems, the receipt data collection has suffered a nUITlber of
interruptions which has lead to the loss of some periods' receipts.
7.3.2. Problems in Data
Although the data collection has proceeded relatively smoothly, it has not been without
problems. Some products' bar codes can be difficult to scan but the quality of the codes and
the scanners have improved over the years. Examples of products that have been difficult
are some brands of beer in cans and frozen food in plastic bags. The problem with bar
codes that are difficult to scan is that the check-out clerks prefer to punch in the price and
product category identification into the computer rather than making repeated attempts to
scan the bar code. The unit sales of such items are then understated. Another reason for
understated unit sales of certain items have been some sales promotions; the monetary sales
amount has then been entered on an aggregate promotional sales category. For sonle sales
promotions, for example "buy three, pay for two", the retailer creates and uses temporary
"EAN" codes. This is a good solution that does not lead to loss of information if the sales
registered at the temporary "EAN" are added to the sales registered at the real EAN before
analysis.
Sales of fresh meat and deli are only registered on an aggregate level in the cash register.
Unit sales volume for these products were consequently not available. We have not
collected information on display activity of fruits, vegetables, and bread. The reason for
this is that our data collection of in-store display has been done on a weekly basis while the
displays in these categories are changed several times during the week.
It may occasionally be difficult to define whether a product has a special display or not.
Some products have regular displays that have the appearance of special displays. Such
regular displays should not be registered as special displays. Therefore, it has been
important to have one person collecting the in-store data for a longer period. As the project
is now on its eigth year of data collection, it has not been possible to have the same person
doing the data collection over the full period. However, great care has been taken to train
new assistants in order to ensure the greatest possible continuity and reduce the variation
induced by the data collection.
The researcher using scanner data should be aware of these types of problems. This
makes the researcher's participation in the data collection process very valuable. However,
it should be emphasized that scanner data are generally very reliable data and that problems
with this kind of data should be regarded as deviations from what would be perfect data
rather than as shortcomings in comparison with other data sources. It can be noted that the
data for the items selected for the empirical illustrations are of very high quality.
- 168 -
7.3.3. The Database
The data collection gives us five data tables: four tables of weekly observations and one
table of transaction data. The common ill-fields in the four weekly tables are the thirteen
digit EAN code and a four digit number identifying the week. The four tables of weekly
observations are:
• item sales table
• special display table
• feature advertising table
• leaflet table
The four tables constitute a relational database that could be used for storing and
retrieving the data. However, the size of the database made data retrieval cumbersome. The
four tables were therefore merged into one single table. One record in this table contain
information about one item one week, and only one record exists with that combination of
EAN code and week number.
In addition to the weekly data, we have collected receipt data. The receipt data, or
transaction data, are stored in a separate data table.
Receipts are uniquely identified by a cash register ill, a four digit transaction number,
and the date. Each record in the receipt data table contains information about one receipt
line. Therefore, a receipt consists of one or more records in the receipt table.
When a customer buys two units of an item this can be represented in two ways on the
receipt. Either the purchase is registered on one line with the number of units purchased
registered or the purchase is registered on two lines. On the receipts produced in a cash
register, both methods are used. This makes analysis difficult and error prone. Therefore,
we have decided to use the first of the two alternative representations. If the same EAN
code appeared more than once on a receipt the lines were merged.
Data about sales promotion are not present in the original receipt data table. This
information had to be retrieved from the weekly database described above. The EAN code
and the ill of the week again served as key variables.
7.4. Description of the Store-Level Data

In order to make the reader familiar with the data, this section describes the store-level
data used in the study. First, the variables are defined. Second, some summary statistics are
provided and the variables' development over time is presented. The regularity of the
patterns that these variables exhibit is noteworthy.
- 169-
7.4.1. Definition of Variables.
Three variables are important for describing the store-level performance. Store traffic is
the number of shopping baskets that are checked out. This is reflected in the number of
receipts produced by the store's EPOS.
In this chapter, we define store sales as the actual store sales (in contrast to store sales
valued at regular price). This is the total amount paid by the store's customers and includes
value added tax. However, the sales model developed in Chapter 5 is defined in terms of
sales valued at regular price. Therefore, store sales valued at regular prices instead of actual
store sales were used in the empirical illustrations (Chapter 8, 9, and 10).
Store sales valued at regular prices are not directly available but can be calculated. As
the exact calculation turned out to be extremely cumbersome, the following approximation
was used: First, actual sales of the focal category were subtracted from actual store sales
which resulted in actual sales of other categories. Sales of other categories valued at regular
prices were then approximated by adding five percent (a figure obtained from comparisons
with the weekly sales data) to the actual sales of other categories. Finally, adding the focal
category's exactly calculated sales valued at regular prices rendered the approximate store
sales valued at regular prices.
The third variable is average basket value, or basket value for short. Basket value is
defined as daily store sales9 divided by daily store traffic. However, it is more useful to
think of store sales as the product of store traffic and basket value because that gives an
idea of how store sales are generated. Benishay (1965) used this approach and regarded the
number of customers and the dollar size of an individual sale as random variables.
7.4.2. Data Frequency and Temporal Aggregation
Temporal aggregation refers to aggregation across periods, e.g., as weekly observations

are aggregated daily observations. The longer the time period, i.e., the higher the degree of
temporal aggregation, the less random noise is left in the data. However, that does not
mean that the coefficient of determination must be higher when aggregated data are used.
Seasonal variation may account for a large part of the variance in the disaggregate series.
When the data have been aggregated, the season can no longer be used to explain the
variance in the aggregate series. Also, the variance of other explanatory variables is
reduced due to the temporal aggregation.
The main benefit of aggregation is a better overview than when disaggregate data are
used. Data may also be aggregated in order to correspond to a relevant decision-level. A
9In this chapter, actual store sales; in following chapters, store sales valued at regular prices.
- 170-
common reason for using aggregate data is, however, that data may not be available at a
more disaggregate level. As aggregation is an irreversible operation on the data, it is
advisable to store data in disaggregate format and aggregate to the preferred-level only for
the analysis.
The drawbacks of aggregation are that the number of observations available for
estimation of the response function decreases and that the parameter estimates may be
biased. An important source of the bias is, according to Foekens, Leeflang, and Wittink
(1994), that the disaggregate response function may be nonlinear while the aggregation is
usually the arithmetic sum.
Scanner receipt data come at a very disaggregate level, since each receipt is marked with
the date, hour, and minute of the transaction. This means that the choice of level of
temporal aggregation is not restricted by the data collection procedure. The choice was
therefore guided by the trade off between the need for a sufficient number of observations
and long enough periods to. reduce the problem caused by observations of zero sales
volume. This made daily observations appear to be the preferable level of temporal
aggregation.
7.4.3. Description of Store-Level Variables
The data were collected between May 18 and November 22, 1992. Due to technical
problems, the period of September 14 to October 4 was lost, and the store was closed on
June 26. Thus, the data sets available for this research contain 168 daily observations.
Table 7.1. presents summary statistics for the three store-level performance variables. It
should be noted that the figures presented in this section refer to actual sales (as opposed to
sales valued at regular prices). In the following chapters, however, sales valued at regular
prices will be used.
Store traffic Basket value Store sales

(Baskets) Index (SEIQ Index (SEIQ Index
Mean 2820 100% 96.21 100 % 277809 100%
Median 2849 101 % 89.86 93% 250916 90%
Standard Deviation 520 180/0 18.59 190/0 95315 34%
Range 3798 135% 129.40 135% 603124 217%
Minirnum 727 26% 42.37 44% 30804 110/0
Maximum 4525 160% 171 .77 1790/0 633928 228%
Table 7.1. Store-level summary statistics. n = 168 daily observations.
The average number of receipts produced by the store's EPOS was 2,820 per day with a
standard deviation of 520, or 18 percent of the average number of baskets. The overall
average basket value was SEK 98.52 during the measurement period. Computing the
- 171 -
average of the daily observations of avera~e basket value gives SEK 96.21 as the average
of the averages. This is the figure occurring in Table 7.1. The standard deviation of daily
average basket value was 18.59, or 19 percent of the average. The daily variation in basket
value and store traffic is thus approximately equal. Store sales, on the other hand, vary
more. The standard deviation of store sales is 95,315, or 34 percent of the average daily
store sales (SEK 278 thousand). The larger relative variance of the store sales indicates that
store traffic and the basket size are positively correlated.
Figure 7.2 shows the variation in daily store traffic over the measurement period. The
pattern is fairly regular with some few exceptions.
5000
4500
4000
3500
CJ 3000
=f!
t; 2500
o
en 2000
1500
1000
500
Figure 7.2. Store traffic over 168 days.
The large dip around period 120 needs some explanation. The explanation is that the
layout of the store was changed and during the reconstruction work, some parts of the store
were closed and not accessible to the customers. This made the store less attractive and
reduced store traffic dramatically. The six observations affected are September 13, and
October 5 to 9, i.e., one observation before and five observations after the period lost due
to technical problems.
Figure 7.3 presents the variation in average basket value over the measurement period.
This variable appears to follow an even more regular pattern than the store traffic variable.
- 172-
180.00
160.00
140.00
! 120.00
~
i 100.00
t=
m
80.00
~
..... 60.00
40.00
20.00
0.00 -tHttt+t+++H+++H+HtH-ftttt1K+tH-HtH-HtHtHtttt+t+++H+ffiH1K+tH-HtHtHtt+H+t++H+++H+Ht++ftttt1-HtH-ttttttHtttt+t+++H+HtH-HtH-ffiH1-HtH+ttH+ttH-H+t1
.... anmC?,..., .... anmC?,..., .... anmC?,..., .... anmC?,..., .... anmC?,..., .... anmC?,..., .... anmC?,..., .... anmC?,..., .... an
........ C\lC\lC\lC?C?'Ot'Ot'Otanan<O<O<O,...,,...,CXlCXlCXlmm~~~;:;:~~~~~~~~~~~~
Figure 7.3. Average basket value over 168 days.
700000.00
600000.00
500000.00
U)
Q)
"iii
en 400000.00
!o
tn
§ 300000.00
1)
~
200000.00
100000.00
Figure 7.4. Store sales over 168 days.
- 173 -
The average basket value also dips around day 120. In other words, not only did fewer
people choose to make their purchases in this store during the reconstruction period, but
those who did visit the store bought less than the usual average. Part of the explanation for
that is that not the whole assortment was available. The store's newspaper and tobacco
kiosk was not affected by the reconstruction. These customers accounted for a larger than
usual share of the store traffic which contributed to the reduction in average purchase
amount.
The correlation between store traffic and basket value is 0.68. Store traffic and store
sales have a correlation of 0.87 and the correlation between store sales and basket value is
0.94. The high positive correlations between store sales and the two other variables are not
surprising since store sales are the product of store traffic and basket value. The correlation
between store traffic and basket value indicates that the two variables tend to move
together; periods when more consumers visit the store, each of them spends a larger
amount on average. Being the product of two positively correlated variables, store sales
vary more than either of the two variables.
The six observations during the reconstruction period can be seen as exceptional
observations or outliers. Whether or not to include these observations in model building
can be discussed. It is likely that the models will not be affected by the inclusion or
exclusion of these observations, because the ratios should not be sensitive to store-level
changes. The coefficient of determination of the store-level model, on the other hand,
would be lower when these observations are included than when they are deleted from the
data. This was also confirmed in tests not reported here. The reason is that the
reconstruction period increases the variance of the dependent variables without affecting
the variance of the independent variables. This automatically reduces the coefficient of
determination. In the empirical illustrations reported in Chapters 8, 9, and 10, the six
observations were excluded in the estimation of the models. However, it should be noted
that all models were estimated with and without the six observations in order to test the
impact of these observations. As expected, the parameters were basically not affected while
the coefficient of determination increased.
We have seen that store traffic, basket value, and store sales follow fairly similar and
regular patterns. It is not difficult to find the source of these patterns: the consumers
purchasing behavior over the days of the week. Table 7.2 presents the averages for the
three store-level variables by day-of-the-week.
- 174-
Store traffic Basket value Store sales
(Baskets) Index (SEIQ Index (SEI() Index
Monday 2806 0.994 81.92 0.850 230740 0.828
Tuesday 2701 0.957 85.33 0.885 234202 0.840
Wednesday 2699 0.956 90.34 0.937 247919 0.890
Thursday 2917 '1.034 102.85 1.067 307699 1.104
Friday 3128 1.108 123.90 1.285 393542 1.412
Saturday 3060 1.084 107.75 1.118 332246 1.192
Sunday 2446 0.867 82.77 0.859 204384 0.733
Average 2822 1.000 96.41 1.000 278676 1.000
Table 7.2. Average store traffic, basket value, and store sales by day ofweek.
Note that during the three heavy shopping days-Thursday, Friday, and Saturday--the
number of shoppers is higher than during the rest of the week. Further, the average
purchase amount is also higher during these days. This means that the total sales volume
are even more concentrated on these three days than the store traffic variation alone would
suggest. Store sales thus fluctuate more over the week than the store traffic and the average
basket value.
7.5. Parameterization of the Response Models

This section describes and discusses the procedures used for estimating the parameters
of the response models. First, the independent variables to be included in the models are
discussed. Then, the transformations used to make the response models linear in the
parameters are presented. The procedure used to select the variables to include in the final
model and to estimate their parameters are described third. Fourth, the interpretation of
statistical significance in this model-building is discussed, and fifth, the indirect modeling
of item sales and its implications for the null hypothesis are discussed.
7.5.1. Selection of Independent Variables
Statistical techniques such as regression analysis can be seen as the bridge between
theoretical models and empirical observations. If the theoretical model fits the data well, it
provides a good representation of the data. The model can be said to correspond to the
facts. It is well known that this does not mean that the model is true in some absolute sense,
but it makes the model useful in several ways. The model can be seen as a shorthand
description of the data set used for estimation. If we have reason to believe that a new data
set would be similar to the first data set, the model would fit this data set, too. The model
can then be used as a description of the population from which the original data set was
drawn. A good model uses the variation in the independent variables to explain (or
- 175 -
represent) a large part of the variation in the dependent variable. In that sense, a high value
of the adjusted coefficient of determination is desirable.
Therefore, the sales promotion variables of other items and day of the week dummy
variables were included in the response models in addition to the sales promotion variables
of the focal brand. Competing stores' marketing variables constitute another type of
variable that would have been desirable to include in the model estimation in order to
explain more of the variance in the dependent variables, but these variables were not
available. However, as long as competitors' promotions are not correlated with the
promotions studied here, leaving them outside the model does not bias the parameter
estimates. That also holds for variables such as season and sales promotions for other items
and categories in the test store.
Scanner data are, as noted above, historical data reflecting the behavior of consumers
and retailers. As retailers tend to use combinations of promotions, scanner data generally
show a high degree of multicollinearity. In the research reported here, the sales promotion
variables were highly correlated. The method that was chosen to handle the
multicollinearity problem was to drop some of the promotion variables. The remaining
variables should thus be interpreted as combinations of the promotions. In particular, the
feature advertising variable was dropped from all models and the special display variable
was dropped in several cases.
The implication of dropping these variables is that the response parameters for the price
deal variable will be inflated. The most important effect of inflating the price deal response
parameters that the optimal deal size determined by the model may deviate from the "true"
optimum. However, the sign of the deviation depends on the specific model parameters in a
complex way and could best be determined by sensitivity studies of the model parameters.
An intuitive idea of the complexity can be gained by realizing that inflating the positive
response for the promoted item at the same time implies that the negative effects on other
items and categories are exaggerated. Retailers intending to use the model in their
promotional decision-making are urged to conduct experiments in order to be able to
obtain better estimated of the response parameters.
7.5.2. Linearizing Transformations before Estimation
The response models used in this study are the exponential model and the logistic model
(see section 5.3). Both these models must be transformed before their parameters can be
estimated with multiple regression. This section presents the transformations.
The general expression of the exponential model is:
Y = exp(a + f3 . X) (7.1)
- 176-
and can be linearized by taking logarithms on both sides. This gives:
In(Y) = a + f3 . X (7.2)
which is linear in the parameters. The error term, e, enters the original equation as
follows:
Y = exp(a + f3 . X + e) (7.3)
The binomiallogit model is in its general form:
Y = _e_x_p{.:....-a_+_f3_·X---.;.)_ (7.4)
1 + exp{a + f3 . X)
That can be made linear in the parameters using the following transformation (see e.g.,
Aldrich and Nelson 1984):
In(~)
1-Y
= a + f3 . X (7.5)
The error tern1 of the original model is thus:
Y = _ex_p.....;...{a_+_f3_'X_+-----'-e) _ (7.6)
1+ exp(a + f3 .x + e)
7.5.3. Estimation Procedure
Multiple regression was used for estimating the models. The regression models were
built using the following procedure. First, a maximum model including all the relevant
variables was specified and estimated. Second, the variables associated with parameters
insignificant at the ten-percent level were eliminated. The new, restricted model was
estimated and further variables might be eliminated. This was repeated until all parameters
in the model were significant at the ten-percent level.
The procedure can thus be described as a stepwise regression using a backward
elimination strategy (see e.g., Kleinbaum, Kupper, and Muller 1988). However, the
automatic stepwise regression procedure of the statistical software could not be used to
implement the variable selection. This is because stepwise regression programs eliminate
variables using undirected hypotheses. In contrast, most of the hypotheses concerning the
parameters in these models are directional; as it is assun1ed that a sales promotion would
not reduce sales of the promoted item, we can interpret negative parameters as
- 177 -
insignificantly positive. Variables having parameters with "wrong" signs were therefore
eliminated.
In the estimation of the item-level models, the decision variables related to the focal
item were treated slightly different from the other independent variables. The focal item's
variables were not eliminated in early steps of the procedure even if the parameters were
insignificant. Only if the sign of the parameter remained wrong when the variable was the
only variable left in the regression model was it eliminated. The reason is that
multicollinearity might make these parameters insignificant in the early steps of the
estimation procedure, while they become significantly positive at later steps. Throwing
these variables out at an early step would, however, be like throwing the baby out with the
bath water, since these are the parameters we really want to estimate.
The estimation procedure can be said to be guided by a face validity criterion. A model
has face validity if its parameters make sense from a theoretical and intuitive perspective.
For example, if the parameter estimate implies that sales promotion reduces sales of the
promoted product, the model does not have face validity. It could of course be a scientific
breakthrough, but chances are good that something is wrong with the model.
7.4.4. Regression Diagnostics
The estimated regression models were examined for deviations from the assumptions of
the classic linear regression model. The condition numbers (see e.g., Kleinbaum, Kupper,
and Muller 1988) indicated that the estimated regressions did not suffer from severe
multicollinearity. Further, plots of the studentized estimated error terms against the fitted
value of the dependent variable showed that there was no problem with heteroscedasticity
(nonconstant error variance) in the data (see e.g., Fox 1991). Ramsey's RESET test, which
was used to test for nonlinearity, did not indicate further nonlinearities after the linearizing
transformations described in Section 7.5.2. However, given the relatively low number of
different deal discounts used, we should not be too confident that the functional form is the
true model.
The Durbin-Watson d statistic indicated that the regression models suffer from positive
first order autocorrelation, Le., positive errors tend to be followed by positive errors and
negative errors tend to follow negative errors. Autocorrelated error terms have two
consequences for the regression results: (1) the standard errors of the parameters are
underestimated which implies that the significance levels are overestimated, and (2) the
amount of variance explained is overestimated. Autocorrelation does not lead to biased
parameter estimates (Kennedy 1992). The main problem with autocorrelation is thus that
the variables included in the model may not be statistically significant.
- 178 -
The regression models were therefore re-estimated using the two step pseudo GLS
(Generalized Least Squares) procedure suggested by Cochrane-Orcutt (see e.g., Ostrom
1978). The result was, as expected, that the amount of variance explained decreased while
the parameter estimates remained (almost) unchanged. More important is that the
parameters (with few exceptions) were still significant at the ten-percent level.
As the results of the pseudo GLS estimation were similar to the results of the OLS
estimates, the OLS results are reported and used in this report.
7.5.5. Parameter Significance

Models used to make conditional forecasts of sales and profits for given values of the
manager's decision variables can be regarded as normative models. When building such
models, it is important that the parameters are "correct". A minimum requirement would be
that the parameters have the right signs. Is it then important that the parameters are
statistically significant? It will here be argued that the answer is no, but that statistical
significance can be useful in order to obtain a parsimonious model.
The idea behind the significance testing of parameters in a regression model is that if the
model were re-estimated on new samples from the same population, the parameters would
be different in each estimation. The standard deviation of the parameter estimates could
then be calculated. The estimate of the standard deviation the parameter would have is
known as the parameter's standard error, and is used for hypothesis-testing. Parameters are
generally tested against the null hypothesis that the population value of the parameter is
zero. Thus, the numerical value of a significant parameter is not more "true" than that of
other, insignificant, parameters. The "truth" of a parameter value is better described by its
standard error, which can be used to estimate the confidence interval of the parameter.
Thus, it is more important that all the parameter values make theoretical sense than that
they are significant, and it might be suggested that all parameters having correct signs
should be included in the model.
However, insignificant parameters are generally close to zero (otherwise they would be
significant) and their variables have little impact on the dependent variable. We can then
use statistical significance is to discard variables having little impact on the dependent
variable. The problem is that the parameter may be insignificant because the variance of the
independent was insufficient during the measurement period.
Parsimony was the reason for using statistical significance as the variable selection
criteria when specifying and estimating the response models in the empirical illustrations.
The general model-building principle of parsimony, Le., that the model should not be more
complex than necessary, has in statistical model building a special reason. Including too
many parameters in the model reduces the number of degrees of freedom, Le., the number
- 179 -
of observations used to estimate the value of each parameter. This increases the standard
errors of the parameter estimates, and consequently widens their confidence intervals.
Including variables that have no influence on the dependent variable increases the
standard errors of the parameter estimates. If such an irrelevant variable is uncorrelated
with the other independent variables of the model, its inclusion does not bias the other
parameters. However, it is unlikely that it is perfectly orthogonal to the other independent
variables, implying that other parameter values may change when the irrelevant variable is
included in the model.
The conclusion of this discussion is that all variables should primarily be selected on the
theoretical plausibility of their parameter estimates. Variables included in the model in
order to make the model explain more of the variance should only be retained in the model
when statistically significant. Statistical significance is less important for the decision
variables.
7.5.6. Modeling Approach and the Null Hypothesis
On a methodological level, the modeling approach and the choice of null hypothesis are
worth some consideration. The indirect modeling approach, i.e., modeling item sales as the
product of four ratios and store traffic, is likely to provide conservative estimates of
promotion profits. This is because promotion for a single item has little, if any, effect on
store traffic and average basket value.
The null hypothesis in statistical estimation of model parameters is that the parameters
are zero, so small effects may be disregarded because of lack of power of the statistical test.
The effects at the item- and category-levels are larger (in relation to the variance of the
dependent variables) and less likely to be victims of type II errors. The item and category
models are ratios. An increase in the focal item's share of category sales automatically
means that other iten1s' or categories' share must decrease. If the lack of power leads us to
the conclusion that the focal item's promotion had no effect on store sales, then the
negative effects on sales of other items and categories become exaggerated, leading to a
conservative estimate of the promotion profits.
If a direct model would be used instead of the indirect model, there is a risk that the
promotion profits would be overestimated. The argument is the same as above. The item-
level positive effect of the sales promotion is fairly large and type II errors are not likely.
There is, however, a large risk that there will not be enough power to reject the null
hypothesis that the sales promotion has no effect on other categories' aggregate sales since
the effect would generally be expected to be negligible. A type II error would occur, with
the consequence that promotion profits would be exaggerated.
- 180-
8. EMPIRICAL ILLUSTRATION 1: COFFEE
8.1. Introduction
This chapter contains a case study of the coffee category in our test store. It is an
application of the model developed in Chapter 5 and simulated in Chapter 6. The chapter
starts with a description of the product category and the data. Then, response models are
specified and estimated. The profit impact of sales promotion is calculated and the optimal
deal discount and the retailer pass-through of trade deals are determined using the sales-
response models. The chapter closes with a section in which the conclusions that can be
dra\vn from the analyses are discussed.
8.2. Description of the Coffee Category

Two reasons were important for selecting the regular ground coffee as one of the case
studies. The first reason is that coffee is a very frequently promoted product category. The
second reason is that coffee is believed to function as a loss leader, Le., a product that
attracts more consumers to the store. As the model allows direct measurement of sales
promotion's impact on store traffic and store sales, a loss-leader promoted product like
coffee would seem suitable for illustrating the model.
There are eight brands of regular ground coffee sold in the test store. Each brand is
available in a number of varieties and three brands are sold in 250 gram as well as the
standard 500 gram package size. Pricing and sales promotion are homogeneous across the
varieties within each brand-size. The brand-size is therefore the lowest level of analysis
that will be used here. The term item will hereafter be used to represent a brand-size
although each brand-size actually consists of more than one SKU (Stock Keeping Unit).
Figure 8.1 presents market-shares of the items in the coffee category.
Figure 8.1 shows that three items dominate the coffee category. The analysis will
thererfore focus on the three items: Item A, Item B, and Item C. Item C has much higher
deal costs than Item A and Item B, which indicates that Item C has bought market shares by
using large discounts. An examination of the store's promotion calendar showed that Item
- 181 -
C, on a number of occasions, was sold with a fifty percent discount to shoppers buying ten
units.
Brand 8: 2509
Brand 8: 5009
I
Brand 7: 5009 I
+
Brand 6: 5009 I
+
(Item C) Brand 5: 5009 I
Brand 2: 2509
Brand 4: 5009
Brand 3: 2509
(Item B) Brand 3: 5009

•I
I
Brand 2: 5009
)I
(Item A) Brand 1: 5009
0% 5% 10% 15% 20% 25% 30% 35% 40% 45%

Share of Category Sales
Figure 8.1. Item sales in the coffee category during the measurement period. The bars
show actual sales (AS) and deal cost (DC).
8.3. Building the Promotional Response Models

This section describes the estimation of response models for the factors of the model
developed in Chapter 5. The specific models used in this chapter are directly based on the
model specification presented in Section 5.3.
First, the model building procedure is described. Then, the plausibility of the modeling
franlework is tested by examination of correlations among the variables and ratios of the
decomposition model (i.e., equation 5.5). Thereafter, regression models are built for the
store-level, the category-level, and the item-level measures. The section closes with an
overview of the regression models, and can be seen as a further test of the modeling
framework.
8.3.1. Model Building Procedure
The regression models were built using the procedure described in Chapter 7. The
nlaximum model (i.e., the model containing all the independent variables that are
considered potentially relevant) differed between the hierarchical levels. In the store and
- 182-
category-level models, the deal discount and the special display were used in the maximum
model, except for Item B, where only deal discount was included. In addition to these, six
dummy variables for the days Tuesday till Sunday were included in the maximum model.
In the item-level models, the maximum models were extended to include both the deal
discount and the special display variable for the focal item. Other items were included with
one of these variables.
Item Bls the price deal and special display variables of were strongly correlated. In fact,
price deal never occurred without a special display. The resulting multicollinearity made
the deal discount parameter assume the wrong sign (Le., negative instead of positive
parameter). As this parameter is that of greatest interest in this study, the special display
variable was dropped from the full model for Item B. It should be remembered that the
price deal parameter thus is biased upwards and should be interpreted as the price deal
response when Item B is on special display.
8.3.2. Initial Tests of the Model
The hierarchical decomposition of item sales into four factors (Le., store traffic, average
basket value, category-to-store sales, and item-to-category sales) is in itself definitional and
cannot be tested. Sales of any item in the store can always be described in terms of these
four factors. However, if we want to use the decomposition to explain variations in item
sales as caused by variations in store traffic, some testing is needed.
It is difficult to design conclusive tests of the validity of the model, but the correlations
between the variables entering the model should give some indication. Although high
correlation between, for example, store-sales and category-sales may be due to a causal
relation between the two variables, there may also be some common cause. Another
alternative explanation for a high correlation between store-sales and category-sales is that
category sales are a part of store-sales, and, if category-sales (of the focal category) make
up a large share of store-sales, the correlation will be high. However, if category sales are
small compared with store-sales, a high correlation would contribute to the credibility of
the model.
At the same time as high correlations cannot conclusively decide the validity, low
correlations per se cannot falsify the model. This is because even if store sales would have
a causal impact on category sales, other sources of variation in category sales may reduce
the correlation between store sales and category sales.
Bearing these caveats in mind, we go on examining the intercorrelations among the
variables (see Table 8.1).
- 183 -
Sf St wt lit
Bt 1.00
St 0.87 1.00
bit 0.38 0.40 1.00
sit 0.29 0.26 0.76 1.00
Table 8.1. Pearson correlations between store traffic (Bt), store sales (St), number of
coffee category baskets (bjt), and coffee category sales (sjt). All correlations significant at
the 1 percent level.
All the correlations were significantly different from zero at the one-percent level or
better. This gives an indication that the model may be useful for explaining variation in
sales. The use of ratios in the decomposed model of item sales reflects the belief that sales
at lower hierarchical levels vary proportionally \vith the next higher level's sales. Forming
the ratio would then neutralize the impact of the denominator. The correlations among the
ratios would then be lower than correlations among the variables. As Table 8.2 shows,
correlations among the ratios are lower than the correlations among the variables.
Bt st/Bt bit/Bt lit/St

Bt 1.00
St/Bt 0.65 1 .00
bjt/Bt 0.13 0.21 1.00
sjt/St 0.11 0.00 0.70 1.00
Table 8.2. Pearson correlations between store traffic (Bt), basket value(St/Bt), category-to-
store baskets (bjt/Bt), and category-to-store sales (sjt/St). All correlations, except those
underlined, were significant at the 1 percent level. Underlined correlations were
insignificant at the 10 percent level.
As noted above, these tests cannot conclusively prove or disprove the model. However,
the observed correlations do strengthen the credibility of the model. Further tests of the
decomposition model are provided by the explanatory power of the sub-models. The
development of the sub-models is the subject of the next few sections.
8.3.3. Store-Level Response Models
This section examines the store-level response to sales promotion in the coffee category.
Three types of independent varaibles are used: (1) price deal variables (deal discount as
fraction of regular price) for the items in the category; (2) special display indicators (coded
o and 1) for the items in the category; and (3) day-of-the-week indicators (coded 0 and 1).
The present analyses are partial in that the independent variables included in the models
are not exhaustive. For example, sales promotion in other categories and other stores' sales
- 184-
promotion are not included among the explanatory variables. It is thus assumed that these
variables are not correlated with the coffee promotion in our test store.
Response models were built for store traffic, store sales, and the average basket value.
The three response models defined at the store-level all had the same structure and the
same variables in the maximum model. The of the general response model equation that
was fitted is (in its linearized version):
K
In(Yt) = a + 'LfJkXkt + Ct (8.1)
k=l
where t indexes the time period and K is the number of independent variables. This type
of response model is referred to as an exponential response model because the dependent
variable is an exponential function of the independent variables. Note that the model
parameters obviously differ between the models. Equation (8.1) is presented to show the
structure of the three models. The variable Yt on the left hand side of (8.1) represents store
traffic expressed as number of baskets, store sales valued at regular prices, and average
basket value, respectively, in the three store-level regression models that were estimated.
Table 8.3 presents the parameters (and their standard errors) of the final models of store
traffic, store sales, and basket value.
Store Traffic Store Sales Basket Value

coefficient s.e. coefficient s.e. coefficient s.e.
Constant 7.95 0.01 12.41 0.02 4.48 0.01
Price Deal
Item A
Item B 0.21 0.10
Item C
Special Display
Item A
Item B
Item C
Day-of-the-Week
Tuesday
Wednesday 0.07 0.02
Thursday 0.27 0.08 0.20 0.02
Friday 0.12 0.03 0.54 0.05 0.40 0.02
Saturday 0.06 0.03 0.33 0.04 0.24 0.02
Sunday -0.13 0.03
Regression Statistics
adj. Rsq. 220/0 52% 71%
n 162 162 162
Table 8.3. Store-level response to sales promotion in the coffee category.
- 185 -
Sales promotion in the coffee category had little impact on all three store-level
dependent variables. This is what can be expected because the coffee category is just one
among a large number of categories in the store. On the other hand, it may be disappointing
for retailers since coffee is traditionally treated as a loss leader category whose purpose is
to attract store traffic. According to Table 8.3, none of the three largest coffee brands' sales
promotion had any impact on store traffic. The insensitivity to sales promotion in the
coffee category could, however, also be a statistical artifact, caused by a more or less
constant level of sales promotion.
Item B's price deal was the only sales promotion that had a positive effect on average
basket value. The standardized coefficient of Item B's price deal variable was around 0.1 in
the basket value model, indicating that the promotion explained about one percent of the
variance in the dependent variable over the measurement period.
The model parameters show that there are important variations in store traffic, store
sales, and the average basket value over the week. Average basket value and store traffic
follow the same pattern as store sales. Store sales are low on Mondays, increase over the
week until Friday. Saturday sales are slightly lower than Friday sales and Sunday sales are
at the same level as Monday sales.
Note that the basket value model's parameters are defined to be the difference between
the sales model's parameters and the store traffic model's parameters (see Section 5.3.2).
The deviations from this found in Table 8.3 are due to the different number of parameters
included in the three models.
Comparing the share of variance explained, Table 8.3 shows that the average basket
value is the variable with the largest share of variance explained by the models (also after
accounting for the number of independent variables included in the model.) One reason for
this may be that store traffic is more stable over the week, its variance therefore relatively
more attributable to other factors than the day-of-the-week and coffee promotion.
8.3.4. Category-Level Response Models
Two category-level response models were built, both representing category-to-store-

level ratios (see Figure 8.2). The logit model was used to model these ratios (compare
Section 5.3). The first response model measures the response in the ratio of the number of
category baskets to the total number of baskets (i.e., store traffic). The category-to-store
sales ratio is modeled in the second response model. The response models had the same
structure and independent variables in both models. The estimation equation of the logit
model is:
- 186-
(8.2)
where t indexes the time period and K is the number of independent variables. Equation
(8.2) should be interpreted in the same way as (8.1), i.e., as an illustration of the nl0del
structure. The independent variables used here are the same as those defined above. The
results of the regressions are presented in Table 8.4.
Both models account for approximately half of the variance of the dependent variables.
A difference between the two models is that four of the day-of-the-week dummy variables
were significant in the category-to-store baskets model while only one was significant in
the share of sales model.
As the category-to-store baskets and category-to-store sales are ratio measures, the a
priori hypothesis may be that there should be no variation over the week. This is what
could be expected in the share of sales model, but in the category-to-store baskets model it
is not necessarily so. The reason is that the larger the shopping basket, the higher the
probability that any product in the store will enter the shopping basket. As the average
basket value varies over the week (as we recall from the previous section), it is logical that
the category-to-store basket ratio varies too.
The category-level response to sales promotion is greater than the store-level response.
It is found that the basket ratio responds to price deals for Item A and Item B, and to
special displays for Item C. The price deal parameter for Item C was not significantly
greater than zero even when Item C's special display variable was omitted from the model.
As explained earlier, Item B's special display variable was not included in the model
because of collinearity. Item B's price deal parameter is thus biased as it accounts for the
effect of price deal in combination with special display. Item A's price deal variable had a
significant impact on the baskets ratio while the effect of its special display variable was
not significant.
In the sales ratio model, on the other hand, Item A's price deal variable and special
display variable both had significant impact. Other sales promotions having significant
impact on the sales ratio were Item B's price deal and Item C's special display.
It is noteworthy that the promotion response parameters are of greater magnitude in the
sales rqtio model than in the basket ratio model. This indicates that the average purchase in
the coffee category is also affected positively by sales promotion.
- 187 -
0.14 0.1
0.09
0.12
0.08
0.1 0.07
g 0.06 g
i 0.08
ia::
a: - - E8sket R:Ilio
!en 0.05 en
CD
'ii - - sales Rtio
~ 0.06 0
0.04
0.03
0.04
0.02
0.02
o mftttttH-o0.01
Figure 8.2. Category development over time.
Category/Store Baskets Category/Store Sales

coefficient 5.e. coefficient 5.e.
Constant -3.75 0.04 -5.06 0.06
Price Deal
Item A 1.12 0.18 1.36 0.34
Item B 3.33 0.29 5.55 0.48
Item C
Special Display
Item A 0.16 0.10
Item B
Item C 0.28 0.04 0.42 0.07
Day-of-the-Week
Tuesday
Wednesday 0.10 0.06 0.19 0.09
Thursday 0.18 0.06
Friday 0.27 0.06
Saturday 0.16 0.06
Sunday
adj. Rsq. 580/0 53%
n 162 162
Table 8.4. Category-level response to sales promotion in the coffee category.
- 188 -
8.3.5. Item-Level Response Models
We have examined the store-level and category-level response to sales promotions for
three items (brand-sizes) in the coffee category. Here, the item-level response will be
studied. Three models were built for each brand, namely item-to-category basket ratio,
item-to-category sales ratio, and average value of item purchase among item buyers.
Hereafter, these will be referred to as the (item) basket ratio, the (item) sales ratio, and the
(item) purchase size models. The two ratios were modeled using logistic regression. The
transformed estimation equation is defined as:
(8.3)
The purchase size was modeled using an exponential model which linearized estimation
equation is:
K
In(Yt) = a + Lf3k X kt + Et (8.4)
k=l
Note that the observations of zero sales of the focal item will be treated as missing
values when estimationg the models using equations (8.2) and (8.3). Therefore, the number
of observations used for estimation of the models will vary. The fact that observations of
zero sales are not used in the estimation implies that the parameters will be biased.
In contrast to the store-level models, the purchase size variable's parameters are not
defined to be the difference between sales and basket ratios (see Section 5.3.2). The
response models will be presented and discussed for one item at a time.
- 189-
Item/Category Item/Category A verage Item
Baskets Sales Purchase
Constant -2.34 0.10 -2.31 0.12 3.24 0.02
Price Deal
Item A 4.59 0.52 5.56 0.66 0.50 0.16
Item B
Item C
Special Display
Item A 1.47 0.15 1.71 0.19 0.24 0.04
Item B -0.97 0.12 -1.31 0.16
Item C -0.34 0.11 -0.54 0.14
Day-of-the-Week
Tuesday
Wednesday
Thursday 0.29 0.15
Friday 0.41 0.15 0.35 0.19
Salurday 0.44 0.15 0.40 0.18
Sunday 0.32 0.15 0.38 0.18
adj.Rsq. 81 % 80% 36°,10
n 161 161 161
Table 8.5. Response ofItem A to sales promotion in the coffee category.
Table 8.5 presents the results for Item A. The basket ratio and sales ratio models are
very successful in accounting for the variance in the dependent variables. These models
explain 81 and 80 percent of the variance. The purchase size model is not as successful,
nevertheless it accounts for 36 percent of the variance. Recall that the situation reversed at
the store-level, with little of the variance in number of baskets explained by the model
while more than half of the variance in basket value was accounted for by the model.
Item A's sales promotion, price deal as well as special display, has a fairly large impact
on the Item A basket ratio and the Item A share ratio. Its effect on the purchase size is
somewhat smaller, however positive, implying that the sales ratio is more responsive than
the basket ratio.
The competitors' sales promotion had significant negative impact on the basket ratio and
the sales ratio. Their impact on purchase size was not significant.
Contrary to what would be expected, day-of-the-week variations in the basket ratio and
the sales ratio were found. The effect of the day-of-the-week was insignificant in the
purchase-size model.
- 190-
Constant -0.95 0.09 -1.04 0.11 3.17 0.02
Price Deal
Item A
Item B 7.52 0.79 6.35 1.00 1.31 0.21
Item C
Special Display
Item A -0.79 0.14 -0.99 0.18
Item B
Item C -1.01 0.12 -1.16 0.15
Day-of-the-Week
Tuesday
Wednesday
Thursday -0.24 0.15 -0.38 0.19
Friday -0.31 0.15 -0.36 0.20
Saturday
Sunday
adj.Rsq. 53% 420/0 19°k
n 162 162 162
Table 8.6. Response ofItem B to sales promotion in the coffee category.
The results for Item B are presented in Table 8.6. The pattern of variance explained is
the same for Item B as for Item A, but the magnitude is smaller. Again, the item basket
ratio is more predictable than the item purchase size.
Item H's promotion was the only significant determinant of the purchase size and its
response parameters in the ratio models were large. Note that although the basket-ratio
response parameter is larger than the sales-ratio parameter, the purchase size responded
positively to sales promotion. This seemingly contradictory result is explained when we
recall that Item B's sales promotion had a larger effect on the category-sales ratio than on
the category-baskets ratio. Purchase size is not the quotient between sales ratio and baskets
ratio, but the quotient between item sales and item baskets.
Like Item A, Item B's basket and sales ratios are affected negatively by competitors'
sales promotion while the purchase size is not significantly affected by other brands'
promotion.
The basket ratio and the sales ratio were found to vary over the week, with both ratios
being slightly lower on Thursdays and Fridays.
- 191 -
coefficient s.e. coefficient s.e. coefficient I.e.
Constant -3.00 0.09 -3.11 0.12 3.17 0.05
Price Deal
Item A
Item B
Item C 2.08 0.60 4.36 0.76 1.15 0.20
Special Display
Item A -0.40 0.21
Item B -0.65 0.18 -0.44 0.22 0.51 0.09
Item C 2.61 0.21 2.24 0.29
Day-of-the-Week
Tuesday
Wednesday
Thursday
Friday
Saturday
Sunday 0.26 0.10
adj.Rsq. 800/0 82% 41 %
n 150 150 150
Table 8.7. Response ofItem C to sales promotion in the coffee category.
The response models for the last of the three brands, Item C, are shown in Table 8.7.
The variation in predictability is again the same as for Item A and Item B. The coefficient
of determination is above 80 percent for the two ratio models and 41 percent for the
purchase size model.
Item C's price deal had significant impact on all three variables. Its impact on purchase
size was positive. Item C's special display had significant impact on both ratios but no
effect on purchase size. An interesting observation is that while the special display
coefficient is higher in the basket ratio model than in the sales ratio model, the opposite
holds for price deal.
Both competitors' promotions hurt the sales ratio. In addition, Item B's promotion hurts
Item C's basket ratio but has a positive impact on the purchase size. This indicates that Item
B's promotion draws unproportionally from small buyers of Item C.
Unlike the models for Item A and Item B, the only day-of-the-week variable found
significant was the Sunday dummy variable in the purchase-size model. The purchase size
was slightly smaller on Sundays than other days. All other day-of-the-week variables were
insignificant.
- 192-
8.3.6. Overview of the Regression Results
The regression results are summarized in this section. First, the explanatory power of the
regression models are compared. Then, the regression models are compared in terms of the
number of independent variables retained.
Table 8.8 presents the amount of variance explained by the regression models. Item CIS
sales ratio model has the highest explanatory power of the 15 regression models. That
model explained 82 percent of the variance in the sales ratio variable.
Baskets Sales Quantity

Store Level 220/0 52% 710/0
Category Level 58% 53% n.a.
Item Level Average 71 % 680/0 32%
Item A 81 % 80% 360/0
Item B 530/0 42°k 190/0
Item C 80% 82% 41 %
Table 8.8. Adjusted R-squares.
Item CIS basket ratio model also explained a large part of the variance. Item A's basket
and sales ratio models both explained about 80 percent of the variance. Item B's models
explained less variance than Item A's and Item CIS models. A pattern observed for all three
items is that the basket and sales ratio models explain more of the variance than the
purchase quantity models.
At the store-level, the pattern is reversed. The store-level model if average basket value
explains 71 percent of the variance, which is more than the 52 percent explained by the
store sales model and the 22 percent explained by the store traffic model.
These observations support the indirect modeling approached used in this report. While
the item-level basket and sales ratios are fairly deterministic according to the regression
results, the variation in store traffic is to a large extent unexplained.
Sales Promotion Dav-of-the-Week

Baskets Sales Quantitv Baskets Sales Quantitv
Store Level 001 334
Category Level 3 4 n.a. 4 1 n.a.
Item Level Average 3.3 3.7 1.7 2.0 1.7 0.3
Item A 442 430
Item B 331 220
Item C 342 001
Table 8.9. Number ofsignificant model parameters.
Table 8.9 presents the number of significant model parameters of different types. At the
store-level, only one sales promotion parameter was significant. The day-of-the-week
- 193 -
dummy variables were the only variables retained in the store traffic and store sales
models. In the store-level quantity model (Le., the average basket value model), one sales
promotion parameter and four of the day-of-the-week dummy variables were significant.
In the item-level models, the more sales promotion parameters than day-of-the-week
parameters were significant. The pattern is clearly the reverse compared with the store-
level models.
- 194-
8.4. Deal Discount, Sales Response, and Profit Impact
This section integrates the sales-response models developed above to a comprehensive
model for each of the three items. For each item, the sales effects of different deal
discounts are exemined at the item-, category-, and store-levels. Sales promotion's effects
on the nUITlber of baskets at the three hierarchical levels are also studied. Note that we are
now studying absolute rather than relative measures. This is because absolute sales, not
relative sales, is what translates into profits.
The absolute-sales responses are entered into the promotion-profit model developed in
Chapter 5. Based on assumptions about the retailer's gross margins, coffee sales
promotion's impact on retailer profits, measured as different hierarchical levels, are
presented. The profit models in this section assume that:
1. The average margin of other categories is 20 percent.
2. The average margin of other coffee items is equal to the margin of the focal
item, 20 percent if not otherwise noted.
3. The manufacturer offers a trade deal of 20 percent if not otherwise noted.
4. There is no requirement that the retailer must pass through any part of the
trade deal to the consumers.
These assumptions were selected after consultations with industry experts. It should be
noted that the qualitative results of the analyses are fairly robust to changes in the gross
margins of the products. Numerical results will however depend on the choice of the
margins.
The store-level profit model enables the retailer to set the optimal size of the price deal.
Assuming that the retailer uses an item-level or a category-level profit model, we estimate
the cost of sub-optimization.
The retailer's pass-through of trade deals to the consumers is of great concern for the
manufacturers. It is shown how the retailer's pass-through depends on the gross margin, the
size of the trade deal, and the level of analysis the retailer uses when setting the deal price
(assuming that the deal discount is set to maximize profits).
- 195 -
8.4.1. Item A
The sales-responses to price deal for Item A are shown in Figure 8.3. Category sales
increase more than item sales, implying that there is no intra-category cannibalization due
to Item A's price deal. However, all of the increase in coffee sales comes from other
categories in the store. In other words, the price deal for Item A leads to complete inter-
category cannibalization. Store sales is thus not affected by Item A's price deal.
Formalized measures of the relations between changes in category sales and changes in
item sales (<1>12 or phi12), and between changes in store sales and changes in category sales
(<1>23 or phi23) were introduced earlier in this report. Table 8.10 presents these measures, as
well as their basket model equivalents ('¥12 or psi12 and '¥23 or psi23), for various deal
discounts.
Deal Discount phi12 phi23 psi12 psi23

2% 2.23 0.00 2.27 0.00
40/0 2.12 0.00 2.18 0.00
6% 2.03 0.00 2.10 0.00
8% 1.93 0.00 2.01 0.00
10 % 1.84 0.00 1.93 0.00
12% 1.75 0.00 1.86 0.00
140/0 1.67 0.00 1.79 0.00
160/0 1.59 0.00 1.72 0.00
18% 1.52 0.00 1.65 0.00
20% 1.45 0.00 1.58 0.00
22% 1.39 0.00 1.52 0.00
24% 1.33 0.00 1.47 0.00
26% 1.27 0.00 1.41 0.00
280/0 1.22 0.00 1.36 0.00
30% 1.17 0.00 1.31 0.00
Table 8.10. Phi and psi for different deal discounts for Item A.
Price deals for Item A lead to larger increases in category sales than item sales. This can
be seen in Table 8.10 in that phi12 is greater than unity for all deal discounts evaluated.
The complete inter-category cannibalization is recognized by a column of zeroes for phi23.
The profit impact of a price deal for Item A is thus likely to be exaggerated if measured at
the item-level or at the category-level. Figure 8.4 shows that retailer profits are maximized
by absorbing the whole trade deal offered by the manufacturer. However, the retailer using
the myopic, item-level, analysis would set the deal discount to 22 percent. This means that
the deal discount would be greater than the (20 percent) trade deal received.
- 196-
800
600
400
---0--- Focal Item
--£r-- Other Items
- - • Category
- - - Store Sales
Deal Discount
Figure 8.3. Sales changes due to price dealfor Item A coffee.
80.00
60.00 .... ......

.............. III
.. .
40.00
20.00
~11. 0.00
---D-- Focal Item
ii
c
0 -20.00 - - • Category
:g
E ---Store
0 -40.00
D:.
-60.00
-80.00
-100.00
-120.00
Deal Discount
Figure 8.4. Promotion profits ofprice deal for Item A measured at different levels
- 197 -
8.4.2. Item B
The sales-responses to price deal for Item B are shown in Figure 8.5. The picture is
rather different from what we saw for Item A. Store sales increase more than category sales
and category sales increase more than sales of the focal item. The reader is, however,
cautioned not to forget that the model parameters are likely to be biased upwards,
exaggerating the sales-response to Item B's price deal. We will proceed with the analysis as
if the model parameters were unbiased.
Table 8.11 presents numeric measures of the category-relative-item and store-relative-
category sales and basket increases. The two columns to the right shows an approximately
one to one relation between the increase in the number of category baskets and the increase
in the number of item baskets. The increase in category baskets is not accompanied by an
increase in store traffic. Despite this, store sales increased more than category sales, which
indicates that the average basket value increased as a consequence of Item B's price deal.
The effect on the average basket value as well as the effect on sales of other items in the
coffee category were stronger than expected.
Deal Discount phi12 phi23 psi12 psi23

2% 2.02 5.50 1.29 0.00
40/0 1.94 5.20 1.24 0.00
6% 1.87 4.91 1.19 0.00
8% 1.79 4.63 1.15 0.00
10 % 1.72 4.37 1 .11 0.00
12% 1.66 4.11 1.08 0.00
14% 1.60 3.87 1.05 0.00
16% 1.54 3.63 1.02 0.00
18% 1.48 3.41 0.99 0.00
200/0 1.44 3.20 0.97 0.00
22% 1.39 3.00 0.95 0.00
24% 1.35 2.81 0.94 0.00
26% 1.31 2.63 0.93 0.00
28% 1.28 2.46 0.92 0.00
300/0 1.25 2.30 0.91 0.00
Table 8.11. Phi and psifor different deal discounts for Item B.
The strong store sales-response to Item B's price deal means that store-level profits of
the promotion are underestimated when the item-level (myopic) analysis is used. Figure
8.10 shows the extreme profitability of Item B's promotion when evaluated at the store-
level. The profit-maximizing deal discount cannot be seen in Figure 8.10; the store-level
profit of Item B's promotion peaks at 34 percent.
Item-level deal discount optimization would lead to a price deal of 26 to 28 percent. Due
to the steep slope of the store-profit curve, the cost of suboptimization would be large.
- 198 -
18000
16000
14000
12000
::
ca
-0- Focal Item
Other Items
~ 10000 ---l!r--
Q) _ _X_ _
___x--x----x
x - - • Category
m
i 8000
....___X - x - Other categories
.c
o .,...--x - - Store Sales
6000 x.....--x
4000
y ~
2000
~p .-
o -bI~~;:~~.~.t:.~·i·~i:-~E=E=::==:=:=:=:=:=:=:
Deal Discount
Figure 8.5. Changes in sales due to price dealfor Item B coffee. I
3000.00
2500.00
;: 2000.00
'0
D- -0- Focal Item
cac
.2 1500.00 - - • Category
'0
E --Store
o
D-
1000.00
500.00
0.00
~ ~ ~ g; ~ ~ ~ ~ ~ ~ ~ ~ ~
&; eft.
~ § t\I
Deal Discount
Figure 8.10. Promotion profits ofprice deal for Item B measured at different levels
- 199-
8.4.3. Item C
The effect of Item C's price deal is simple to describe. The increase in sales of Item C
comes entirely from other items in the coffee category. In other words, intra-category
cannibalization is complete and category sales are not affected by the promotion. The
effects on other categories' sales are zero. Figure 8.11 illustrates these effects.
Table 8.12 puts exact numbers on lack of effect of Item C's price deal on store- and
category-sales. The store-relative-category increase is undefined because the category
increase was zero.
The implication of the sales effect pattern for promotion profits is not difficult to
conjecture. Gross profits from other items drop because their sales decrease, and the
increase in sales of Item C is not strong enough to inlprove category-level gross margins.
The store and category-level profit impact coincide, because the price deal for Item C
has no impact on other categories' sales. Figure 8.12 depicts the profit impact of Item C's
price deal.
The item-level promotion profit is positive but small. Compared to the category- and
store-level, item-level analysis gives a too positive picture of the true promotion profits.
Deal Size phi12 phi23 psi12 psi23

2% 0.00 n.a. 0.00 n.a.
40/0 0.00 n.a. 0.00 n.a.
60/0 0.00 n.a. 0.00 n.a.
8% 0.00 n.a. 0.00 n.a.
10 % 0.00 n.a. 0.00 n.a.
12% 0.00 n.a. 0.00 n.a.
14% 0.00 n.a. 0.00 n.a.
16% 0.00 n.a. 0.00 n.a.
18% 0.00 n.a. 0.00 n.a.
20% 0.00 n.a. 0.00 n.a.
22% 0.00 n.a. 0.00 n.a.
24% 0.00 n.a. 0.00 n.a.
260/0 0.00 n.a. 0.00 n.a.
280/0 0.00 n.a. 0.00 n.a.
300/0 0.00 n.a. 0.00 n.a.
Table 8.12. Phi and psi for different deal discounts for Item C.
- 200-
200
150
100
~ 50 --{}-- Focal Item

m
en ---l:s- Other Items
.5
G) - - • Category
en
c
ftS
s::. ~ - x - Other Categories
(,) -50 - - Store Sales
-100
-150
-200
Deal Discount
Figure 8.11. Sales changes due to price deal for Item C coffee.
5.00
0.00 ~--__iiiiiiiiii;±::::=--t----i----+--+-----j-----+---+--+--+---+--=::="¢-::::::-t----i
-5.00
:::: -10.00
'0
Ii: ~FocaJltem
m -15.00
c
o - - • Category
~ -20.00 --Store
o
Ii:
-25.00
-30.00
-35.00
-40.00
Deal Discount
Figure 8.12. Promotion profits ofprice deal for Item C measured at different levels.
Category- and store-level profits coincide.
- 201 -
8.5.1. Level of Analysis
This section deals with the question of how the level of analysis that the retailer uses for
optimizing promotion profits affects the size of the deal discount. The effect of a deal
requirement imposed on the retailer by the manufacturer will also be examined. All
amounts are expressed in SEK per day.
Item-, category-, and store-level pro~otion profits were evaluated for each percentage
point of deal discount between zero and 100. The deal with the highest item-level profits is
called Deal1 while category and store-level optimal deals are called Deal2 and Deal3.
The store-level promotion profit of Deal3 is Profit3. The store-level profits of the sub-
optimal price deals are denoted Profit1 and Profit2, respectively. The cost of myopic deal
optimization (opportunity loss) can then be denoted Loss1 and Loss2, where Loss1 =
Profit3 - Profit1 and Loss2 = Profit3 - Profit2 (see equation 5.66 in Section 5.5.3).
Table 8.13 provides the numerical results of the analysis.
Item DealReg Deal1 Deal2 Deal3 Loss 1 Loss2 Profit3

Item A 0°10 22°10 220/0 4°1o 40.10 40.10 0.90
Item B 0°10 27°10 280/0 30°10 65.25 33.38 2562.97
Item C 0°10 15°tlc> 0°10 0°10 7.24 0.00 0.00
Item A 1 °10 22°10 22% 4% 40.10 40.10 29.39

Item B 1 °10 270/0 280/0 30°10 65.25 33.38 2645.37
Item C 1 °10 150/0 10/0 1 °10 7.11 0.00 13.34
Item A 100/0 220/0 220/0 10°10 37.35 37.35 26.64

Item B 100/0 270/0 280/0 300/0 65.25 33.38 2645.37
Item C 10°10 15% 10°10 10°10 3.95 0.00 10.18
Table 8.13. Deal requirement, deal discounts, loss due to myopia, and optimal deal profit.
We first examine the results for Item A when the deal requirement is zero. The item-
level and the category-level optimal deal discount are both 22 percent, while store-level
optimal deal discount is only 4 percent. The 4 percent deal increases the retailer's profits by
SEK 0.90 compared with absorbing all of the trade money and selling Item A at regular
price. Using item-level or category-level optimization would cost the retailer SEK 40.10.
A deal requirement of 1 percent does not change the optimal deal discount for Item A.
The store-level promotion profit, however, increases to SEK 29.39. The increase reflects
the value of the trade money absorbed by the retailer if the trade deal could be fully
absorbed (compare Sections 6.5.3). The loss caused by myopic deal maximization is
unaffected by the one percent deal requirement.
- 202-
Setting the deal requirement at 10 percent increases the store-level optimal deal to 10
percent. The decrease in promotion profits caused by the deal requirement is fairly small,
SEK 29.39 less SEK 26.64, or SEK 2.75. The opportunity loss caused by myopic deal
optimization is at the same time reduced by this amount.
While myopic optimization leads to larger than optimal price deals for Item A, the effect
of myopic optimization is the opposite for Item B. This is because of the positive sales-
response of other categories and other items in the focal category. The deal discount and
pass-through are large irrespective of the level of analysis. Thus, deal requirements have no
effect on the optimal deal discounts. The change in promotion profit between the zero and
the one percent deal requirement is the value of the trade money the retailer could have
absorbed.
For Item C, it is only the item-level optimal deal discount that is unaffected by the deal
requirenlents. The reason is that item-level optimal deal discount is 15 percent (i.e., larger
than the 10 percent deal requirement) while the category-level and the store-level optimal
deal discounts are 0 when no deal is required.
Category-level and store-level optimization lead to the same optimal deal for Itenl C.
Consequently, there is no opportunity loss due to category-level optimization. The
opportunity loss caused by item-level deal optimization decreases with the size of the deal
requirement.
8.5.2. Trade Deal and Margins
This section examines how the size of the trade deal affects the size of the deal discount,
under the assumption that the retailer maximizes store-level promotion profits. It is also
illustrated how the profit nlargin of coffee affects the pass-through of the trade deal to the
consumers. Twenty percent margins of the focal item, the focal category, and other
categories are regarded as the base case and is marked by a fat line in the figures.
Figure 8.13 depicts the (store-level) optimal price deal discount for Item A as a function
of the size of the trade deal for different profit margins of the coffee category.
Small trade deals will not be passed through at all but are fully absorbed by the retailer
when the coffee margin is 20 percent. The retailer would find it optimal to pass through a
small part of the trade deal when the trade deal exceeds 15 percent. In fact, with a trade
deal above 15 percent, the size of the deal discount is a linear function (with a slope close
to unity) of the trade deal. The relative pass-through is therefore larger for large trade deals
than for small trade deals.
The first 15 percent of the trade deal can (in this particular case) be seen as an entry
ticket for the manufacturer. The implication for the manufacturer is that small trade deals
have no effect on consumer purchases because they fail to induce action from the retailer's
- 203-
side. However, there may be other reasons for the manufacturer to use small trade deals,
such as, for example, to load the retailer with inventories in order to keep competitors out
of the store, or to improve the retailer's margin without having to reduce the regular
wholesale price.
Figure 8.13 shows the effect on the optimal deal discount of different margins. The
higher the coffee margins, the larger the optimal deal discount. This is because the retailer
need to be less concerned with inter-category cannibalization when the margin of the coffee
category exceeds that of other categories.
The analysis highlights the importance of large trade deals in order to induce the retailer
to promote Item A if the coffee margins were low. Figure 8.13 shows that very large trade
deal discounts would be necessary to induce the retailer to promote Item A if coffee
margins were 10 percent. The manufacturer would have to use other methods than
discounting the merchandise to make the retailer offer a price deal on Item A, if margins
were 10 percent or lower.
Figure 8.14 shows that the retailer would pass-through all of the trade deal and sacrifice
some of the regular margin for Item B in order to maximize store-level promotion profits.
Even if gross margins in the coffee category were only 10 percent, and in the absence of
a trade deal, the retailer could increase profits by a promotional discount for Item B of
more than 20 percent. The gross profits from sales of Item B would obviously be negative,
but the increase in store-wide gross profits would outweigh the item-level loss.
The results for Item C are presented in Figure 8.15. The minimum trade deal required to
induce the retailer to a small price deal is 15 percent. The retailer will pass through about
half of the trade money for trade deals larger than 30 percent. As for Item A, relative pass-
through is higher the higher the trade deal.
The pass-through is very sensitive to variations in profit margins of the coffee category.
If margins were ten percentage points lower, only very large trade deals would have any
effect on the price deal. Raising the margins by ten percent, on the other hand, makes
relative pass-through for small trade deals greater than one. The size of the price deal
would increase at a slower rate than the size of the trade deal. The relative pass-through of
large trade deals is, thus, lower than the relative pass-through of small trade deals.
- 204-
35%
30%
c
S 25%
u
Margin
(/)
i5
- - • 10%
m20%
.
C
.!!
--20%
"iii - - - 30%
~ 15%
--40%
ca
E
g 10%
5%
Trade Deal Discount
Figure 8.13. Item A: Store-level optimal deal discount for margin by trade deal.
60%
50%
C
:s Margin
o
"~ 40%
C
caG) - - • 10%
c --20%
.!! 30%
~a:
30%
--40%
ca
E 20%
a
o
10%
0% +-+-t-+-+---+--+--+--+--+-+-+-+-+-+-+-+----/--+-+-+-l----+----+-----1f-----+--t--J-J----t--J--+-+-+-t-+-+-t-+-+---j
Trade Deal Discount
Figure 8.14. Item B: Store-level optimal deal discountfor margin by trade deal.
- 205-
16%
14%
C 12%
::I Margin
0
(,)
--
U)
i5 10%
(ij • 10%
CD
C --20%
.!!
'n;
8%
30%
Q)
a: --40%
(ij 6%
e
a0 4%
2%
0%
Trade Deal Discount
Figure 8.15. Item C: Store-level optimal deal discountfor margin by trade deal. Note that
the all four margins lead to the same optimal deal discount.
- 206-
8.6. Conclusions
This empirical illustration of the application of the framework model and its sub-models
found evidence supporting the credibility of the framework. The correlation between store
traffic and store sales was, as expected, found to be high. High correlations were also found
between store traffic and category baskets, and between store-sales and category-sales.
Importantly, the correlations between the ratios in the model were lower than the
correlations of the variables. All this taken together support the model framework.
Likewise, the coefficients of determination of the sub-models indicate that the modeling
approach makes sense.
It was found that the promotion profit looks different depending on the perspective.
Although not always true, item-level promotion profit often give an exaggerated estimate
of the promotion profits. In the absence of effects on store sales, item-level deal
optimization leads to larger than optimal price deals. The opposite can be the case if the
price deal affects store traffic.
Thus, cannibalization (within and across product categories) is an important issue for
the retailer who tries to optimize promotion. Item-level deal optimization fails to take
positive and negative store-level effects into account. This generally reduces the
profitability of promotions and may lead the retailer to run unprofitable promotions.
It was shown that the retailer's pass-through of trade money can be analyzed using the
proposed framework. Strong cannibalization leads to low pass-through of small trade deals.
Interestingly, the relative pass-through of large trade deal discounts is higher than the
relatively pass-through of small trade deal discounts. The profit margin of the category was
also found to influence the pass-through of trade deals. The higher the margin of the focal
product category, the more of the trade deal will be passed through by the retailer.
The model can help retailers make better use of sales promotion. However, the
manufacturer should also find the model useful as a part of a model for optimizing trade
deals. The consumers' benefits of the model are perhaps more indirect; better handling of
sales promotion first improves retail profits. Then, competition among stores will push the
excess profits down by giving the consumers more service or lower prices.
The numerical results presented above should be regarded with some caution. The
parameter estimates are not as stable as desired. This is partly due to the multicollinearity
found in this kind of data due to retailers' (probably wise) practice of using price deals and
special displays in combination and rarely one without the other.
- 207-
Appendix
S.A.I. Summary Statistics
Item Brand Size price Pi; PiVtd si; siV,; de dD

Item A 1 500g 23.00. 3533 21 % 125971 20% 24362 42
2500g 15.50 479 3% 9223 1% 0 0
Item B 3500g 20.80 4049 24% 111727 180/0 10088 34
3250g 12.00 575 3% 7164 1% 0 0
4500g 21.50 480 3% 11739 20/0 0 0
4250g 11.50 105 1% 1422 0% 0 0
Item C 5500g 20.00 4684 28% 261048 41°A, 107316 56
6500g 23.50 1416 8% 46366 70/0 6342 35
7500g 23.00 1468 90/0 53673 8% 8160 42
8500g 34.83 340 20/0 5596 1% 0 21
8 2509 20.95 7 0% 147 0% 0 0
Total 20.60 17136 101°A, 634075 1000/0 156267 230
Table 8.14. Summary statisticsfor the store's coffee sales: number of item baskets (bij),
item-to-category baskets (bijlbj), sales valued at regular price (s), deal cost (de), number of
days with special display (dp). Note that the item-to-category baskets shares do not sum to
100 percent. This is because some category baskets contain more than one brand-size.
Symbol Value
Category purchase probability bjt/Bt 3.590/0
Category's share of actual store sales (sjt-de)/A St 1.020/0
Actual category sales per basket (sjt-de)/Bt 1.01
Actual basket value ASt/Bt 98.52
Actual category sales per category basket (sjt-de)/bjt 28.10
Purchase incidences per category basket sumfbiit )/bjt 1.01
Table 8.15. Summary statistics for the store's coffee sales.
- 208-
8.A.2. Store and Category-Level Time-Series
10000 1000000
~!
~
1000 100000 1--
-St
8t
'---------
100 -HttH#ttttt_#ttttt -Htttlt_-HtttltHHHH_HHHH-Htttlt 10000
Figure 8.16. Store traffic (Bt) and store sales (St) over time (daily observations).
Logarithmic scales.
1000
100
fIJ
i~I
~ --bjt
--sjt
as
o
10
Figure 8.17. Category baskets (bjt) and sales-response (sjt) ofthe coffee category over
time (daily observations). Logarithmic scales.
- 209-
!I
~
16
(,)
mI
--sjt
Figure 8.18. Store sales (St) and coffee sales-response (sjt) over time (daily observations).
Logarithmic scales.
10000 1000
:J
~I
()
~e 1000 100
m
~
--Bt
~ --bjt
~ 16
(,)
Figure 8.19. Store traffic (Bt) and coffee category baskets (bjt) over time (daily
observations). Logarithmic scales.
- 210-
8.A.3. Item-Level Time-Series
90.00 ....,....-----------------------tr-----------r-
I
80.00
," 0.9
0.8
70.00
c ,.. 0.7
g 60.00 "I ' o
:;
E
c( 0.6 ';. r--------,
o ... ·s11/b11
.,
CD
~ 50.00
I C)
.c
(,)
0.5 ~ --b11/bjt
~ ,
~
CD
C)
40.00
, I'
.',
, .'. 0.4 E
o --s11/sjt
L.- --'
f!
~ 30.00
,"
~U'JII"
0
•
:~ ~
c(
: ~ ~~': :'\""'. : I I•• , ....,
I'
" ' , , . ',
'.,
0.3
20.00
0.2
10.00 0.1
Figure 8.20. Item A's development over time. Item sales to item baskets (sijtlbjt) on the left
hand scale. Item-to-category baskets (bijtlbjt) and item-to-category sales (sijt/sjt) on the
right hand scale.
70.00
0.9
60.00
0.8
C 50.00 0.7
::::s o
o ,"I :;
,,
E •
r:
c( .' 0.6 ';.
.',.o..' I'
r-------,
~ 40.00 o .... s31/b31

m
.c
e .,
JII
0.5 ~
en
- - b31/bjt
". ,." o
,. ... "'
::::s
.. \:
I •
'; 30.00
~
"
....
.,. 0
'.' '".... I •• , " ... 0.4 E

o --s31/sjt
~
c( 20.00
I,
I 0.
'
aI" ..
\ ......
'.,"
:':
.a ..
I' "
I
..
• •'
• • ',"
~"
I -- ,
,.
' ....., .. "
0.3
~
0.2
10.00
0.1
0.00
,... (0,...(0 .... ( 0 , . . . ( 0 , . . . ( 0 , . . . ( 0 , . . . ( 0 , . . . (0,... (0,... ( 0 , . . . ( 0 , . . . ( 0 , . . . <0 .... ( 0 , . . . (0 .... ( 0 , . . . (0
,...,...C\IC\1C')C')~~IOIO(o<O,...,...CXlCXlO>O>~~;:;:~~~~~~~~~~
Figure 8.21. Item B's development over time. Item sales to item baskets (sijt/bjt) on the left
right hand scale.
- 211 -
250.00
0.9
200.00 0.8
'E 0.7
~ o
o
E
~ 150.00 0.6
~~ "'-.-.-.-.-S-51-/b-51---'
CD
UJ
CO tft
.c
(J
0.5 ~ - - b51/bjt
~
D. o --s51/sjt
& 100.00 0.4 E'---------'
f!
~
~ 0.3
50.00 0.2
0.1
Figure 8.22.Item C's development over time. Item sales to item baskets (sijtlbjt) on the left
right hand scale.
- 212-
9. EMPIRICAL ILLUSTRATION 2: CEREALS
9.1. Introduction
This chapter is an empirical illustration of the model applied to the breakfast cereals
product category. The structure of the chapter is similar to that of Chapter 8, starting with a
description of the product category, followed by promotion sales-response models. The
sales-response models are then used in the promotion-profit model. The chapter ends with
a summary of the conclusions from the analysis.
Not only the structure of the chapter is similar to the preceding chapter's; the tools and
concepts are also borrowed from Chapter 8. These will not be explained again, and the
reader is advised to refer to Chapter 8 for more detail.
9.2. Description of the Cereals Category

Figure 9.1 illustrates the brand sales in the breakfast cereals market in the Jest store.
There are seven brands of breakfast cereals available in the test store. Brand 1 dominates
the category with nearly fifty percent of the sales, or three times the sales of the closest
competitor.
The breakfast cereals category is divided into four sub-categories, here labeled Com
Flakes & Co, Crunchy, Muesli, and Natural. Com Flakes & Co is the largest (64 percent)
and the most difficult to describe of the four. In addition to com flakes and frosted com
flakes, it consists of a large number of products similar in their bulkiness and light weight
to cereals such as roasted and sweetened com. The best way to describe this sub-category is
by a negative definition, i.e., everything that does not qualify for any of the other three sub-
categories. The sub-category Crunchy consists of sweetened ready-to-eat mix of grains,
with dried fruits or nuts accounted for 13 percent of cereal sales. The Muesli product
market is here defined as the ready-to-eat mix of grains, sometimes with dried fnlits or
nuts. Its share of sales was 22 percent. The sub-category labeled Natural consists of grains
that can either be used as ingredients when mixing one's own muesli or be eaten pure.
Natural has a market share of approximately 1 percent of the total cereals category. Table
9.1 provides an analysis of cereals' sales broken down by sub-category and brand.
- 213 -
Brand 8
Brand 7
Brand 6
Brand 5
Brand 4
Brand 3
Brand 2
Brand 1
0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50%

Figure 9.1. Brand sales in the cereals category. The bars represent sales valued at regular
price divided into actual sales (AS) and deal cost (DC).
The two largest brands, Brand 1 and Brand 2, focus on the Com Flakes & Co sub-
category. Their combined sales share is above 95 percent in this sub-category. Brand 3 has
a market share exceeding 90 percent in the Crunchy sub-category. Market shares are more
evenly split among participants in the Muesli sub-category, with Brand 5 as the leading
brand with 37 percent of sub-category sales. All brands except Brand 6 clearly focus on one
of the sub-categories, while Brand 6's sales is almost evenly split between Com Flakes &
Co and Muesli.
Brand Corn Rakes&Co CrunchY Muesli Natural TOTAL

Brand 1 46.4% 0.90/0 2.30/0 49.60/0
Brand 2 14.90/0 1.9% 0.20/0 16.9%
Brand 3 11.6% 11 .60/0
Brand 4 6.40/0 6.4%
Brand 5 8.3% 0.00/0 8.3%
Brand 6 2.20/0 2.6% 0.9% 5.8%
Brand 7 0.80/0 0.8%
Brand 8 0.70/0 0.7 %
TOTAL 64.20/0 12.5% 22.2% 1.10/0 100.0%
Table 9.1. Sales of brands and sub-categories.
Table 9.2 provides some summary statistics for the cereals category and the Com Flakes
& Co sub-category.
- 214-
Description Svmbol Value
Average basket value StlBt 103.46
Cereals· share of sales sjtlSt, 0.6%
Corn Rakes & Co's share of Category Sales subsjtlsjt 64.20/0
Cereal to total baskets bjtlBt 2.30/0
Corn Rakes & Co to Cereal Baskets subbjtlbjt 67.5°10
Average purchase amount subsjtlsubbjt 25.72
Deal cost in percent of sales (Corn Rakes & co) subdbjtlsubsjt 5.8°k
Corn Rakes & co share of store sales subsjtlSt 0.4°10
Corn Rakes & co to total baskets sUbbjtlBt 1.5%
Table 9.2. Descriptive measures of the cereal category.
Although the cereals only account for 0.6 percent of the store sales, cereals can be found
in 2.3 percent of the shopping baskets. The corresponding statistics for the Corn Flakes &
Co sub-category to the cereals category are 64.2 percent and 67.5 percent.
9.2.1. Selection of Items for Analysis
Being the most important sub-category in terms of sales, the Corn Flakes & Co sub-
category was selected as the focus of the following analysis. Thus, category hereafter refers
to the Corn Flakes & Co sub-category of the total breakfast cereals category.
Within this sub-category, six items (SKUs, Stock Keeping Units) were chosen for
detailed analysis. The two criteria used for selection of items were that (1) the items should
be important in terms of sales volume and (2) that the retailer promoted the item during the
measurement period. The combined market share of these six items was 61 percent of the
sub-category. The items are presented in Table 9.3.
Relative Market
Item Brand Description Size Dea/Cost Share
Item A Brand 1 Corn Flakes 500g 13.3% 10.20/0
Item B Brand 1 Corn Flakes 750g 6.90/0 13.3%
ItemC Brand 1 Other 375g 6.5% 8.2%
Item D Brand 1 Sweetened Corn Flakes 500g 5.8°10 16.0%
Item E Brand 2 Corn Flakes 375g 11.0% 6.7°10
Item F Brand 2 Roasted and Sweetened Can 500g 5.60/0 6.60/0
Total 60.80/0
Table 9.3. Items selected for the further analysis. Market shares in percent ofthe Corn
Flakes & Co sub-category.
The reason for using single items in the analysis is that sales promotion was not
homogeneous across items within the same brand-variety-size. Aggregating items would
thus mean aggregation of the sales promotion which leads to biased estimates of the sales
response. This is especially important when the sales response is assumed to be non-linear
- 215 -
while aggregation is usually linear, i.e., a sum or a weighted sum (Foekens, Leeflang, and
Wittink 1994).
Table 9.3 shows that each of the items selected for the analysis commands a market
share of more than six percent in the test store. Item D is the best selling article in the
breakfast cereals category. However, although the brand's total sales of normal com flakes
is larger, it is split on more package sizes than its sales of sweetened flakes.
- 216-
9.3. Building the Promotion Response Models
9.3.1. Data Manipulations

As in the coffee case study, sales valued at regular prices of other categories than the
focal category were estimated as actual sales plus five percent.
The regular prices of a number of items in the breakfast cereal category were changed
during the measurement period. As suggested in Chapter 5, regular prices from one week
(week 25, 1992) were used as weights when computing sales volumes of the items and the
category. Deal size, however, was calculated as the deal's fraction of the actual regular
price of the item.
Price deal and special display were almost always used in combination with each other.
Due to the high degree of multicollinearity, special display was omitted from the regression
models. The price deal parameters should consequently be interpreted as the sales response
to a price deal in conlbination with a special display.

It was argued in Chapter 8 that one way to test whether the decomposition of item sales
makes sense is to examine the correlations among the variables. Table 9.4 shows that the
correlation between store traffic and store sales is high. The correlations between store
traffic and sub-category baskets and sub-category sales are also high.
Bt St sUbbjt sUbsjt
Bt 1.00
St 0.87 1.00
subbit 0.49 0.45 1.00
subsit 0.51 0.47 0.96 1.00
Table 9.4. Pearson correlations between store traffic (Bt), store sales (St), number of Corn
Flakes & Co baskets (subbjt), and Corn Flakes & Co sales (subsjt).
Also the correlations between store sales and sub-catgegory baskets and sub-category
sales are high. The correlation between sub-category baskets and sub-category sales is very
high. These correlations indicate that the decomposition model may be useful.
Further evidence is provided by Table 9.5 which shows the correlations between store
traffic and basket value, category-to-store baskets, and category-to-store sales. If store
traffic is an important determiniant of store sales, category baskets, and category sales, then
we would expect the correlations to be lower than those in Table 9.4.
- 217 -
Bt SYSt subbjYBt subsjYSt
Bt 1.00
StlBt 0.64 1.00
subbitlBt 0.10 0.11 1.00
subsjtlSt -0.23 -0.42 0.80 1.00
Table 9.5. Pearson correlations between store traffic (Bt), average basket value (St/Bt),
sub-category-to-store baskets (subbjt/Bt), and subcategory-to-store sales (subsjt/St).
Table 9.5 shows that the correlation between store traffic and basket value (i.e., the ratio
between store sales and store traffic) is 0.64 which is lower than the correlation between
store traffic and store sales. This supports the model. It is not a problem that the correlation
between store traffic and basket value is still fairly high. The high correlation indicates that
store traffic and average basket value may be driven by a common underlying factor, such
as the day-of-the-week.
The correlations between store traffic and category-to-store baskets and category-to-
store sales also support the model. These correlations are much lower than the correlations
between store traffic and category baskets, and between store traffic and category sales.
In summary, the correlations presented in this section clearly indicate that the
decomposition model may be useful. Further evidence will be provided by the response-
model estimation.
9.3.3 Store-Level Response Models

The exponential model structure was used for the three models estimated at the store-
level. The regression equations were described in equation (8.1) in Chapter 8. The
variables are defined in the same way as above.
The maximum model included price deal variables for all six items and six dummy
variables indicating the day of the week. After estimating the maximum models, variables
not significant at the 20 percent level were dropped before re-estimating the model.
Regression coefficients were examined for significance at the 10 percent level and the new
models were estimated. This was iterated until all coefficients were significant at the 10
percent level. As explained in the coffee case study, regression coefficients having "wrong
signs" were treated as insignificant and consequently dropped from the model. The general
null hypothesis was that the parameters were equal to zero. In the store traffic and store
sales models, the null hypothesis for the price deal paranleters wase that the parameters
were non-positive. No directed hypotheses were specified in the basket-size model. The
results of the model estimation are presented in Table 9.6.
- 218 -
Store Traffic Store Sales Basket Size
coefficient s. e. coefficient s. e. coefficient s. e.
Constant 7.92 0.01 12.43 0.02 4.47 0.01
Price Deal
Item A
Item B
Item C 0.62 0.14 0.84 0.21
Item D
Item E
Item F
Day-of-the-Week
Tuesday 0.06 0.02
Wednesday 0.10 0.02
Thursday 0.07 0.03 0.24 0.04 0.22 0.02
Friday 0.13 0.03 0.49 0.04 0.41 0.02
Saturday 0.07 0.03 0.28 0.04 0.26 0.02
Sunday -0.11 0.03 -0.16 0.04
adj. Rsq. 330/0 59% 710/0
n 160 160 160
Table 9.6. Store-level response to sales promotion in the cereal category.
The only sales promotion found to have a positive effect on store traffic and store sales
was Item C's price deal. The regression results indicate some variation over the days of the
week. The most important variation seems to be in the average basket size. The explained
variance in this dependent variable was 71 percent. Store traffic, on the other hand, showed
the smallest variation over the week, with only 33 percent of its variance explained.
9.3.4 Category-Level Response Models
We shall now examine how the Corn Flakes & Co category responds to sales
promotion. Before building regression models, the dependent variables will be examined
graphically. The development over time of the ratio of category-to-store baskets and the
category-to-store sales ratio are illustrated in Figure 9.2.
It can be seen that the two ratios follow each other closely over the measurement period.
The correlation between the two measures is 0.80 whether the six exceptional observations
are included or not.
The two ratios were modeled using logistic regression as previously described in
Chapter 8 (see equation 8.2). First, the maximum models, including all variables appearing
in Table 9.7 were estimated. Variables whose effect on the dependent variable was
insignificant at the 20 percent level were discarded and the new, restricted, regression
equation was re-estimated. Variables were then eliminated until all remaining parameters
were significant at the ten percent level. The formal null hypotheses associated with the
price deal variables were that sales promotion would have no positive effect on the
- 219-
category-to-store ratios. The hypotheses pertaining to the day-of-the-week dummy
variables were that they would have no effect on the ratios.
It was found that the sales ratio varied more over the days of the week than did the
basket ratIo. The unexpected finding in this result was that we know from the store-level
response models that the average basket size varies over the week and it would be logical
that the probability of any category entering the shopping basket would be positively
related to the basket size. According to the regression results, the cereals category accounts
for a larger proportion of store sales on Mondays than on other days.
Table 9.7 also shows that sales promotion for five of the six items studied here had a
positive impact on the category-to-store sales ratio. The category-to-store traffic ratio was
positively affected by sales promotion for four of the items.
The models explained approximately half of the variance in the (transformed) dependent
variables, with 48 percent of the variance in the basket ratio and 52 percent of the variance
in the sales ratio explained.
- 220-
0.04 0.009
0.035 0.008
0.007
0.03
0.006
0.025
o
ia:: 0.005
- - Basket Ratio
i 0.02
--Sales Ratio
=
m
0.004
0.015
0.003
0.01
0.002
0.005 0.001
Figure 9.2. Category development over time.
Category/Store Category/Store
Baskets Sales
coefficient s.e. coefficient s.e.
Constant -4.21 0.02 -5.44 0.05
Price Deal
Item A 1.14 0.19 1.38 0.21
Item B 0.69 0.30
Item C 0.96 0.27 0.96 0.28
Item D 0.61 0.29 0.69 0.30
Item E 0.94 0.25 0.47 0.26
Item F
Day-of-the-Week
Tuesday -0.09 0.05 -0.19 0.06
Wednesday -0.13 0.06
Thursday -0.28 0.06
Friday -0.44 0.06
Saturday -0.27 0.05 -0.53 0.06
Sunday -0.25 0.05 -0.26 0.06
adj. Rsq. 480/0 52%
n 160 160
Table 9.7. Category-level response to sales promotion in the cereal category.
- 221 -
9.3.5 Item-Level Response Models
At the item level, three models were built for each item. Item-to-category baskets and
item-to-category sales were both modeled using the logistic regression model also used at
the category level (see equation 8.3). For modeling the average item purchase among
shoppers buying the item, the exponential model was used (equation 8.4).
The maximum models included all variables found in Table 9.8 and were identical
across all six items. Final models were found using the procedure described above, with the
only difference the fomlal null hypotheses for sales promotion for items other than the
focal item; their promotion would have a positive or zero effect on the ratios.
Regression results will be presented and discussed for one item at the time, starting with
Item A. Table 9.8 shows that Item A's basket and sales ratios did not vary due to the day of
the week. The day-of-the-week had only small effects on the purchase amount of Item A
among shoppers buying Item A on Saturdays and Sundays.
Item A's price deal had a fairly strong impact on the item's basket ratio and sales ratio.
The impact on average purchase amount was small. Two of the other items' sales
promotion, Item B and Item D, affected Item A's basket and sales ratios. None of these had
any significant impact on the average purchase amount. Item E's sales promotion had a
small but significant positive impact on the average purchase amount.
While the purchase amount model only explained 22 percent of the variance, the ratio
models explained 53 percent and 55 percent of the variance in the basket ratio and the sales
ratio, respectively.
The regression models for Item B are less successful than those for Item A in terms of
explained variance. Only 4 percent of the variance in average purchase size could be
explained, while the basket ratio and sales ratio models each explained 36 percent of the
variance in their dependent variables. The final regression models for Item B are presented
in Table 9.9.
The response parameters for Item B's price deal are alnlost identical in the basket-ratio
and the sales-ratio models. Their magnitude indicates that the price deal has a strong
impact on these ratios. Item B's sales promotion had no significant impact on the average
purchase amount of Item B. In fact, the only variable found to have a significant impact on
the average purchase amount of Item B was the dummy variable for Thursday.
Item E was the only other item whose sales promotion had an inlpact on the ratios.
Neither the basket ratio nor the sales ratio was affected by the day-of-the-week dummy
variables. Thus, there was no evidence that these ratios would vary over the days of the
week.
- 222-
Constant -2.55 0.07 -2.60 0.07 3.18 0.00
Price Deal
Item A 5.93 0.55 6.13 0.55 0.14 0.04
Item B -1.91 0.92 -2.14 0.92
Item C
Item D -2.77 0.83 -3.10 0.83
Item E 0.12 0.05
Item F
Day-of-the-Week
Tuesday
Wednesday
Thursday
Friday
Saturday 0.03 0.01
Sunday 0.03 0.01
adj.Rsq. 550/0
n 149
Table 9.8. Item A: Item-level response to sales promotion in the cereal category.

coefficient s. e. coefficient s. e. coefficient s. e.
Constant -2.65 0.08 -2.34 0.08 3.52 0.01
Price Deal
Item A
Item B 8.47 1.09 8.66 1.09
Item C
Item D
Item E -3.29 0.92 -2.96 0.91
Item F
Day-of-the-Week
Tuesday
Wednesday
Thursday 0.04 0.02
Friday
Saturday
Sunday
adj.Rsq. 360/0 360/0
n 145 145
Table 9.9. Item B: Item-level response to sales promotion in the cereal category.
- 223-
Table 9.10 presents the regression results for Item C. Again, none of the day-of-the-
week dummy variables had any significant impact on the ratios and only one of them had a
small impact on the average purchase amount of the item.
Item CIS own promotion affected the two ratios approximately equally, with a somewhat
larger effect on the sales ratio. The promotion also had a small significant impact on the
average purchase amount.
Other items whose sales promotion affected Item CIS basket and sales ratios were Item
A and Item D. Item D's price deal had a significant impact on the average purchase amount
of Item C as well.
In comparison with the models for Item A and Item B, the models for Item C performed
worse in terms of variance explained in the ratio models. The models of Item CIS basket
and sales ratios accounted for 29 percent each.
The model of the average purchase amount of Item C among buyers of the item
explained 15 percent of the variance. Again, the average purchase-amount model was the
model explaining the smallest part of the variance in its dependent variable.
This pattern was repeated for Item D, whose average purchase amount model accounted
for 9 percent of the variance while the basket-ratio model accounted for 43 percent and the
sales-ratio model for 42 percent of the variance in their respective dependent variables (see
Table 9.11).
The regression results for Item D, presented in Table 9.11, show that the day of the
week had little effect in any of the three models. Only the parameter for the Thursday
dummy variable was significant in the two ratio models.
Item D's price deal had a strong impact on the basket and sales ratios and a small
positive impact on the average purchase amount of the iten1. Item D's sales ratio was
negatively influenced by price deals for Item A, Item B, and Item C. In addition to these
three items' price deals, the basket ratio was also negatively influenced by Item E's price
deals.
- 224-
coefficient s.e. coefficient s.e. coefficient s. e.
Constant -2.63 0.07 -2.54 0.06 3.32 0.01
Price Deal
Item A -2.49 0.58 -2.39 0.57
Item B
Item C 4.84 0.85 4.96 0.84 0.13 0.07
Item D -1.89 0.91 -1.91 0.90 0.20 0.07
Item E
Item F
Day-of-the-Week
Tuesday
Wednesday
Thursday
Friday
Saturday 0.03 0.01
Sunday
adj.Rsq. 290/0 150/0
n 144 144
Table 9.10. Item C: Item-level response to sales promotion in the cereal category.

coefficient s.e. coefficient s.e. coefficient s. e.
Constant -1.78 0.07 -1.69 0.07 3.32 0.00
Price Deal
Item A -1.80 0.55 -2.19 0.50
Item B -1.90 0.81 -2.22 0.82
Item C -2.90 0.74 -3.04 0.75
Item D 6.88 0.79 7.02 0.80 0.17 0.04
Item E -1.28 0.66
Item F
Day-of-the-Week
Tuesday
Wednesday
Thursday 0.25 0.13 0.22 0.13
Friday
Saturday
Sunday
adj.Rsq. 430/0 420/0 90/0
n 148 148 148
Table 9.11. Item D: Item-level response to sales promotion in the cereal category.
- 225-
The regression results for Item E are presented in Table 9.12. Item E's basket-ratio and
sales-ratio models are the ratio models explaining the smallest share of variance of the six
items' models. The basket-share model accounts for 23 percent of the variance while the
share of explained variance in the sales share model is 27 percent. The average purchase
item purchase model is again the model explaining the smallest share of the variance with
12 percent. This is, however, better than the model for Item C.
The Friday dummy variable was the only day-of-the-week dummy variable having
significant impact in any of the three models for Item E. The sales ratio was on average
somewhat higher on Fridays than the rest of the week.
Item E's price deal had significant positive impact in all three models. Its impact on the
ratios was fairly strong. Item A's and Item D's price deals had significant impact on Item
E's sales ratio. Item A's sales price deal also had a significant negative impact on Item E's
basket ratio as well as on the average item purchase of Item E. The average item purchase
was positively affected by Item F's price deal.
The average item purchase of Item F, on the other hand, was not affected by any items'
price deal, as shown in Table 9.13, presenting the regression results for Item F. The only
variable having a significant impact on average item purchase of Item F was the Thursday
dummy variable. The model accounts for only 2 percent of the variance, making it the
worst performing model in the cereal category. The ratio models explain 45 percent of the
variance.
None of the day-of-the-week dummy variables had any significant impact on the basket
ratio or the sales ratio.
Item F's price deal had very strong impact on the basket and sales ratios. These ratios
were negatively affected by sales promotion for Item A and Item D. The sales ratio was
also affected by Item C's price deals.
- 226-
Constant -2.45 0.03 -2.99 0.08 2.75 0.01
Price Deal
Item A -1.88 0.72 -2.12 0.73 -0.24 0.11
Item B
Item C
Item D -1.81 1.03
Item E 6.07 0.91 6.50 0.91 0.45 0.10
Item F 0.33 0.16
Day-of-the-Week
Tuesday
Wednesday
Thursday
Friday 0.29 0.17
Saturday
Sunday
adj.Rsq. 120/0
n 144
Table 9.12. Item E: Item-level response to sales promotion in the cereal category.

Constant -2.87 0.06 -2.71 0.06 3.35 0.01
Price Deal
Item A -3.83 0.70 -3.74 0.70
Item B
Item C -1.35 0.76
Item D -1.94 0.76 -1.63 0.85
Item E
Item F 10.23 1.07 9.72 1.08
Day-of-the-Week
Tuesday
Wednesday
Thursday 0.04 0.02
Friday
Saturday
Sunday
adj.Rsq. 20/0
n 137
Table 9.13. Item F: Item-level response to sales promotion in the cereal category.
- 227-
This section summarizes the regression results in terms of how successful the models
are in explaining the variance and the number and type of variables that were retained in
the final models.
The explanatory power of a regression model can be expressed as the share of the
variance of the dependent variable the model explains. Table 9.14 presents the amount of
variance explained by the 23 regression models.
Baskets Sales Quantitv

Store Level 330/0 59% 71 %
Category Level 480/0 52% n.a.
Item Level Average 38% 39% 11 %
Item A 530/0 550/0 22%
Item B 36% 36% 4%
Item C 290/0 29% 15%
Item D 430/0 42% 90/0
Item E 23% 270/0 12%
Item F 450/0 45% 20/0
We recognize that the store-level average purchase quantity (or purchase amount) model
is the model with the highest adjusted R-square, i.e., it is the most successful model in
terms of variance explained. The store-traffic model explains less variance than the store
sales model.
The two category-level models explain about fifty percent of the variance. The sales
ratio model is slightly better than the basket-ratio model.
The item-level models show a consistent pattern in that the purchase-quantity models
perform considerably worse than the basket and sales ratio models. On average, the
purchase-quantity models explain 11 percent of the variance. The average for the basket
ratio models is 38 percent and the average for the sales-ratio models is 39 percent.
The results suggest that the models concerned with the number of shoppers perform
approximately equally well at all three hierarchical levels. The sales-related models
perform better at the store- and category-levels than at the item level. The purchase-
quantity models, on the other hand, perform badly at the item level while performing very
well at the store level.
Table 9.15 presents the number of significant model parameters of the regression
models. Only one sales promotion parameter was significant in the store-level models of
store traffic and store sales. The average basket value was not affected by sales promotion
in the cereals category. Sales promotion thus had little impact on the store-level measures.
- 228-
The day-of-the-week dummy variables, on the other hand, were important for explaining
variance of the store-level measures.
Sales Promotion Day-of-the-Week

Baskets Sales Quantity Baskets Sales Quantity
Store Level 1 1 0 445
Category Level 4 5 n.a. 4 6 n.a.
Item A 332 002
Item B 220 001
Item C 332 001
Item D 541 1 1 0
Item E 233 010
Item F 341 001
Table 9.15. Number of significant model parameters.
The opposite pattern is observed for the item-level models. Few day-of-the-week
dummy variables were significant in the item-level models. On average, the item-level
models contain less than one of the day-of-the-week dummy variables. At the same time,
three sales promotion variables were retained in the item-level models of the basket ratio
and the sales ratio. The average number of sales promotion variables in the item-level
purchase-quantity model was 1.5.
It can be concluded that these observations are encouraging and strengthen our
confidence in the modeling framework.

In this section we investigate the promotion effects on sales of the focal item, other
items in the focal category, category sales, and sales of other categories than the focal one,
as well as store sales. Recall that the item- and category-level response models developed
in Section 9.3 model relative-responses, Le., they model changes in ratios rather than
changes in absolute levels.
Nevertheless, profits do not come from relative sales but from absolute sales. In order to
get absolute item sales, we need to integrate the results of the relative models. This is also
necessary in order to get absolute category sales and sales of other items in the focal
category. Table 9.16 presents the model parameters for each item.
One item at a time is selected as the focal item. The model parameters found in Table
9.16 are inserted into the relative-sales response models. The size of the deal discount is
increased in steps of 2 percentage points from 0 to 30 percent of the regular price. For each
deal discount, the sales responses are evaluated.
- 229-
Store Traffic Basket Value Category Item
Focal Item Constant Deal Constant Deal Constant Deal Constant Deal
Item A 7.92 0.00 4.47 0.00 -5.44 1.38 -2.60 6.13
Item B 7.92 0.00 4.47 0.00 -5.44 0.69 -2.34 8.66
Item C 7.92 0.62 4.47 0.00 -5.44 0.96 -2.54 4.96
Item D 7.92 0.00 4.47 0.00 -5.44 0.69 -1.69 7.02
Item E 7.92 0.00 4.47 0.00 -5.44 0.47 -2.99 6.50
Item F 7.92 0.00 4.47 0.00 -5.44 0.00 -2.71 9.72
Table 9.16. Summary ofmodel parameters.
In addition to the sales-response model parameters presented above, the following

assumptions were used when evaluating promotion profits. The profit margin of the focal
item, other items in the category, and other categories were assumed to be 20 percent of the
regular prices. The manufacturer was assumed to offer the retailer a 20 percent trade deal,
measured in percent of the regular retail price of the focal item. There was no requirement
from the manufacturer that the retailer should pass anything of the trade deal through to the
consumers. The retailer received the trade deal for merchandise sold only.
9.4.1. Item A
We shall now investigate how a price deal for Item A affects sales at the three
hierarchical levels. Figure 9.3 shows that sales of Item A increase as a result of the price
deal. The price deal for Item A also causes sales of other items in the focal category to
increase. This means that category sales increase more than the sales of the focal item.
Thus, Item A's price deal causes no intra-category cannibalization. All incremental sales of
Item A are also incremental sales to the category. In contrast, the inter-category
cannibalization is complete. All incremental category sales come from other categories in
the store. Store sales are not affected by the price deal for Item A.
The promotion profits seen from the item-, category-, and store-perspective of different
deal discounts are presented in Figure 9.4. The item-level promotion profits rise with the
size of the deal discount and peak at a deal discount of 24 percent. The category-level
promotion profits are larger than the item-level promotion profits for all deal discounts
investigated. For deal discounts between 0 and 24 percent, the category-level promotion
profits rise with the size of the deal discount. Thereafter, larger deal discounts decrease the
category-level promotion profits.
The store-level promotion profits are close to zero for deal discounts smaller than 10
percent. Larger deal discounts lead to negative promotion profits, with the size of the loss
rapidly increasing with the size of the deal discount.
- 230-
600
III III
III
400
III'" III
III III III
III III III •

III III
200 III III
en
G)
---0- Focal Item
'ii
en ---tr- Other Items
.5
G) 6 • • • Category
C)
c
as - x - Other Categories
J:
(J
- - - Store Sales
·200
-400
-600
Deal Discount
Figure 9.3. Sales changes caused by Item A's price deal.
60.00
.. III ..
.......... . .. . ..
III •
40.00
20.00
~
a. 0.00 ---0- Focal Item
'ii
c
o • • • Category
:g ---Store
E -20.00
o
D:
-40.00
-60.00
-ao.oo
Deal Discount
Figure 9.4. Promotion profits ofItem A's price deal.
- 231 -
9.4.2. Item B
The sales changes resulting fronl a price deal for Item B are presented in Figure 9.5.
Item B's price deal has stronger impact on sales of Item B than on the category sales,
although category sales also increase due to Item Bls price deal. There is some, but not
complete, intra-category cannibalization. Inter-category cannibalization, on the other hand,
is complete. In other words, the incremental category sales come from other categories.
Store-sales are therefore not affected by Item B's price deal.
The changes in sales of Item B, sales of other items in the category, and other categories'
sales result in the promotion profits depicted in Figure 9.6.
The item-level promotion profits are higher than the category-level promotion profits,
and the category-level promotion profits are higher than the store-level profits for all deal
discounts. Small deal discounts lead to small promotion profits irrespective of the level of
analysis. Large deal discounts, on the other hand, appear fairly profitable at the item-level,
but are negative when measured at the category- or store-level.
- 232-
800
600
400
en
Q)
~Focalltem
m
......... .. .. . . ......
U) ~ Other Items
.5
Q) 200 .. .. • Category
en
c
~ - x - Other Categories
.c
0
- - - Store Sales
o-
-200
-400
Deal Discount
Figure 9.5. Sales changes caused by Item B's price deal.
60.00
40.00
20.00
.. .. .. . . ............... . . .. . ..
~
e
~
0.00
--0----- Focal Item
m
c
0 -20.00 .... • Category
:g ---Store
E
e
~
-40.00
-60.00
-80.00
-100.00
Deal Discount
Figure 9.6. Promotion profits ofItem B's price deal.
- 233-
9.4.3. Item C
Figure 9.7 presents changes in sales for Item CIS price deals. The effect on sales of other
categories and on store sales are very large compared to the effect on sales of the focal item
and the- focal category. Separate scales were therefore applied to make the item and
category effects visible in Figure 9.7.
Item CIS price deal has a strong impact on store sales by increasing sales of other
categories considerably. Promotion for Item C causes neither intra- nor inter-category
cannibalization, leading to the promotion profits presented by Figure 9.8.
As with the store-sales response, the store-level promotion profits are of a greater order
of magnitude than the item- and category-level profits. Again, separate scales were used.
The store-level profits are higher than the category-level profits for all sizes of the deal
discount. All three levels of analysis report positive promotion profits. While item- and
category-level promotion profits peak at deal discounts smaller than 30 percent (the
maximum deal discount investigated here), the store-level promotion profits rise with the
deal size and show no sign of leveling out. The implication is that the retailer could make
larger profits by very large deal discounts for Item C. Some caution is called for, however.
The store traffic and store-sales responses to Item CIS price deal are too large to be realistic,
and may be statistical artifacts. A retailer could test this by running sonle price deals of, for
example, 20 percent deal discount and re-estimating the response models.
- 234-
700 50000
45000
600
40000
500 35000 fI)

.!!
as Focal Item
.m
c'l 400
30000 ! ---0--
----tr- Other Items

.5 S
!, 25000 ~ • • • Category
Ii
~
300 - x - Other Categories
.c 20000
o as
.c - - - Store Sales
o
200 15000
10000
100
5000
Deal Discount
Figure 9.7. Sales changes caused by Item C's price deal.
70.00 10000.00
9000.00
60.00
8000.00
iE
50.00 7000.00 e
D.
~ 1;
a 6000.00
c
0
Q: 40.00 :g Focal Item
..
--0-
1;
c E
o 5000.00 e • Category
~
D.
30.00 'i ---Store
e
D.
4000.00 >
~
3000.00 !0
20.00
en
2000.00
10.00
1000.00
0.00 0.00
~ ~ ~ g; ~ ~ § ~ ~ ~ ~ ~
?fl. ?fl. ?fl. ?fl.
0
~ ~ ~
Deal Discount
Figure 9.8. Promotion profits ofItem C's price deal.
- 235-
9.4.4. Item D
Sales responses for Item D's price deals are depicted by Figure 9.9. Store sales are
unaffected by Item D's price deal while the promotion does affect category sales, implying
that the .inter-category cannibalization is complete. Cannibalization within the category is
moderate; between one third and half of the incremental sales of the focal item represent
incremental category sales.
The promotion profits due to deal discounts on Item D are illustrated in Figure 9.10.
Small deal discounts have almost no effect on the store-level promotion profit, while lager
discounts are unprofitable.
The category-level promotion profits are positive (but small) for deal discounts up to 20
percent. Larger deal discounts are unprofitable at the category level. The promotion profits
measured at the item-level are positive for all price deals examined, with a peak at deal
discounts of 20 percent.
- 236-
700
600
500
400
300 ---0- Focal Item

I
...... . . . ....
'ii
en 200
••• I
----tr-- Other Items
.5
Q)
tn
c 100
ca
...... • • • Category
.c
0
o- - - - Store Sales
-100
-200
-300
-400
Deal Discount
Figure 9.9. Sales changes caused by Item D's price deal.
40.00
.. .. . . ...........
20.00
.. .. lit
0.00
~
e
a. -20.00 ----D---- Focal Item
'ii
c
0 • • • Category
:g
-40.00 ---Store
E
0
~
-60.00
-80.00
-100.00
Deal Discount
Figure 9.10. Promotion profits ofItem D's price deal.
- 237-
9.4.5. Item E
The inter-category cannibalization is complete also for price deals for Item E (Figure
9.11). Item E's price deal has a large impact on Item E's sales and some impact on category
sales as well, resulting in the profit impact shown in Figure 9.12.
The store-level promotion profits are close to zero when the deal discount for Item E is
below 10 percent. Larger deal discounts imply negative store-level promotion profits. Deal
discounts smaller than 26 percent generate small profits at the category level. At the item
level, all deal discounts examined appear profitable.
- 238-
300
250
200
.....
.... .. .. ......
150
VI
Q)
16 100
en .... .... .. . .. --0--
------t:r----
Focal Item
Other Items
.5
Q) 50 - - • Category
C)
c
ca - x - Other Categories
.c o-
0
- - - Store Sales
-50
-100
-150
-200
Deal Discount
Figure 9.11. Sales changes caused by Item E's price deal.
20.00
10.00
. .- .. ......
--1--.. .
0.00 -0-.-.'""'"1--... _ .;;;;;;;;::=:+----+---1--4---1---+--1--+-.-..--+-----1
~
'0
a: -10.00 ----D-- Focal Item
16
a
:g
• - • Category
E -20.00 ---Store
o
a:
-30.00
-40.00
-50.00
Deal Discount
Figure 9.12. Promotion profits ofItem E's price deal.
- 239-
9.4.6. Item F
Item Fls price deals lead to complete intra-category cannibalization. All incremental
sales of the focal item come from other items in the category. Category sales are therefore
not affected by price deals for Item F. The price deal for Item F has no effect on store sales
either.
The promotion profits are presented in Figure 9.14.
The category- and the store-level promotion profits coincide for Item Fls price deal,
because the price deal has no effect on sales of other categories. The store-level promotion
profits of small deal discounts are close to zero because the item-level promotion profits
outweigh the reduced gross profits of other items in the focal category. Larger deal
discounts are profitable only at the item level.
- 240-
600
400
200
---0-- Focal Item
J
.5
-fr-- Other Items
8. 0 _.*-II@!~-*---«--:~->.O'-~-*-->.o-~:~--_,*"O:<-X-,",*:'_*-*-*-* • • • Category
IU ~ ~ ~ ~ ~ ~ - x - Other Categories
- - Store Sales
·200
·400
·600
Deal Discount
Figure 9.13. Sales changes caused by Item F's price deal.
60.00
40.00
20.00
--0---- Focal Item
• - • Category
--Store
-80.00
Deal Discount
Figure 9.14. Promotion profits ofItem F's price deal. Note that the category- and store-
level promotion profits coincide.
- 241 -
9.4.7. Summing Up
To summarize, we have studied the impact of price deals on sales for six items. In all six
cases, the price deal had a positive impact on sales of the promoted item. In one of the
cases (Item F), intra-category cannibalization due to the price deal was complete. Moderate
intra-category cannibalization was observed in three cases (B, D, and E), while the
remaining two items (A and C) did not cannibalize on sales of other items in the category.
Thus, sales promotion had a positive impact on category sales for five out of six items.
Only one of the items' price deal (C), had a positive impact on store sales. Inter-category
cannibalization was in that case non-existent while the remaining five cases showed
complete inter-category cannibalization.
The item-level profits were larger than the store-level profits for five out of six items
(for all deal discounts). For one item (C), the store-level promotion profits were larger than
the item-level profits, irrespective of the size of the deal discount. The category-level
profits were larger than the store-level profits for four items (A, B, D, E) and equal to the
store-level profits for one item (F). In one case, Item C, the store-level promotion profits
were larger than the category-level profits. The category-level profits exceeded the item-
level profits for two items (A and C). In one case (Item E) the category-level profits were
equal to the item-level profits for deals smaller than 10 percent. For deals larger than 10
percent, the item-level promotion profits were larger than the category-level profits for
Item E. For three items (B, D, and F), the item-level promotion profits were larger than the
category-level profits for all deal discounts.
- 242-
In this section we examine how the deal discount set by the profit-maximizing retailer
depends on the level of analysis, i.e., whether the retailer maximizes item-, category-, or
store-level promotion profits. The store-level promotion profits are regarded the as most
relevant function for the retailer (see Chapter 5). Thus, deals determined to maximize item-
or category-level promotion profits are regarded as sub-optimal. The cost of such sub-
optimization is then the difference in store-level promotion profitsof the optimal and the
sub-optimal deal. This cost is important for the retailer as it indicates the value of analyzing
the promotion profits at the store level.
For each level of analysis, the optimal deal was found by evaluating, for each full
percentage point between zero and 100, the promotion profits resulting from a deal of that
size. These are called Deal], Deal2, and Deal3, depending on whether the item-, category-,
or store-level promotion profits were used. The promotion profits of Deal3 are called
Profit3, while the sub-optimization cost of setting Deal1 or Deal2 are called Loss1 and
Loss2. Table 9.17 presents the results for the six items analyzed.
Item Dea/Reg Dea/1 Dea/2 Dea/3 Loss1 Loss2 Profit3

Item A 0% 23% 230/0 60/0 -25.27 -25.27 0.94
Item B 0°;'0 23% 140/0 7°;'0 -36.37 -4.48 3.01
Item C 00/0 22% 27% 1000/0 -30595.38 -28880.57 37600.61
Item D 0% 190/0 11 % 4°;'0 -28.67 -4.77 1.18
Item E 0% 230/0 160/0 5% -16.73 -4.80 0.43
Item F 0% 23°;'0 8% 8% -27.72 0.00 2.81
Item A 10/0 23% 23% 60/0 -25.27 -25.27 15.30

Item B 1°;'0 230/0 14% 7% -36.37 -4.48 21.27
Item C 10/0 22°;'0 27% 1000/0 -30595.38 -28880.57 37615.79
Item D 1% 19°;'0 110/0 4% -28.67 -4.77 33.54
Item E 10/0 23% 160/0 5% -16.73 -4.80 10.38
Item F 10/0 23% 8°;'0 80/0 -27.72 0.00 15.76
Item A 100/0 23% 23% 10% -24.33 -24.33 14.37

Item B 10 % 230/0 14% 10% -35.83 -3.95 20.73
Item C 100/0 22% 27°;'0 100% -30595.38 -28880.57 37615.79
Item D 10 % 19% 11% 100/0 -25.31 -1.41 30.18
Item E 10% 230/0 160/0 10% -15.92 -3.99 9.56
Item F 100/0 230/0 10% 100/0 -27.49 0.00 15.53
Table 9.17. Deal sizes set using item-level (DeaI1), category-level (DeaI2), and store-level
(DeaI3) analysis. Profit3 is the profits ofthe optimal deal evaluated at the store-level.
Loss] and Loss2 are losses due to sub-optimization by setting the deal size to Deal1 or
Deal2.
- 243-
We first analyze the results presented in the upper section of Table 9.17, Le., when there
is no deal requirement imposed by the manufacturer.
In five out of six cases, item-level deal optimization leads to larger deal discounts than
what category- or store-level deal optimization would suggest. In the sixth case (Item C),
category-level deal optimization suggests a somewhat larger deal discount than the item-
level analysis. Store-level analysis suggests that the retailer should give the customers the
item for free. The retailer should however not take this suggestion too literally for two
reasons. First, the parameter estimates of the response model may exaggerate the store sales
response. Second, since the suggested deal discount of 100 percent is far larger than the
largest observed deal discount for Item C, it falls outside the scope of the model. It cannot
be regarded as a valid estimate of the optimal size of the deal discount. It may, however, be
interpreted as an indication that large deal discounts may be profitable.
Disregarding the results for Item C, the maximum store-level profits are small, with a
n1aximum of SEK 3.01 per day (Item B). Again, it should be noted that these promotion
profits are incremental profits to the store compared to the profits reached by a deal of size
zero, where the retailer is assumed to absorb all of the trade deal offered. If the
n1anufacturer required the retailer to pass through at least a part of the trade deal, the
promotion profits would also include the trade deal money that the retailer would pocket
without any deal discount, had the manufacturer not required any deal. A deal requirement
of one percent would not change the optimal deal discounts but would increase the
promotion profit, as can be seen in the middle section of Table 9.17. Note that the
opportunity losses due to myopic decision-making remain the same as with a zero deal
requirement.
The lower section of Table 9.17 shows the effects of a deal requirement of 10 percent.
Such a deal requirement has no effect on the deal size determined using item-level
promotion profits as the function to maximize since Deal] is greater than 10 percent for all
six items. For one out of six items, the deal requiren1ent increased Deal2, the optimal deal
discount when the category-level promotion profits are maximized. The deal discount
chosen by retailers maximizing store-level promotion profits increased for five out of six
items. For each of these five items, the optimal deal discount equals the deal requirement.
A comparison between the store-level promotion profits (Profit3) with the one percent
deal requirement and the profitswith the 10 percent deal requirement shows that the
retailer's promotion profits are hurt by the higher deal requirement. The loss due to myopic
optimization of the deal size decreases as a consequence of the higher deal requirement.
- 244-
9.5.2. Trade Deal and Margin
This section examines how the size of the trade deal affects the optimal deal discount.
Only results from store-level analyses are presented because these, and not the item-level
or category-level analyses, represent optimal deal sizes seen from a retailer perspective.
In order to investigate how sensitive the pass-through of trade promotions is to the profit
margin and of the cereals category, the optimal deal size was determined for four levels of
the profit margin. The 20 percent profit margin is regarded as the base case and has been
marked by a thick line in the figures.
Figure 9.15 presents results for Item A. At a profit margin of 20 percent, the retailer will
optimally absorb all trade money for trade deals below 15 percent (of retail price). Larger
trade deals will be partially passed through. Examination of Figure 9.15 reveals that the
retailer should pass through all of the trade money exceeding 15 percent of the retail price.
If the profit margins were 10 percent, the retailer would find it advantageous to absorb all
of the trade deal.
At gross margins higher than 20 percent, the retailer would pass through a larger share
of the trade deal to the consumers. It can be seen that the relative pass-through is more
dependent on the profit margin for small trade deals than for large trade deals.
The pass-through of large trade deals is found to be less sensitive to profit margins than
the pass-through of small trade deals Item B. Figure 9.16 shows that a trade deal of 40
percent of retail price of Item B leads to deal discounts between 21 and 26 percent,
depending on the profit margin. The corresponding range for a trade deal of 25 percent is 7
to 17 percent. The range for trade deals of five percent is again smaller, because a 5 percent
trade deal only induces a deal discount if the retailer's profit margin is 30 percent or higher.
The results for Iten1 C, presented in Figure 9.17, are fairly uninteresting because the
model suggests that the deal discount should always be 100 percent. Deal discounts of 100
percent are not very realistic and, as noted above, fall outside the range of the model.
The relative pass-through for Item D (Figure 9.18) ranges between zero, for trade deals
lower than 15 percent, and close to 50 percent for trade deals of 40 percent in the base case.
The relative pass-through increases with the size of the trade deal for all profit margins
examined. The pass-through and the relative pass-through are both more sensitive to the
profit margin for small trade deals than for large trade deals.
For Item E, the pass-through and the relative pass-through are very sensitive to the
retailer's profit margin (see Figure 9.19).
The pass-through is also more sensitive to the size of the trade deal when the retailer's
profit margins are thin than when the margins are fat. In the base case, i.e., when the profit
margins are 20 percent, the retailer will pocket all trade money for trade deals smaller than
- 245-
15 percent. The optimal deal discount then rises with about one percentage point for each
percentage point increase of the trade deal.
The results for Item F differs from the previous results in that the deal discount is totally
insensitive to the profit margin (Figure 9.20). This is because price deals for Item F leads to
complete intra-category cannibalization and the profit margin of all items (including Item
F) are assumed to be equal.
The retailer will not initiate any price deal if the trade deal offered is lower than 11
percent. Above 11 percent, the deal discount is an approximately linear function of the size
of the trade deal discount. The relative pass-through is consequently higher the larger the
trade deal.
In summary, it has been found that the size of the trade deal is important for the size of
the optimal deal discount. The larger the trade deal, the larger the deal discount in the store.
Small trade deals often fail to induce the retailer to deal the item. In such cases, the relative
pass-through of the trade deal to the consumers is larger if the trade deal is larger.
Generally, the size of the deal discount is also dependent on the profit margin of the
category, with higher profit margins leading to larger deal discounts. The sensitivity of the
deal discount is higher for small than for large trade deals.
- 246-
~ ~
rJQ0 rJQ.
;:: Optimal Retailer Deal Discounl ;:: Optimal Retailer Deal Discounl
~ 0 C1I o C1I
I\)
o
I\)
C1I
Co)
0 ~ 0 C1I
....
o
....
C1I
I\)
0
I\)
C1I
Co)
o ~
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
~ ~
~ 0% ~ 0%
~ 1% ~ 1%
2% 2%
~ 3% ~ 3%
~ ~
~ 4%
~ 4%
5%
5%
~ 6% ~ 6%
7% 7%
~ 8%
~ 8%
9%
§o
~ 9%
10%
~
§. 10%
~ 11% ~ 11%
~ ~
12% 12%
~ 13% ~ 13%
~ 14% ~ 14%
~ ~
15% 15%
~ ~
§o -f 16%
i1 17%
N·
~
-f 16%
i1 17%
i i
~
18% 18%
~
Ic ~~~
~ 19%
~ !!. 20% ~
N
~
~
C
§
21%
22%
a
~ R 22%
21%
~ ~ ~
g 23%
-J § 23%
~ - 24% .. ~ a. 24%
~ 25% ~ ~ 25%
~ 26% "" ~ 26%
S· 27%
28%
s· 27%
28%
~ ~
~ 29% ~ 29%
~ 30% ~ 30%
?
~
31%
32%
?
~
31%
32%
-.
("';)
~
<:;:)
~
33%
34%
35%
-.
("';)
~
<:;:)
~
33%
34%
35%
36% 36%
~ 37% ~ 37%
;;. 38% ;;. 38%
~ 39% ~ 39%
~
40%
~
40%
~ • ~ •
~
~
r-
~
0
~
I Co)
0
~
0
~
I·
I\)
•
0
~
I:
II
ca:r ~
~
r- ~
0
~
I Co)
0
~
0
I ·•
I\)
'#.
0
~
I:
ca
:r
100%
90%
80%
C
j
0 Margin
u 70%
en
is
a;
CI)
60%
.. • 10%
Q
-20%
.9! 50%
!
CI)
30%
a: 40% --40%
a;
E
a0 30%
20%
10%
0%
Trade Deal Discount
Figure 9.17. Item C: Optimal deal size by profit margin as a function ofthe trade deal.
25%
20%
C
:3 Margin
o
u
en
is
'i 15% • • -10%
CI)
Q ~ ~ --20%
.9l ~
~ ~ - - - 30%
~a: 10% --40%
a;
E
ao
5%
Trade Deal Discount
Figure 9.18. Item D: Optimal deal size by profit margin as afunction of the trade deal.
- 248-
30%
25%
§ Margin
o
.~ 20%
Q
~
- • • 10%
--20%
R 15% - - - 30%
Ia: --40%
'ii ,
E 10%
ao I
I
I
5% I
I
I
I
Trade Deal Discount
Figure 9.19. Item E: Optimal deal size by profit margin as a function ofthe trade deal.
25%
20%
c Margin
I
is
'ii 15% • • • 10%
•
Q --20%
.ii - - 30%
j
a: 10% --40%
'ii
E
ao
5%
Trade Deal Discount
Figure 9.20. Item F: Optimal deal size by profit margin as afunction of the trade deal.
- 249-
9.5. Conclusions
This chapter started with the development of sales response models for six items in the
Com Flakes & Co sub-category of the breakfast cereals category. It was found that price
deals for the six items had little impact on store traffic and store sales for five of the six
items, implying complete inter-category cannibalization. The store-level response for the
sixth item's price deal was very strong according to the model parameter estimates.
However, it is likely that the store-level response was overestimated. On the other hand, as
noted in Chapter 7, the choice of the null hypothesis favors the finding of no effect.
The price deal for the focal item had a positive effect on category sales for five out of
six items while sales of the focal item was always positively influenced by the price deal.
Complete intra-category cannibalization was found in only one case. Three of the items'
price deals caused moderate intra-category cannibalization (i.e., item sales rise more than
category sales), while two items' price deals caused no intra-category cannibalization.
Myopic optimization of the deal size usually leads to larger price deal discounts than
what would be optimal from a store-level perspective. If store traffic and store sales are
responsive to promotion for an item, then the myopically determined deal discount may be
smaller than optimal. Category-level optimization generally leads to smaller deal discounts
than item-level optimization. As these are often closer to the optimal deal, the sub-
optimization cost is smaller when using category-level optimization than when using item-
level optimization.
The size of the optimal deal discount is positively related to the size of the trade deal
discount and to the profit margin of the category. As small trade deals may fail to induce a
retail promotion, the relative pass-through is in general higher for larger trade deals
(assuming that the retailer optimizes the deal size at the store level). Another way the
manufacturer can use to influence the relative (and absolute) pass-through is by imposing a
deal requirement, that the retailer only receives the trade deal if a price deal is executed in
the store. Such a deal requirement changes the economic profits of the price deal, because
the promotion profits are measured on an incremental basis.
- 250-
10. EMPIRICAL ILLUSTRATION 3: PASTA
10.1. Introduction
This chapter contains the third empirical illustration of the proposed model. The model
is here applied to six items in the pasta category. Analysis and presentation of results
follow the format of the previous chapter. First, the pasta category is described. Then,
sales-response models are built at the store-, category-, and item-levels. The sales-response
models are then integrated to one sales-response model per item. These models are used to
study the profit impact of sales promotion and the optimal deal discount. The retailer pass-
through of trade deals is also examined.
10.2. Description of the Pasta Category

The test store sells six brands of pasta, two of which dominate the category. Sales of the
third largest brand are less than half of the second largest, but almost three times the sales
of the fourth brand (see Figure 10.1).
Some descriptive measures for the pasta category are given in Table 10.1. Although
pasta only accounts for 0.4 percent of store sales, 2.6 percent of the shoppers buy from the
category. The average category purchase was SEK 17.09. Sales promotion is rather
infrequent in the category, leading to a low average deal cost (3.0 percent of category sales
valued at regular prices).
Description Svmbol Value

Average Basket Value Sf/at 103.46
Pasta's share of sales sjt/St 0.4%
Pasta baskets to store traffic bjt/Bt 2.6%
Average purchase amount sjt/bjt 17.09
Deal cost in percent of sales dei!lsi! 3.00/0
Table 10.1. Descriptive measures for the pasta category.
- 251 -
Brand 6
Brand 5
Brand 4
Brand 3
Brand 2
Brand 1
0% 5% 10% 15% 20% 25% 30% 35% 40% 45%

Figure 10.1. Brand sales in the pasta category. The bars represent sales valued at regular
prices divided into actual sales (AS) and deal cost (DC).
Breaking category sales down by variety and brand, the results are presented in Table
10.2. Macaroni is the largest variety or sub-category with approximately one-third of
category sales. Spaghetti is second largest with 28.5 percent followed by tagliatelle etc.
(e.g., farfalle, penne, rigate) with 18.9 percent. The tortellini and the lasagne sub-categories
are fairly small.
Brand SpaghetU Tagliatelle etc. Torte/lini Lasagne Macaronj TOTAL

Brand 1 9.5% 32.5 % 42.0%
Brand 2 9.9 % 11.40/0 6.5% 5.80/0 33.5 %
Brand 3 5.6 % 4.9 % 3.1% 13.5%
Brand 4 0.9 % 2.5 % 3.4%
Brand 5 1.7% 1.0 % 2.7%
Brand 6 2.0% 1.7 % 1.2% 4.9 %
TOTAL 28.5% 18.9% 10.5 % 8.3% 33.70/0 100.0 %
Table 10.2. Market shares of brands and sub-categories.
The macaroni sub-category is totally dominated by the largest brand with 96 percent of
the sub-category's sales. Brand 1 has a strong position in the spaghetti sub-category as well
with sales close to that of the sub-category leader, Brand 2. The third largest brand in the
pasta category, Brand 3, holds position three in the spaghetti sub-category with close to 20
percent of the sub-category's sales.
Brand 2 holds large shares of the tagliatelle etc. sub-category as well as the tortellini and
lasagne sub-categories. In fact, the tortellini etc. sub-category is Brand 2's most important
- 252-
sub-category in terms of sales. Lasagne is the sub-category where Brand 2 has its largest
sub-category market share but, as this sub-category is the smallest, its importance for Brand
2's total sales is not very large.
Brand 3's most important sub-category in terms of sales is the spaghetti sub-category,
while tortellini is the sub-category where Brand 3 has its largest market share with almost
30 percent of sub-category sales.
10.2.1. Selection of Items for Analysis
Two items from each of the three dominating brands were selected for the further
analysis. In addition to large sales, these items have been promoted more than other items
with similar sales volumes. The six items are presented in Table 10.3.
Relative Market
Item Brand Variety Size Deal Cost Share Rank
Item A Brand 1 Macaroni 1000g 10.80/0 8.2°k 1
Item B Brand 1 Macaroni 4.5kg 4.60/0 6.10/0 3
Item C Brand 2 Tagliatelle 500g 4.5% 4.6% 6
Item D Brand 2 Spaghetti 500g 9.3% 3.8% 9
Item E Brand 3 Tortellini 500g 7.6% 3.10/0 11
Item F Brand 3 Spaghetti 1000g 3.10/0 2.60/0 12
Total 28.40/0
Table 10.3. Description of the six items selectedfor further analysis. Market shares in
percent of category sales.
Item A is the best-selling item in the pasta category with 8.2 percent of category sales. It
is also the item having the highest relative deal cost, i.e., the item with the largest
percentage of total sales (valued at regular price) that is given to the consumers in the form
of as price deals. Item B, ranking third in terms of sales, is the same product as Item A but
in a larger package. Item C and Item D are number six and number nine in the overall
market share rating in the category. Item E ranks eleventh while Item F has rank 12. In
total, the test store sells 55 different items in the pasta category.
10.3. Building the Response Models
10.3.1. Data Manipulations
Store sales valued at regular prices were approximated using the same procedure as in
the previous chapters, i.e., the actual sales of other categories were increased by five
percent and added to the exactly calculated category sales valued at regular prices. The
- 253-
approxmiated store sales valued at regular prices were then used in the calculations of
average basket value and the pasta category's share of store sales. The average basket value
estimated in this chapter is the same as in Chapter 9, Le., SEK 103.46.
As described in chapter 9, a period of store traffic and store sales due to a reconstruction
of the store's layout led to biased estimates of the importance of store traffic for store sales.
These observations were consequently discarded before the models were estimated.

Before starting the model building, we will examine the correlations between store
traffic and the category-level variables and between store traffic and the category-to store
ratios. Table 10.4 presents the first set of correlations, those between store-level and
category-level variables.
Bt Sf bjt
Bt 1.00
Sf 0.87 1.00
bit 0.52 0.46 1.00
sjt 0.54 0.52 0.91 1.00
Table 10.4. Pearson correlations between store traffic (Bt), store sales (St), number of
pasta baskets (bjt), and pasta sales (sjt).
Store traffic and store sales are, as noted before, strongly correlated. The correlations
between store traffic and category baskets and between store sales and category sales are
also high. We can also observe that the number of category baskets is highly correlated
with category sales. All these correlations indicate that the indirect modeling approach may
be useful when modeling category sales.
Bt $VBt bjtlBt siVSt

Bt 1.00
St/Bt 0.64 1.00
bit/Bt -0.13 -0.08 1.00
sjt/Sf -0.41 -0.58 0.76 1.00
Table 10.5. Pearson correlations between store traffic (Bt), average basket value (St/Bt),
category-to-store baskets (bjt/Bt), and category-to-store sales (sjt/St).
Table 10.5 provides further support for the modeling approach. The correlation between
store traffic and the ratio of category-to-store baskets (store traffic) is low. Somewhat
disturbing however is the fairly strong negative correlations that appear for the category-to-
store sales ratio. Still, what is more important is how successful the regression models are
in explaining the variance in the dependent variables.
- 254-
10.3.3. Store-Level Response Models
The exponential model structure was used for the store-level models. The linearized
• <
form of the model was described jn Chapter 8 (equation 8.1). The variables also follow the
definition in Chapter 8.
The estimation procedure follows the pattern described in the Chapter 7. For each
dependent variable, a maximum model including price deal variables for all six items and
six dummy variables indicating the day-of-the-week was specified. Variables not
significant at the 20 percent level in the initial estimation were dropped and the models
were re-estimated. Variables were dropped and the models re-estimated until all remaining
variables were significant at the 10 percent level. Recall that a variable with the wrong sign
was regarded as insignificant. Table 10.6 presents the final regression results.
Store Traffic Store Sales Basket Size

Constant 7.94 0.01 12.44 0.02 4.46 0.01
Price Deal
Item A 0.40 0.11 0.55 0.16 0.15 0.08
Item B
Item C
Item D
Item E
Item F
Day of Week
Tuesday 0.06 0.02
Wednesday 0.10 0.02
Thursday 0.24 0.04 0.22 0.02
Friday 0.11 0.03 0.49 0.04 0.41 0.02
Saturday 0.06 0.03 0.28 0.04 0.26 0.02
Sunday -0.13 0.03 -0.16 0.04
adj.R.sq. 280/0 58% 720/0
n 161 161 161
Table 10.6. Results ofstore-level response models.
Only Item A's price deal has a significant impact on store traffic, store sales and average
basket value. Price deal for Item A increases store traffic and the average basket value. This
means that its effect on store sales is larger than the effect on store traffic.
Day-of-the-week effects are found significant for all three store-level measures.
Consistent with the findings in the previous chapters, the basket-value model is the most
successful of the three store-level models in explaining the variance of the dependent
variable. The store sales model also has a high coefficient of determination while the
amount of variance explained in the store-traffic model is less than 30 percent.
- 255-
10.3.4. Category-Level Response Models
Relative category-level responses to sales promotion are examined in this section.

Figure 10.2 depicts the development over time of these variables. The two category-level
ratios were modeled using the logistic function. The estimation equation was described in
Chapter 8 by equation (8.2).
Table 10.7 presents the final regression models obtained after application of the same
procedure which was used in Section 10.3.3. None of the independent variables of the
maximum model were retained in the final category-to-store baskets model. The model
obviously did not explain any of the variance in the baskets ratio.
The sales ratio is more predictable than the basket ratio for the pasta category. The sales-
ratio model explained 34 percent of the variance in the dependent variable. Item E was the
only item having a significant impact on the category sales-ratio. Other variables that
affected the sales-ratio were the day-of-the-week dummy variables. These show that pasta
takes a larger share of Sunday, Monday, and Tuesday baskets' value than of other days'
baskets.
- 256-
0.045 0.009
0.04 0.008
0.035 0.007
0.03 0.006
0.025 0.005
- - Basket Ratio
- - Sales Ratio
0.02 0.004
0.015 0.003
0.01 0.002
0.005 0.001
Figure 10.2. Category ratios' development over time.
Category/Store Category/Store
Baskets Sales
coefficient s.e. coefficient s.e.
Constant -3.62 0.02 -5.30 0.03
Price Deal
Item A
Item B
Item C
Item D
Item E 1.38 0.41
Item F
Day of Week
Tuesday
Thursday -0.26 0.05
Friday -0.37 0.05
Saturday -0.29 0.05
Sunday
adj.R.sq. 0%
n 161
Table 10.7. Category-level response to sales promotion in the pasta category.
- 257-
10.3.5. Item-Level Response Models
Three item-level models were built for each item. The basket-ratio and sales-ratio
models both assumed the logit model structure, while the exponential model was used for
the average item-purchase model (see Section 8.3.5).
The maximum models included all variables found in Table 10.8, and were identical
across all six items. The final models were found using the procedure described above,
with the only difference the formal null hypotheses for other items sales promotion for
other than the focal item; promotion would have a positive or zero effect on the ratios. A
difference between the models in this chapter and those built for the breakfast cereals is
that special display for the focal item was included in the maximum model.
Table 10.8 presents the regression results for Item A. Contrary to the category and store-
level models, the basket-ratio model explains the most variance. This was also found in the
cereals and coffee categories. All three models explain approximately 50 percent of the
variance in the dependent variable.
None of the day-of-the-week dummy variables was retained in the final models for Item
A. Item A's own price deal and special display had significant impact on the basket-ratio
and the sales-ratio. The average item purchase was only effected by the price deal, which
I
increased the average purchase quantity.

Item B's price deal had significant negative impact on Item A's basket ratio and sales
ratio.
Item B's response models are presented in Table 10.9. Price deal and special display for
Item B had significant positive effect on all three dependent variables. Item A's price deal
affected Item B's basket and sales ratios negatively while Item CIS price deal had positive
effect on the average purchase of B.
Three of the day-of-the-week dummy variables were significant in the sales-ratio model,
while only one was significant in the basket-ratio and average item-purchase models. The
models for Item B differ from Item A's models in that the average purchase model accounts
for more variance than does the basket-ratio model.
- 258-
coefficient I.e, coefficient I.e, coefficient I.e,
Constant -2.89 0.05 -2.78 0.05 2.93 0.00
Price Deal
Item A 3.95 0.69 4.04 0.72 0.19 0.02
Item B -2.36 0.81 -2.03 0.84
Item C
Item D
Item E
Item F
Display Focal Item 1.18 0.30 1.17 0.31
Day of Week
Tuesday
Wednesday
Thursday
Friday
Saturday
Sunday
adj.R.sq. 520/0 500/0 470/0
n 153 153 153
Table 10.8. Promotional sales-response, Item A.

coefficient I.e, coefficient se, coefficient s,e..
Constant -2.03 -2.63 0.08 2.17 0.00
Price Deal
Item A -1.66 0.63 -1.77 0.62
Item B 2.53 1.30 2.88 1.29 0.23 0.08
Item C 0.11 0.05
Item D
Item E
Item F
Display Focal Item 0.96 0.37 0.97 0.37 0.07 0.02
Day of Week
Tuesday 0.02 0.01
Wednesday
Thursday
Friday -0.42 0.17
Saturday -0.35 0.17
Sunday -0.32 0.17 -0.49 0.17
adj.R.sq. 240/0 29°16 32('1/0
n 139 139 139
Table 10.9. Promotional sales-response, Item B.
- 259-
Turning to the item-level models for Item C, we recognize the more normal pattern of
more variance explained by the basket-ratio model than by the sales-ratio model, which in
tum explains more of the variance than the average purchase model. None of the
independent variables of the maximum model was retained in the final average purchase
model (Figure 10.13).
Item C's price deal was the single sales promotion variable found to have significant
impact on the basket and sales ratios. Three day-of-the-week dummy variable were retained
in the final ratio models, indicating some variation over the week.
Table 10.10 presents the final models for Item D. Price deal for Item D had significant
impact on all three item-level measures, while other items' price deals and Item D's special
display failed to influence any of the three measures. One of the day-of-the-week dummy
variables had significant impact on the sales ratio. The basket ratio, on the other hand, did
not show any significant day-of-the-week effects. In contrast, the average item purchase of
Item D varied over the week with four of the dummy variables significant.
The average item purchase model accounted for a larger share of the variance than did
the sales-ratio model. The worst performance in terms of variance explained was shown by
the basket-ratio model. This is the same pattern as was found for Item B.
- 260-
coefficient s,e coefficient s,e, coefficient s,e.
Constant -3.42 0.07 -3.23 0.07 3.00 0.01
Price Deal
Item A
Item B
Item C 3.79 0.68 3.89 0.69
Item D
Item E
Item F
Display Focal Item
Day of Week
Tuesday -0.36 0.15 -0.36 0.16
Wednesday
Thursday
Friday
Saturday 0.31 0.14 0.27 0.14
Sunday 0.34 0.14 0.27 0.14
adj.R.sq. 00/0
n 143
Table 10.10. Promotional sales-response, Item C.

coefficient s,e, coefficient s,e, coefficient s,e,
Constant -3.09 0.06 -3.52 0.06 2.32 0.02
Price Deal
Item A
Item B
Item C
Item D 3.02 0.56 3.85 0.59 0.92 0.12
Item E
Item F
Display Focal Item
Day of Week
Tuesday 0.09 0.03
Thursday 0.08 0.03
Friday
Saturday 0.06 0.03
Sunday 0.07 0.03
adj.R.sq. 230/0 31 0/0
n 151 151
Table 10.11. Promotional sales-response, Item D.
- 261 -
Item E follows the more common pattern with more of the variance in the basket-ratio
and the sales-ratio explained by the model than did the average item purchase model. The
two ratio models explain close to 50 percent of the variance while the average item
purchase model explained 9 percent (Table 10.12).
None of Item E's final models contain day-of-the-week dummy variables. The item's
own price deal had a large positive effect on the basket- and sales-ratios. However, it failed
to influence the average item purchase. Item CIS price deal influenced the average item
purchase positively while Item B's promotion had a negative impact on the sales ratio. The
basket- and sales-ratios were both influenced negatively by Item F's price deal.
The results for Item F appear in Table 10.13. The item's own price deal had significant
impact on the sales- and basket-ratios while its effect on the average item purchase was
insignificant. Item A's price deal had a small positive impact on the average item purchase.
No other promotional variables were retained in the final item-level models.
The Friday dummy variable was found significant in the sales model. Other day-of-the-
week dummy variables were insignificant in all three models.
Item F's models were not very successful in terms of variance explained. The best fitting
model was the basket-ratio model with a coefficient of determination of 13 percent. The
sales-ratio model explained 11 percent of the variance while the average item purchase
model only reached five percent.
- 262-
Item/Category Item/Category Average Item
coefficient s.e. coefficient s.e. coefficient s.e
Constant -4.07 0.06 -3.23 0.07 3.63 0.02
Price Deal
Item A -2.10 1.01
Item B 0.72 0.26
Item C
Item D
Item E 11.77 1.51 11.57 1.59
Item F -7.05 1.77 -5.41 2.09
Display Focal Item
Day of Week
Tuesday
Wednesday
Thursday
Friday
Saturday
Sunday
adj.R.sq. 47% 46% 90/0
n 71 71 71
Table 10.12. Promotional sales-response, Item E.

coefficient s e. coefficient s. e. coefficient s. e.
Constant -3.69 0.06 -3.59 0.07 2.87 0.00
Price Deal
Item A 0.08 0.03
Item B
Item C
Item D
Item E
Item F 5.84 1.37 4.99 1.39
Display Focal Item
Day of Week
Tuesday
Wednesday
Thursday
Friday -0.29 0.16
Saturday
Sunday
adj.R.sq. 130/0 110/0 5%
n 121 121 121
Table 10.13. Promotional sales-response, Item F.
- 263-
We shall now summarize the results of the regression models in order to get a better
overview of the important results and to get an indication of the validity of the indirect
modeling approach. First, we discuss the models' explanatory power. Then, the types of
variables significant in the models are discussed.
The explanatory power of a regression model can be expressed as the share of the
variance of the dependent variable the model explains. Table 10.14 presents the amount of
variance each of the 23 regression models explained.
Baskets Sales Quantity

Store Level 28% 580/0 720/0
Category Level 0% 340/0 n.a.
Item Level Average 290/0 30% 210/0
Item A 52% 50 0k 47%
Item B 24% 290/0 32%
Item C 24% 23% OOk
Item 0 16% 230/0 310/0
Item E 47% 460/0 90/0
Item F 13% 11 % 50/0
The store-level model of average purchase amount is the model with the largest share of
explained variance. This is important support for the indirect modeling approach because it
indicates that the-ratio (i.e., store sales divided by store traffic) is a relevant measure. The
store-level regressions show that store sales can be broken down into store traffic and
average purchase amount, where variations in store traffic are less deterministic than
variations in the average purchase quantity.
The second ratio in the sales decomposition model is the category-to-store-sales ratio.
The model explained about one-third of the variance in this ratio. Variations in the ratio of
category-to-store baskets could not be explained by the model.
In comparison with the store-level n10dels, the item-level models explain less of the
variance in the dependent variables (i.e., the ratios). A further difference between the store-
level models and the item-level models is that while the store-level model of basket value
(average purchase amount) explained more of the variance than the models of store traffic
and store sales, the opposite was found at the item-level; the models of average item-
purchase amount item per buyer generally explained less variance than the models of item-
to-category baskets and item-to- category sales.
- 264-
Sales Promotion Day-of-the-Week
Baskets Sales Quantity Baskets Sales Quantity
Store Level 1 1 1 345
Category Level o 1 n.a. o 4 n.a.
Item A 331 000
Item B 333 131
Item C 1 1 0 330
Item 0 1 1 1 014
Item E 2 3 1 000
Item F 1 1 1 010
Table 10.15. Number of significant model parameters.
Table 10.15 presents the number of significant model parameters of the regression
models. It can be seen that more day-of-the-week dummy variables entered the store-level
models than on average entered the itenl-level models. The opposite pattern was found for
the promotion variable. More promotion variables entered the average item-level model
than entered the store-level models. This lends support for the modeling approach in that
variations over the week are caught mainly by the store-level variables while sales
promotion mainly affect the item-level ratios.

This section takes the sales-response models of the previous section and investigates the
resulting promotion-profits for the retailer, assuming that the retailer's margins are 20
percent for the focal item, other items in the category, and other categories. It is also
assumed that the retailer is offered a trade deal worth 20 percent of regular retail price for
items sold, with no requirement that the item should be dealt.
The item- and category-level models developed above are relative to the category, the
store sales and traffic. In order to get the absolute sales and basket response we need to
integrate the relative models. The integrated sales-response models are then used to study
the absolute sales-response to different levels of price deals for the six items (A through F).
Table 10.16 summarizes the model parameters for the six items.
Store Traffic Basket Value Cetagory Item

Focal Item Constant Deal Constant Deal Constant Deal Constant Deal Displav
Item A 7.94 0.40 4.46 0.15 -5.30 0.00 -2.78 4.04 1.17
Item B 7.94 0.00 4.46 0.00 -5.30 0.00 -2.63 2.88 0.97
Item C 7.94 0.00 4.46 0.00 -5.30 0.00 -3.23 3.89 0.00
Item D 7.94 0.00 4.46 0.00 -5.30 0.00 -3.52 3.85 0.00
Item E 7.94 0.00 4.46 0.00 -5.30 1.38 -3.23 11.57 0.00
Item F 7.94 0.00 4.46 0.00 -5.30 0.00 -3.59 4.99 0.00
Table 10.16. Summary ofmodel parameters by focal item.
- 265-
The table entries should be interpreted as follows. Setting Item A as the focal item we
are interested in the sales responses caused by Item A's sales promotion. The first row in
Table 10.16 gives all the model parameters needed, assuming that the responses are
evaluated for a Monday and that the other five items are not promoted.
10.4.1. Item A
Evaluating the sales changes due to price deals for Item A from zero percent to 30
percent of the regular retail price for Item A results in Figure 10.3.
The store-level sales effects are of a greater magnitude than the category and item-level
sales responses. Therefore, separate scales were used for the store-level effects and the
category- and item-level effects in Figure 10.3. In fact, item- and category-level effects are
fairly unimportant for Item A compared with the store-level effects. This is because Item
A's price deal had a large effect on store traffic and basket value. However, this result
should be interpreted with some care because the effect appears to be overestimated. A
retail manager would be wise to obtain more data to validate the extreme traffic building
effect of Item A's price deal.
The profit impact of Item A's price deal is depicted in Figure 10.4. The store-level profit
impact was so much larger that it was drawn on another scale than item- and category-level
promotion-profits. Category-level profits are larger than item-level profits for all evaluated
deal discounts. Thus, item-level profit evaluation underestimates the profitability of a price
deal for Item A. The item-level profit maximum occurs at a smaller deal discount than the
category-level maximum. Store-level promotion-profits are still increasing at the deal
discount 30 percent indicating that deals greater than 30 percent are even more profitable.
- 266-
250 45000
40000
200
35000
VI
CD
'ii
30000 en
VJ Focal Item
~ 150 'i ---0--
en !
CD
-fr-- Other Items
.5
CD
en
c
ca
25000
20000
0
~
-- • Category
t3 100
8»
c - - - Store Sales
15000 ca
J::.
0
10000
50
5000
Deal Discount
Figure 10.3. Sales changes caused by Item A's price deal.
16.00 9000.00
14.00 8000.00
12.00
7000.00
10.00
6000.00 :t!
i
e
Q.
8.00
5000.00 ~
e ----0--- Focal Item
'ii
0
c
:g
6.00 aca - - • Category
E 4000.00 ~ - - - Store
0 4.00 o
~ ~
3000.00
2.00
2000.00
0.00
-2.00 1000.00
-4.00 0.00
Deal Discount
Figure 10.4. Promotion-profits ofItem A's price deal.
- 267-
10.4.2. Item B
In contrast to Item A, none of the other five items' price deals had any effect on store
sales. This may be an overly negative view of the price deal effects, but is more likely to be
correct than the large positive effects of Item A's promotion.
Figure 10.5 shows the changes in sales due to Item B's price deal. Category and store
sales are not affected by the price deal. The only effect is that Item B cannibalizes on the
other items' sales in the category.
The profit impact of price deals for Item B is, according to Figure 10.6, a sad story for
the retailer, with negative promotion-profits for all deal discounts at all levels.
Although item-level promotion-profits are negative, item-level evaluation of the profit
impact gives an overly positive estimate. The category and store-level promotion-profits
are more negative than item-level promotion-profits. Irrespective of the level of analysis
used, the retailer would choose not to price deal Item B. Note that category- and store-level
promotion-profits coincide because of complete intra-category cannibalization.
- 268-
100
80
60
40
en
CL)
--0--- Focal Item
"ii 20
en ----f:r- Other Items
.5
CL)
0)
*-*-*-*--*->.<--*-*-*-* - .. Category
c
CG ~ - x - Other Categories
.c -20
0
- - Store Sales
-40
-60
-80
-100
Deal Discount
Figure 10.5. Sales changes casued by Item B's price deal.
5.00
0.00 n-=--O--O----O-==M::---t----t---t---f-----i~___t-___+-__+-__+-_+-____i
-5.00
~ -10.00
Ii: --0---- Focal Item
"ii
5 -15.00 • • • Category
:g --Store
E
o
Ii: -20.00
-25.00
-30.00
-35.00
Deal Discount
Figure 10.6. Promotion-profits ofItem B's price deal.
- 269-
10.4.3. Item C
The sales responses due to Item C's price deal are presented in Figure 10.7. Again, store
and category sales are unaffected by the promotion. All incremental item sales come from
other items in the pasta category. The strength of the response is similar to that of Item B
although Item C has a lower base sales as can be seen by comparing the intercepts of the
item-level models.
The results for Item C are slightly better than for Item B. Figure 10.8 shows that item-
level promotion-profits are small but positive. Category- and store-level promotion-profits
are equal and negative for all positive deal discounts. It would thus be optimal not to price
deal Item C. Retailers using item-level analysis may decide on a price deal of 12 or 14
percent.
- 270-
100
80
60
40
VI --0-- Focal Item

CD
ii 20
en -----tr- Other Items
.5
CD o -
m
c
as
.c -20 ~ - x - Other categories
0
- - Store Sales
-40
-60
-80
-100
Deal Discount
Figure 10.7. Sales changes caused by Item C's price deal. Note that the category and store-
level changes are zero.
5.00
0.00 -Q-&i.....ii;;;;;;:::::::::i----t-----t----j----t--t----t----t----t---=:y-..::::::::::-t-------t---j-----j
-5.00
-------0-- Focal Item
-10.00 • • • Category
--Store
-15.00
-20.00
-25.00
Deal Discount
Figure 10.8. Promotion-profits ofItem C's price deal. Note that store and category-level
profits coincide.
- 271 -
10.4.4. Item D
The picture for Item D is very much the same as we have seen for Items Band C. The
incremental sales of the focal item come entirely from other items in the category, Le., the
intra-category cannibalization is complete. The amount of sales stolen from other items in
the category is smaller for Item D than for the two previous items. The sales changes are
depicted in Figure 10.9.
The profit impact of price deals for Iten1 D is very small for all deal discounts evaluated.
Again, category- and store-level profits are equal while item-level profits are somewhat
higher. Item-level analyses would lead the retailer to set price deals which are too large for
Item D. However, the promotion-profits are not very sensitive to price deals for Item D.
- 272-
80
60
40·
:I 20 -0-- Focal Item

'iii
tn ------tr- Other Items
.5
CD - - • Category
C)
i ~ - x - Other Categories
.c
o -20 - - Store Sales
-40
-60
-80
Deal Discount
Figure 10.9. Sales changes caused by Item D's price deal. Note that the category and
store-level changes are zero.
2.00
0.00 -()IIii......oI:::::::-_t----t---+---t----+----t---+-_t-----t-----'T"'''"-<::::::---+----t---t--------l
-2.00
-4.00
e= -6.00
i::
D. -----D- Focal Item
'iii
6 -8.00 - • • Category
:g --Store
E
e -10.00
D.
-12.00
-14.00
-16.00
-18.00
Deal Discount
Figure 10.10. Promotion-profits ofItem D's price deal. Note that category- and store-level
profits coincide.
- 273-
10.4.5. Item E
Figure 10.11 shows that the sales changes due to Item E's price deal are more complex
than for item's B, C, and D. Item E's price deal has no effect on store sales but does have a
positive effect on category sales. The category thus cannibalizes on other categories' sales.
Price deals smaller than 18 percent of regular retail price cause no intra-category
cannibalization. Indeed, sales of other items in the category also increase somewhat from
the price deal for Item E when the price deal does not exceed 18 percent. Larger price deals
lead to moderate intra-category cannibalization.
The promotion-profits of Item E's price deals differ depending on the level of analysis.
Store-level promotion-profits are lower than item- and category-level profits for all sizes of
the price deal. Small price deals lead to small, but positive, store-level promotion-profits,
while larger deals rapidly tum profits negative (see Figure 10.12).
Price deals up to 18 percent are more profitable at the category-level than at the item-
level. Thereafter, item-level profits continue to rise while category-level promotion-profits
start to decline.
- 274-
1000
800
600
,. ,. ,. ,.
.. ,. ,. " ,. ,.
400
UJ
II)
,. ,. ,. -0-- Focal Item
1i
en 200 ---l:r-- Other Items
.5
II) - - • Category
C)
C
as o - - x - Other Categories
.s:.
0
- - - Store Sales
-200
-400
-600
·800
Deal Discount
Figure 10.11. Sales changes caused by Item E's price deal.
100.00
80.00
. ,. . . . ... ...
60.00
40.00
,. ... ,. ,. ,. ~
,,
~
20.00
Q. Focal Item
.-
---0-
1i 0.00
c
0
:g g; ";ft
~ ~ ~ ~
• Category
-20.00
E ---Store
0
Q: -40.00
-60.00
-80.00
-100.00
-120.00
Deal Discount
Figure 10.12. Promotion-profits ofItem E's price deal.
- 275-
10.4.6. Item F
Sales responses to Item F's price deal are, again, like those of Items B, C, and D. Figure
10.13 shows that price deals for Item F lead to complete intra-category cannibalization.
Store sales and sales of other categories are unaffected by Item F's price deals.
We have now seen the sales response to the six items' price deals. In the next section we
will investigate the profit impact of the promotions.
Figure 10.14 shows the results for Item F'g price deals. Due to complete intra-category
cannibalization, the category and store-level promotion-profits are identical. Category- and
store-level promotion-profits are negative for all deal discounts. Item-level promotion-
profits are positive but small for deal discounts up to 30 percent.
- 276-
100
80
60
40
Focal Item
~
-0--
'ii 20
U) --t:r-- Other Items
.5
- - • Category
8»
c
«I
"t!-
o - x - Other Categories
cJ -20
- - Store Sales
-40
-60
-80
-100
Deal Discount
Figure 10.13. Sales changes caused by Item F's price deal. Note that category and store-
level changes are zero.
5.00
0.00 -£?'I'i--..-....iiiiiiiiiiiiiii-±:=--t--t------t----+---+--i-----j,--------t---+--i--r---r
-5.00
--0---- Focal Item
- - • Category
--Store
-10.00
-15.00
-20.00
Deal Discount
Figure 10.14. Promotion-profits ofItem F's price deal. Note that category- and store-level
profits coincide.
- 277-
10.4.7. Summing Up
In summary, it was seen that the category-level and the store-level promotion-profits
coincide in four out of six cases (Items B, C, D, and F) due to complete intra-category
cannibalization. The category- and store-level profits were negative in these four cases.
The category-level profits were greater than the store-level profits in one case (Item E)
and lower in one case (Item A). The sales promotion had a positive impact on category
sales in both cases. For Item E, there was some degree of intra-category cannibalization,
while the inter-category cannibalization was complete. This made the category-level profits
greater than the store-level profits. Item A's sales promotion, on the other hand, caused
neither intra- nor inter-category cannibalization; sales of other items in the focal category
as well as sales of items in other categories increased. This made the store-level
profitsexceed the item- and category-level profits.
For all other items, the item-level profits were larger than the store-level profits. Due to
intra-category cannibalization, the item-level profits were greater than the category-level
profitsin all cases but two (Item A and Item E). For Item E, the moderate intra-category
cannibalization led to almost equal category- and item-level profits for small deal
discounts. Item A's sales promotion caused sales of other items in the focal category to rise.
The category-level profits were therefore greater than the item-level profits.
- 278-
We shall now investigate how the deal discount depends on the level of analysis used by
the retailer. The cost of myopic optimization is also examined.
It is assumed that gross margins are 20 percent for the focal item, the focal category, and
for other categories. It is also assumed that the manufacturer offers the retailer a trade deal
worth 20 percent of the retail price (implying a 25 percent discount on the wholesale price).
Three price deals were determined for each item: the item-level optimized deal (Deall);
the category-level optimized deal (Deal2); and the store-level optimized deal (Deal3). In
accordance with the central argument of this thesis, the store-level optimized deal is what
we regard as the optimal deal while the item and category-level optimization is myopic.
The promotion-profits of the store-level optimized deal are called Profit3.
The cost of setting another price deal discount than the optimal deal determined at the
store level can be estimated as the difference between Profit3 and the store-level
promotion-profits of such a price deal. The cost of item-level deal optimization is thus the
difference between Profit3 and the store-level promotion-profits of a deal of size Deall.
This cost of myopic deal optimization is denoted Lossl. In the same way, Loss2 is the cost
of category-level deal optimization. Table 10.17 presents the results for the six items in the
pasta category.
The results show that for all items except Item A, item-level optimization leads to larger
deal discounts than category-level optimization. Item A is different because of the store
traffic effects noted in the previous section. Also noted earlier is that the size of the store-
level effect is exaggerated, which is indicated by the suggested optimal deal of 100 percent
for Item A. The store-level profits of giving away Item A for free are estimated to SEK
30,000 per day, which is obviously too optimistic an estimate.
Items B, C, D, and F all failed to affect category and store sales. The retailer's optimal
deal for these items is therefore to absorb all of the trade deal and sell the items at their
regular prices. The store-level promotion-profits are, according to Table 10.17, zero for
these items. It should be noted, however, that this figure does not include the profits the
retailer makes by pocketing the trade deal.
Although Item E's price deal does have an impact on category sales, item-level deal
optimization leads to larger than optimal deals. Item-level optimization sets the price deal
at 27 percent, while category-level optimization sets the deal at 20 percent and store-level
optimization leads to a price deal of 11 percent. The 11 percent deal increases gross
margins by SEK 6.44 above what the retailer would obtain when absorbing the trade deal.
- 279-
Item DealReg Deal1 Deal2 Deal3 Loss1 Loss2 Profif3
Item A 00/0 160/0 190/0 1000/0 29831.79 28953.58 34290.25
Item B 00/0 3% 0% 0% 1.28 0.00 0.00
Item C 00/0 130/0 00/0 OOk 3.98 0.00 0.00
Item 0 00/0 13% 00/0 0% 3.00 0.00 0.00
Item E 0% 27% 200/0 110/0 73.51 15.62 6.44
Item F 0% 190/0 0% 00/0 5.68 0.00 0.00
Item A 1% 16% 190/0 100 % 29831.79 28953.58 34304.34

Item B 1% 30/0 10/0 10/0 0.89 0.00 15.82
Item C 1% 13% 10/0 1% 3.85 0.00 9.05
Item 0 10/0 130/0 1% 1% 2.91 0.00 6.84
Item E 1% 270/0 20% 11 % 73.51 15.62 15.62
Item F 1% 19% 1% 1% 5.66 0.00 6.46
Item A 100/0 160/0 190/0 100 % 29831.79 28953.58 34304.34

Item B 10% 0
10 k 10 % 10% 0.00 0.00 10.58
Item C 10 % 13% 10 0k 100/0 1.45 0.00 6.65
Item 0 10 0k 130/0 10 % 10% 1.10 0.00 5.03
Item E 100/0 270/0 200/0 110/0 73.51 15.62 15.62
Item F 100/0 19°k 10 % 10% 4.44 0.00 5.24
Table 10.17. Optimal deal discount, loss due to myopia, and optimal deal profits.
If the manufacturer would demand that the retailer passes through some part of the trade
deal to the consumers in order to receive the trade money, the promotion-profits, and
sometimes the optimal deal, will change. If the optimal deal discount does not change
because of the deal requirement, the increase in promotion-profits (compared with the no
deal requirement case) reflects the retailer's profits from absorbing all of the trade deal and
selling the item at regular price. If the deal requirement alters the optimal deal although the
deal requirement is low, the change in promotion-profits can only be regarded as an
approximation of the value of the pocketed trade money. The results when the
manufacturer requires a one percent or a ten percent price deal are presented in Table
10.17.
The loss due to myopic optimization changes only when the myopically determined deal
discount change or when the store-level optimal deal (Deal3) changes as a result of the deal
requirement, or both. A requirement of a one-percent price deal has no impact on item-
level optimal deal, but increases the optimal deal determined at store- or category-level, to
one percent in four out of six cases.
The item-level optimal deal (Deal1) changes for one item (Item B) as a result of a ten
percent deal requirement. The ten percent deal requirement also raises the store- and
category-level optimal deal for four out of six items. The change in promotion-profits by
increasing the deal requirement from one percent to ten percent is generally small.
- 280-
10.5.2. Trade Deal and Margins
This section investigates how the optimal deal discount and pass-through depend on the
gross margin and the trade deals, given that the retailer uses store-level deal optimization. It
is assumed that the gross margin of other categories is 20 percent and that the gross margin
of the focal item is equal to that of other items in the pasta category.
The results for Item A are presented in Figure 10.15. Due to the strong store-level effect,
the model claims that the retailer should give the consumers Item A for free during the
promotion period, whatever the trade deal. As already noted, this result should be taken as
an indication that a large price deal may be profitable, rather than as a suggestion that the
retailer should set the price at zero.
When intra-category cannibalization is complete, as in the case of Item B (Figure
10.16), the retailer will in general find it more profitable to absorb all of the trade deal than
to pass it through to the consumers. Only substantial trade deals will lead to price deals in
the store, and even then, the pass-through is small. Note that the gross margin of the focal
item (and other items in the focal category) has no impact on the optimal deal discount.
Item C is similar to Item B in that the intra-category cannibalization is complete. Figure
10.17 indicates that there is some difference between items Band C since the pass-through
is much larger for Item C than for Item B. The difference is that Item C's relative sales
response is stronger than Item B's. Although intra-category cannibalization is complete, the
incremental sales of Item C mean that higher effective margin sales (i.e., Item C's regular
margin and the trade deal) are substituted for the regular margin sales.
The regular gross margin of Item C (and other items in the pasta category) has no effect
on the pass-through.
Figure 10.18 presents the results for Item D. The optimal deal for Item D as a function
of the trade deal is identical to that for Item C. Going back to the model parameters for
items C and D (Table 10.16) we observe that the two items' models are virtually identical.
Item E (Figure 10.19) differs substantially from items B, C, and D. The optimal pass-
through is in general large and dependent on the regular gross margin of items in the pasta
category. Although price deals for Item E have no impact on store sales, it is profitable for
the retailer to let the pasta category cannibalize on other categories' sales when the pasta
category has higher gross margin than other categories. When gross margins in the pasta
category are equal to other categories' margins (i.e., 20 percent), the trade deal must exceed
8 percent (of retail price) before the retailer will deal Item E. Lower pasta margins
obviously lead to smaller pass-through of the manufacturer's trade deal.
The pass-through pattern for Item F resembles that of Item B, C, and D, the difference
being that the retailer's pass-through is somewhat higher for Item F. This is due to the
stronger relative-sales-response for Item F (see Figure 10.20).
- 281 -
f}~ ~~
~~ ~ c;"
Optimal Retailer Deal Discoun1 ~~ Optimal Retailer Deal Discoun1
~~
~
0
'#. '#.
......
~
I\)
?fl.
I\)
?fl.
U)
'#.
U)
'#. ~ 0 o
I\)
o
U)
o
.j:o.
o
01
o
0)
o o"""
(Xl
o
co
o g
~ ~ '#. '#. '#. '#. '#. '#. '#. '#. '#. '#. '#.
0%
~ 1%
~ 0%
~ ~
1%
?- 2%
~ 2%
3%
~
4% ~ 3%
~ 5%
~ 4%
~ 6%
~ 5%
~ 7% ~ 6%
7%
8%
~ ~
8%
9%
9%
~ 10%
§" 11% §" 10%
11%
~
~
12%
~ 12%
13%
f} 14% f} 13%
14%
~
~
~
15%
16%
f:.. 15%
~
~
t"'.l
D1 17% ~"
~
D1
16%
17%
~ i 18% ~ i 18%
~ C 19% ~
~ ~ ~ 19%
~
m. 20%
~ e. 20%
~ 52 21% ~
52 21%
~ 22%
tv
00 ~ § 23%
~ § 22%
tv ~ ~ § 23%
... 24% ... 24%
~
~ 25%
~
c.s:
~
26%
-"
~
25%
26%
27%
~ 28%
~ 27%
~ ~ 28%
~
29%
~ 30%
29%
5" 31% 5" 30%

31%
~ 32% ~ 32%
t"'.l t"'.l
33% 33%
~ ~
34% 34%
?
~
35%
36%
?
~
35%
36%
~ ~
~ 37% ~ 37%
c"
~
38% c"
~
38%
39%
39%
~ 40%
~ 40%
~ I
~ I
I. I.
~ ~
I
3: I
3:
~
~ D)
~ I ca
S'
I caS'
~
~
.j:o.
0
'#.
U)
0
'#.
I\)
0
'#.
C;
'#.
f} .j:o.
000
'#.
U)
'#.
I\)
'#.
......
0
'#.
14%
12%
c Margin
S 10%
~
Q
- - • 10%
I
.
Q
.!!
8%
--20%
- - - 30%
Ia: 6%
--40%
"iii
E
~ 4%
2%
Trade Deal Discount
Figure 10.17. Item C: Optimal deal discount by profit margin as a function of the trade
deal.
14%
12%
c Margin
S 10%
~
Q
~
- - • 10%
8%
--20%
.!!
- - - 30%
Ia: 6%
--40%
'i
E
;;
8' 4%
2%
Trade Deal Discount
Figure 10.18. Item D: Optimal deal discount by profit margin as a function of the trade
deal.
- 283-
35%
30%
c
g 25% Margin
fA
Q
~
- - • 10%
20%
--20%
.!!
- - - 30%
~a: 15%
--40%
ca
.5
g 10%
5%
Trade Deal Discount
Figure 10.19. Item E: Optimal deal discount by profit margin as a function of the trade
deal.
20%
18%
16%
C
;::,
0
Margin
() 14%
-.
0
Q
ca
G)
12% • 10%
C --20%
.!! 10%
~a:
30%
8% --40%
ca
E
a
0
6%
4%
2%
0%
Trade Deal Discount
Figure 10.20. Item F: Optimal deal discount by profit margin as afunction of the trade
deal.
- 284-
10.5. Conclusion
It has been seen that price deal in the pasta category has little impact on store traffic and
store sales. Only one of the six items' price deals had a significant impact on store sales.
This item did not affect the category's share of store sales while one item affected the
category's share of store sales without affecting store sales. Four items' sales promotion led
to complete intra-category cannibalization.
In these four cases, the retailer would maximize profits by absorbing all of the price deal
for most levels of trade deal. Only large price deals would lead to a partial pass-through of
the trade deal to the consumers. Retailers using item-level deal optimization would use
price deals which are too large.
Item and category-level deal optimization would lead to larger than optimal price deal
also for the item whose price deal affected relative category sales. The sub-optimization
would, however, be smaller for this item than for the four items whose promotions lead to
complete intra-category cannibalization.
Contrary to the other items, the item whose price deal had an impact on store sales
would receive a too small price deal if the retailer used item-level analysis when setting the
deal discount.
- 285-
11. CONCLUSION
11.1. Summary
The purpose of this study was model the profit impact of retailer promotion. The
motivation for this study was the observation that the profit impact of sales promotion is
poorly understood. In addition, most previous research about sales promotion have
assumed a manufacturer's rather than a retailer's perspective. As the retailer's objectives
with sales promotion differ from those of the manufacturer, there is-a large need for the
development of a retailer-oriented, promotion-profit model.
11.1.1. Promotion Profit as Change in Gross Profits

It was noted that the economic profit of any action is the change in profits caused by that
action. The profit of a sales promotion is thus the difference in the profits that would be
obtained if the promotion were launched and the profits that would be obtained in the
absence of the promotion. Measuring the profits of a promotion ex post thus involves an
estimation of what the profit would have been without that promotion. Ex ante prognoses
of the profit of a promotion require two estimates: the profits with and without the
promotion. The difference between these estimates is equal to the difference in gross
profits, assuming that fixed costs are not affected by the promotion.
For the firm selling only one item, gross profits change as a consequence of sales
promotion in two ways. The profit margin per unit sold may change and the sales volume
may change. While the change in per unit profit margin is readily available, the change in
sales volume must be estimated.
Sales promotion typically reduces the profit margin per unit while increasing the sales
volume. It is important to note that the lower margins not only pertain to the incremental
sales caused by the promotion, but also to the sales that would have occurred in the absence
of the promotion. The promotion profits are thus dependent on sufficient incremental sales
to overcome the negative effect caused by lower margins on baseline sales.
For firms selling more than one item, the profit of the promotion for one item (the focal
item) is not only dependent on the change in profits from sales of the promoted item, but
from the change in gross profits from sales of all other items as well. The change in gross
- 287-
profits from other items is only a consequence of changes in sales volumes, since the per
unit gross profits of these items are not affected by the promotion of the focal item. In order
to measure the profit of a promotion for the focal item, we need to estimate how sales of
other items will change.
Sales of other items than the promoted item may increase, decrease, or be unaffected as
a consequence of the promotion. When sales of other items decrease, we have substitution
effects, implying that the promoted item steals from the sales of other items. This is also
known as sales cannibalization. When sales of other items increase, we have
complementarity between the items. If sales of other items are unaffected by the promotion,
the profit of the promotion is simply the change in gross profits from the focal item.
If the number of items the firm sells is not too large, the sales effect on each individual
item can be estimated. The profit of the promotion can then be calculated as the sum of the
change in gross profits from sales of the promoted item and the change in gross profits
from sales of each other item. However, this is not feasible when the number of items is
large.
Grocery retailers carry large assortments consisting of several thousands of items. It is
well known that there are interdependencies among many of these items. The substitution
effects are perhaps the most obvious, but the complementarity effects nlay also be
important. The combination of large assortments and the interdependencies among the
items makes measurement of promotion profits in grocery retailing difficult, because the
total change in gross profits from other items may be large, even if the effect on each item
is small and difficult to measure.
11.1.2. The Sales Model
The solution suggested in this study is based on the notion that grocery shoppers usually
buy more than one item when shopping. Indeed, most grocery shoppers buy baskets of
groceries; the reason is probably the fixed cost in the form of the effort to visit the store.
This behavior gives rise to a general complementarity among all items that enter the
shopping baskets, and highlights the importance of store traffic, i.e., the number of
shoppers or shopping baskets.
Positive effects on sales of other items than the promoted item may then come from an
increase in store traffic caused by the promotion. The size of the average basket is thereby
important to observe because if the incremental shoppers only buy the promoted item, then
sales of other items would not be positively affected.
The shopping basket is also important for substitution effects. We can think of the
content of the shopping basket as three kinds of items: the focal item, its closest substitutes,
and all other items. The focal item's closest substitutes are the other itenls in the same
- 288-
category as the focal item. This category is cslled the focal category and we can calculate
its share of the basket's value. Multiplying by the number of shopping baskets gives the
sales of the focal item.
This calculation makes it possible to express the average content of the focal item in the
baskets as the product of the four factors: (1) the store traffic, (2) the average basket value,
(3) the focal category's share of store sales, and (3) the focal item's share of category's sales.
Expressed as an equation, sales of the focal item are:
(11.1)
where
= sales (valued at regular price) of item i, category j, period t

= category sales (valued at regular prices) category j, period t
= store sales (valued at regular prices) period t
= store traffic, Le., total nunlber of shopping baskets, in period t
Sales of other items in the focal category and sales of other categories than the focal
category can then also easily be calculated. By building models for each of these four
factors, we can estimate the change in sales of the focal item, other items in the focal
category, and of other categories caused by sales promotion (or other changes in the
marketing mix).
11.1.3. The Profit Model
The estimates of the changes in sales for the promoted item, other items in the same
category, and other categories can then be used to calculate the promotion profit. The
retailer's promotion profit is thus the sum of three terms: (1) the change in the gross profit
from the focal item, (2) the change in the gross profits from sales of the other items in the
focal category, and (3) the change in the gross profits from sales of items in other
categories. This measure of promotion profit is the way to measure promotion profits that
is most relevant to retailers, and is called store-level promotion profits or just promotion
profits. Expressed as an equation, we have:
~n3 == ~nl + (~n2 - ~nl) + (~n3 - ~n2) (11.2)

~ . "---v----' "---v----'
(I) focal Item (2) other items in focal category (3) other categories
where
- 289-
= change in item-level gross profits
= change in category-level gross profits
= change in store-level gross profits
The first term, the change in gross profits from the focal item, is called item-level
promotion profits, while the sum of the two first terms is called category-level promotion
profits. Both item- and category-level promotion profits are myopic in that they do not
consider the full impact of the promotion.
The store-level promotion profits of a promotion for the focal item can be calculated
using the formula:
foca~ item other items in focal category other categories
f1n3=~1~+&1(t8_DEAL)+S;(t8_t'-DEAL~+ ~ +~
item level pro~otional profit
store level promoti~nal profitcategories
(11.3)
where
* baseline sales of focal item valued at regular price

S1 =
&1 = change in sales of focal item caused by promotion
&2 = change in sales of focal category caused by promotion
Lls3 = change in store sales caused by promotion
m. = gross margin of item (percent of regular retail price)
~ = average effective gross margin of other items in the focal category
~ = average effective gross margin of products in other categories
DEAL = deal discount on focal item (percent of regular retail price)
t* = trade deal (percent of regular retail price) received by the retailer if
focal item is not promoted by the retailer
t8 = trade deal (percent of regular retail price) received by the retailer if
the focal item is promoted by the retailer
The trade deal received by the retailer is a function of the trade deal offered by the
manufacturer and the requirements the manufacturer imposes on the retailer. For example,
the manufacturer may offer a trade deal of 20 percent (in percent of the retail price) if the
retailer displays the item and offers the consumers at least a 10 percent deal. Formally, such
requirements can be modeled in the following way:
- 290-
t = {TROFFER if (DEAL ~ DEALREQ) and (DISPLAY ~ DISPREQ) (11.4)
o otherwise
where
TROFFER = trade deal offered on focal item (in percent of regular retail price)
DEALREQ = minimum deal required to obtain the trade deal
DISPREQ = indicator of display requirement to obtain the trade deal
The strength of substitution effects, or cannibalization, were defined formally using the
changes in item, category, and store sales. The intra-category cannibalization measure was
defined as:
(11.5)
An intra-category cannibalization measure of zero indicates complete intra-category

cannibalization, Le., that all incremental sales of the promoted item came from other items
in the same category. If the measure is unity, the promoted item does not cannibalize at all
on other items in the same category, and values greater than unity indicate that sales of
other items in the category also increase.
Similarly, inter-category cannibalization was defined as:
(11.6)
Similar measures were also defined to describe the relation between changes in the
number of buyers of the focal item and the change in the number of buyers of the focal
category, as well as the relation between the change in the number of buyers of the category
and the change in store traffic.
11.1.4. Myopic Measures and Myopic Optimization
The promotion measurement level generally affects the numerical outcome of the profit
measurement (see equation 11.3). Due to strong substitution effects within the focal
category (intra-category cannibalization), the item-level profit is almost always greater than
the category level promotion profit. Whether the category-level promotion profit is smaller
or larger than the store-level promotion profit depends on the degree of inter-category
cannibalization. If the promotion increases store traffic, it is likely that the category-level
measure understate the promotion profit. In such cases, the most myopic measurement
level (Le., the item-level) may actually be closer to the store-level promotion profit.
- 291 -
That myopic measures of promotion profits differ from store-level promotion profits is
in itself a reason to measure promotion profits at the store level. Even more important is
that the promotion chosen by the profit-maximizing retailer depends on the level of
measurement. Store-level deal optimization is, thus, the selection of the deal that
maximizes store level pronl0tion profits, while myopic optimization selects the deal
maximizing a myopic measure of promotion profits.
Myopic optimization leads to sub-optimization when the myopic optimization leads the
retailer to select another promotion than store-level optimization. The cost of the myopic
optimization is then the opportunity loss incurred by using a less than optimal promotion.
This cost is the difference in store-level promotion profits of the optimal deal and of the
myopic deal.
11.2. What is New?

The main contribution of this research is the system of models for analyzing retailer
profits of sales promotion. This system consists of two types of models: the sales-response
models and the profit models. In relation to previous sales-response models, the model
developed here can be seen as an extension and adaption of the indirect modeling approach
described by Naert and Leeflang (1978) to the grocery retailer's situation. Its most
important new feature is that it explicitly includes store traffic. A contribution of this
research that is closely related to the sales model is the usage of receipt data for the
measurement of the pronl0tion effects. The sales-model framework is conceptually related
to the DuPont model used for analyses of accounting data, in that both models decompose
the measure of interest into a chain of key ratios. Changes in the measure of interest can
then be analyzed in terms of changes in the key ratios.
The profit model developed here is based on the principle that the profit of an action is
given by the change in profits that this action causes. This is a basic economic principle,
borrowed from economic theory and operations research, that has been used by other
researchers studying promotion profits. What is new here is the specification of the
retailer's costs and revenues of the promotion. Also new is that the model identifies three
levels (item-, category-, and store-level) at which promotion profits can be measured. The
store level is the relevant profit measurement level for the retailer, while the item and
category levels are regarded as myopic. The observation that the promotion profits at the
store level and at the myopic levels are different is new, as is the definition of the cost of
myopic promotion optimization.
The formal definition of cannibalization in retailing is a contribution of the present
research. Consequently, the illustration of the impact of cannibalization on promotion
- 292-
profits is also new. This research also highlighted the importance of store traffic for
promotion profits.
Previous studies have shown that retailer promotion can be seen as a function of trade
promotion. This study extended those findings and showed that the deal size and the
retailer's pass-through depends on the:
• size of trade deal
• retailer's level of optimization
• gross margins for focal item, other items, and other categories
• cannibalization and store traffic effects
• deal requirements
The pattern of the impact of these factors is fairly complex, and there is important
interaction among these factors.
11.3. Generalizations
The model and measurement method can be generalized to other types of marketing
activities undertaken by retailers. The model is most directly generalizable to frequently
occurring marketing events because of the need for data. However, effects of larger
changes could be evaluated ex-post by using statistical methods, or ex-ante, by using the
model to calculate the profit impact of the change, given the managers' assunlptions.
The model may also be generalized to other settings than grocery retailing. The model
should be most relevant for situations where the customer buys a basket of products, such
as for example, department stores, fast food restaurants, and banks.
The numerical values of the response functions are not supposed to be generalized.
First, the sales response may differ across stores, product categories, and items. Second, the
product categories and items were systematically, rather than randomly, chosen. Third,
scanner data makes such generalizations unnecessary, as each retailer should use their own
data to estimate the models. This should preferably be implemented as an ongoing tracking
and experimentation system, re-estimating the models frequently.
11.4. Limitations
The research reported here has a number of limitations that should be noted. First, the
model and method presented seek to increment profits rather than pursue profit
maximization. If there has been little variation in the controllable variables, the model
works most reliably when small changes from a reference level are considered when
- 293-
making predictions based on historical data. The model might not cover the global
optimum when it is parameterized with historical data. Over time, the model needs to be
re-estimated. The path mayor may not lead to a global optimum, but the lack of a
guarantee that the global optimum will be reached should not prevent us from seeking
improvement.
Second, the proposed model does not consider simultaneous optimization of all
products' promotions. That would be too difficult a task. The model is in any case an
improvement from current pratice, which only regards sales or possibly gross margin of the
promoted item. The model focuses the attention on the consequences for the store gross
nlargin of sales promotion for one item or a group of items.
A third limitation of the model is that it is a one-period model, considering only current
profits. The general model can easily be generalized to a multi-period setting, but would be
rather difficult to estimate empirically. It may be that the indirect modeling approach used
here would hamper such multiple-period modeling.
Fourth, it is assumed that the nlargins of other items in the focal category are
homogeneous or that the effect on other items sales are proportional to their sales.
Similarly, margins of other categories are assumed to be homogeneous or to have a sales
effect on other categories which is proportional to their sales. Although these are
reasonable approximations that make the model parsimonious, they constitute limitations.
However, as noted in the model development, the model can be extended to accommodate
groups of items with different margins or which are differently affected by the promotion,
or both. Such groups may be groups of items within the focal category, or groups of other
categories. An interesting way to extend the model would be to use the hierarchical
attraction models developed by Foekens et ale (1992) in order to model sales of all items
within the focal category. Their approach could also be used to model asymmetries in the
inter-category cannibalization.
11.5. Suggestions for Further Research

In addition to the possibilities for improvements of the model that the above-identified
limitations of the present research indicate, future research in the following areas would be
valuable.
Future research may use the models developed here to study the promotion profits for
different types of products. One typology that appears to be useful for such research was
suggested by Pellegrini (1989). Pellegrini makes a distinction between type A and type B
products, where type A products can have an impact on the store choice while type B
products have no impact on the store choice and are simply bought in the store visited.
- 294-
Such studies would also be important in order to examine whether the effects are
homogeneous across categories and stores. If the effects are fairly homogeneous across
some dimension, knowledge could be accumulated and generalized to other stores and
other products. That would be very useful for theory development and for retail-chain
management.
It would also be interesting to investigate the effects of different types of sales
promotion, for example, whether the consumers meet the promotion in-store or out-of-
store, and whether the promotion gives the consumers a direct incentive to buy or if it only
serves to remind the consumers of the product. This may affect the sales response pattern
and the profits of sales promotion.
The deterministic-response models used here could be replaced by stochastic-response
models to simulate the variance of promotion profits. This may be interesting for the
retailer for two reasons. First, a smaller average profit may be preferable to a large average
profit when the variance of the latter is high (though the retailer may spread the risks).
Second, the profit of the average response is in general different from the average profit
from the stochastic response. This is because profit is not a linear function of response.
Stochastic simulation of sales makes inventory costs interesting to model.
It would also be interesting to combine the models developed here with panel data. The
linkages between individual and aggregate behavior could then be studied. Alternatively,
the individual behavior may be simulated and the aggregate outcome of the simulated
behavior could be compared with the aggregate outcome as measured with receipt data.
Future research could also adapt the model to better correspond to a chain-store
perspective. That would mean better modeling of the store choice and should include
cannibalization among stores.
Another way that the models developed here could be used in future research is as input
in models of trade deal profits. That may lead to game-theoretic models of manufacturer
and retailer behavior.
A further area for future research is the examination of other functional forms of the
response models than the logit and exponential models used in this study. Research
investigating alternative estimation procedures and techniques would also be of interest.
11.6. Managerial Implications

The most important managerial implication of the present research is that retail
managers must be aware of the interdependencies among the many items in retail
assortments. This means that it should be important to develop and install systems that can
collect and process the relevant data.
- 295-
This research showed the importance of store traffic for sales and profits. It appears that
retail managers should preferably install point-of-sales systems that can collect receipt data.
If receipt data is not available to the retailer, the item sales data should at least be
complemented with the number of receipts produced by the scanner system. This figure
would be available from most cash registers.
For tracking and marketing control purposes, it is important that store traffic is explicitly
modeled. For example, the responsibility for insufficient sales of a promoted product
should not fall on the given category if the reason for the low sales is that store traffic was
low. The system of models developed in this report could be the base for such a tracking
and control system.
A furthert implication of the present research is that there may be problems associated
with the trend towards category management and the store-in-the-store management
principle where the store's categories are run as profit centers and employees receive
bonuses tied to category performance. Category management is based on the assumption
that the sales of different product categories are independent. When this assumption does
not hold, category management will be sub-optimal. If the category is important for store
traffic, it is likely that the store's category managers will use less promotions than optimal
for the store. On the other hand, inter-category cannibalization implies that a store run as a
portfolio of category profit centers will over-use sales promotions. The proposed model
should help the retailers to measure the profit impact of sales promotion, thus giving them
the possibility to improve the design of promotions and a way to determine the sub-
optimization cost, if any, of category nlanagement.
Measurement of the profit impact of sales promotion should also be important for
retailer who has other objectives than maximizing profits from the current period's
promotions. Such retailers will find the model useful for estimating the opportunity cost of
the chosen promotion, and possibly for avoiding unnecessarily costly sales promotions.
Retailers who do not have access to the necessary data could use the model framework
presented in this report as a decision-calculus model. The decision calculus methodology
developed by Little (e.g., Little 1975b) starts with a model parameterized using judgmental
data. Decision calculus provides a consistent method to calculate the implications of the
assumptions managers make. The assumptions are made explicit which makes comparisons
between alternative plans easier and more objective. The sensitivity to changes in
assumptions and decision variables can be calculated.
Problem understanding and learning may be improved by using decision calculus
models and judgment of multiple managers can be integrated. Lodish (1982) reports that
the decision-support system for retailers that he developed using decision-calculus
nlethodology led to new marketing research projects to get better estimates of the response
- 296-
functions. For a discussion of decision calculus and managerial judgment, see also
Chakravarti, Mitchell, and Staelin (1979, 1981) and Little and Lodish (1981).
The decision-calculus usage of the promotion-profit model may be important for pre-
testing sales pronlotions and other marketing activities that have not been used before.
Since historical evidence of the sales response to such activities does not exist, scanner
data would not help the retailer. Judgmental data is then the best available alternative. That
may also be the situation for the manufacturers, who may not have direct access to retail-
level scanner data. However, manufacturers would do better to forge alliances with retailers
to gain access to retail level receipt data. Last but not least, this research shows that
manufacturers should use retailer-oriented data and models to understand the needs of their
immediate customers.
- 297-
REFERENCES
Aaker, David A. (1973), "Toward a Normative Model of Promotional Decision Making,"

Management Science, 19 (6), 593-603.
Abraham, Magid M. and Leonard M. Lodish (1987), "PROMOTER: An Automated

Promotion Evaluation System," Marketing Science, 6 (Spring), 101-123.
Abraham, Magid M. and Leonard M. Lodish (1990), "Getting the Most Out of Advertising
and Promotion," Harvard Business Review, May-June 1990,50-60.
Abraham, Magid M. and Leonard M. Lodish (1993), "An Implemented Systen1 for
Improving Promotion Productivity Using Store Scanner Data," Marketing Science,
12 (Summer), 248-269.
Achabal, Dale D., Shelby McIntyre, and Stephen A. Smith (1990), "Maximizing Profits
from Periodic Department Store Promotions," Journal ofRetailing, 66 (Winter), 383-
407.
Ahn, Kwang Ho and Avijit Ghosh (1989), "Hierarchical Models of Store Choice,"
International Journal of Retailing, 4 (5), 39-52.
Aldrich, John H. and Forrest D. Nelson (1984), "Linear Probability, Logit, and Probit
Models," Sage University Paper series on Quantitative Applications in the Social
Sciences, 07-045, Beverly Hills and London: Sage Publications.
Allenby, Greg M. and Peter E. Rossi (1991), "Quality Perceptions and Asymmetric
Switching Between Brands," Marketing Science, 10 (Summer), 185-204.
Bass, Frank M. (1974), "The Theory of Stochastic Preference and Brand Switching,"
Journal ofMarketing Research, 11 (February), 1-20.
Bass, Frank M., Morshe M. Givon, Manohar U. Kalwani, David Reibstein, and Gordon P.
Wright (1984), "An Investigation Into the Order of the Brand Choice Process,"
Marketing Science, 3 (Fall), 267-87.
Bawa, Kapil (1990), "Modeling Inertia and Variety Seeking Tendencies In Brand Choice
Behavior," Marketing Science, 9 (Summer), 263-78.
- 299-
Bawa, Kapil, Aradhna Krishna, and Jane T. Landwehr (1989), "Consumer Response to
Retailers' Marketing Environments: An Analysis of Coffee Purchase Data," Journal
ofRetailing, 65 (Winter), 471-495.
Beckwith, Neil E. (1973), "Concerning the Logical Consistency of Multivariate Market

Share Models," Journal ofMarketing Research, 10 (August), 341-4.
Bemmaor, Albert C. and Dominique Mouchoux (1991), "Measuring the Short-Term Effect
of In-Store Pronl0tion and Retail Advertising on Brand Sales: A Factorial
Experiment," Journal ofMarketing Research, 28 (May), 202-14.
Benishay, Haskel (1965), "A Disaggregative Model for the Generation of Sales," Journal
ofMarketing Research, 2 (February), 74-79.
Blattberg, Robert C., Gary D. Eppen, and Joshua Lieberman (1981), "A Theoretical and
Empirical Evaluation of Price Deals for Consumer Nondurables," Journal of
Marketing, 45 (Winter), 116-129.
Blattberg, Robert C. and Stephen J. Hoch (1990), "Database Models and Managerial
Intuition: 50% Model + 50% Manager," Management Science, 36 (August), 887-899.
Blattberg, Robert C. and Alan Levin (1987), "Modelling the Effectiveness and Profitability
of Trade Promotions," Marketing Science, 6 (Spring), 124-146.
Blattberg, Robert C. and Scott A. Neslin (1990), Sales Promotion: Concepts, Methods, and
Strategies, Englewood Cliffs, NJ: Prentice Hall.
Blattberg, Robert C. and Kenneth J. Wisniewski (1989), "Price-Induced Patterns of

Competition,," Marketing Science, 8 (Fall), 291-309.
Bolton, Ruth N. (1989), "The Relationship Between Market Characteristics and

Promotional Price Elasticities," Marketing Science, 8 (Spring), 153-69.
Bolton, Ruth N. (1989), "The Robustness of Retail-Level Price Elasticity Estimates,"

Journal ofRetailing, 65 (Summer), 193-219.
Brodie, Roderick and Comelis A. de Kluyver (1984), "Attraction Versus Linear and
Multiplicative Market Share Models: An Empirical Evaluation," Journal of
Marketing Research, 21 (May), 194-201.
Bucklin, Randolph E. and Sunil Gupta (1992), "Brand Choice, Purchase Incidence, and
Segmentation: An Integrated Modeling Approach," Journal of Marketing Research,
29 (May), 201-15.
Bucklin, Randolph E. and James M. Lattin (1991), "A Two-State Model of Purchase
Incidence and Brand Choice," Marketing Science, 10 (Winter), 24-39.
- 300-
Bucklin, Randolph E. and James M. Lattin (1992), "A Model of Product Category
Competition Among Grocery Retailers," Journal ofRetailing, 68 (Fall), 271-293.
Bultez, Alain and Philippe Naert (1988), "SH.A.R.P: Shelf Allocation for Retailers' Profit,"
Marketing Science, 7 (Summer), 211-231.
Bultez, Alain, Els Gijsbrechts, Philippe Naert, and Piet Vanden Abeele (1989),
"Asymmetric Cannibalism in Retail Assortments," Journal of Retailing, 66
(Summer),153-192.
Chakravarti, Dipankar, Andrew Mitchell, and Richard Staelin (1979), "Judgment Based
Marketing Decision Models: An Experimental Investigation of the Decision Calculus
Approach," Management Science, 25 (3),251-263.
Chakravarti, Dipankar, Andrew Mitchell, and Richard Staelin (1981), "Judgment Based
Marketing Models: Problems and Possible Solutions," Journal of Marketing, 45
(fall), 13-23.
Chapman, Randall G. (1986), "Assessing the Profitability of Retailer Couponing with a

Low-Cost Field Experiment," Journal ofRetailing, 62 (Spring), 19-40.
Chevalier, Michel (1975), "Increase in Sales Due to In-Store Display," Journal of

Marketing Research, 12 (November), 426-31.
Chevalier, Michel (1975), "Substitution Patterns as a Result of Display in the Product

Category," Journal ofRetailing, 51 (Winter), 65-72.
Chevalier, Michael and R.C. Curhan (1976), "Retail Promotion as a Function of Trade
Promotion: A Descriptive Analysis," Sloan Management Review, 18 (Fall), 19-32.
Chiang, Jeongwen (1991), "A Simultaneous Approach to the Whether, What and How
Much to Buy Questions," Marketing Science, 10 (Fall), 297-315.
Chou, Ya-Lun (1984), Statistical Analysis, second edition, London: Holt, Rinehart and
Winston.
Chow, Gregory C. (1960), "Test of Equality Between Sets of Coefficients in Two Linear
Regressions," Econometrica, 28 (3), 591-605.
Cooper, Lee G. and Masao Nakanishi (1988), Market-Share Analysis: Evaluating

Competitive Marketing Effectiveness, Boston, Dordrecht, London: Kluwer Academic
Publishers.
Corstjens, Marcel L. and David A. Gautschi (1983), "Formal Choice Models in

Marketing," Marketing Science, 2 (1), 19-56.
- 301 -
Cox, Anthony D. and Dena Cox (1990), "Competing on Price: The Role of Retail Price
Advertisements in Shaping Store-Price Image," Journal of Retailing, 66 (Winter),
428-45.
Craig, C. Samuel, Avijit Ghosh, and Sara McLafferty (1984), "Models of the Retail
Location Process: A Review," Journal of Retailing, 60 (Spring), 5-36.
Curhan, Ronald C. (1974), "The Effects of Merchandising and Temporary Promotional

Activities on the Sales of Fresh Fruits and Vegetables in Supermarkets," Journal of
Marketing Research, 11 (August), 286-294.
Currim, Imran and Linda G. Schneider (1991), "A Taxonomy of Consumer Purchase
Strategies in a Promotion Intensive Environment," Marketing Science, 10 (Spring),
91-109.
Curry, David J. (1989), "Single Source Systems: Retail Management Present and Future,"
Journal of Retailing, 65 (Spring), 1-20.
Curry, David J. (1993), The New Marketing Research Systems: How to Use Strategic
Database Information for Better Marketing Decisions, New York, NY: John Wiley
& Sons, Inc.
Dalal, S.R., J.C. Lee, and D.J. Sabavala (1984), "Prediction of Individual Buying Behavior:
A Poisson-Bernoulli Model With Arbitrary Heterogeneity," Marketing Science, 3
(Fall),352-67.
Dhebar, Anirudh, Scott A. Neslin, and John A. Quelch (1987), "Developing Models for
Planning Retailer Sales Promotions: An Application to Automobile Dealerships,"
Journal ofRetailing, 63 (Winter), 333-364.
Dodson, Joe A., Alice M. Tybout, and Brian Sternthal (1978), "Impact of Deals and Deal
Retraction on Brand Switching," Journal of Marketing Research, 15 (February), 72-
81.
Doyle, Peter and Zeki Gidengil (1977), "A Review of In-store Experiments," Journal of
Retailing, 53 (Summer), 47-62.
Doyle, Peter and John Saunders (1985), "The Lead Effect of Marketing Decisions,"
Journal ofMarketing Research, 22 (February), 54-65.
Doyle, Peter and John Saunders (1990), "Multiproduct Advertising Budgeting," Marketing
Science, 9 (Spring), 97-113.
Engel, James F., Roger D. Blackwell, and Paul W. Miniard (1986), Consumer Behavior,
fifth edition, New York, NY: The Dryden Press.
- 302-
Eppen, Gary D. and Floyd Jerome Gould (1985), Quantitative Concepts for Management:
Decision Making without Algorithms, 2nd edition, Englewood Cliffs, N.J.: Prentice-
Hall Inc.
Feichtinger, Gustav, Alfred Luhmer, and Gerhard Sorger (1988), "Optimal Price and
Advertising Policy for a Convenience Goods Retailer," Marketing Science, 7
(Spring), 187-201.
Foekens, Eijte W., Peter S.H. Leeflang, and Dick R. Wittink (1992), "Asymmetric Market
Share Modeling in a Market with Many Items," working paper, European Institute
for Advanced Studies in Management, Brussels.
Foekens, Eijte, Peter S.H. Leeflang, and Dick R. Wittink (1994), "A Comparison and an
Exploration of the Forecasting Accuracy of Nonlinear Models at Different Levels of
Aggregation," International Journal ofForecasting, forthcoming.
Fotheringham, A. Stewart (1988), "Consumer Store Choice and Choice Set Definition,"
Marketing Science, 7 (Summer), 299-310.
Fox, John (1991), "Regression Diagnostics," Sage University Paper series on Quantitative
Applications in the Social Sciences, 07-079, Beverly Hills and London: Sage
Publications.
Ghosh, Avijit, Scott Neslin, and Robert Shoemaker (1984), "A Comparison of Market
Share Models and Estimation Procedures," Journal of Marketing Research, 21
(May),201-210.
Gijsbrechts, Els (1993), "Prices and Pricing Research in Consumer Marketing: Some
Recent Developments," International Journal of Research in Marketing, 10, 115-
151.
Givon, Morshe (1984), "Variety Seeking Through Brand Switching," Marketing Science, 3
(Winter), 1-22.
Grover, Rajiv and V. Srinivasan (1992), "Evaluating the Multiple Effects of Retail
Promotions on Brand Loyalty and Brand Switching Segments," Journal of Marketing
Research, 29 (February), 76-89.
Guadagni, Peter M. and John D. C. Little (1983), "A Logit Model of Brand Choice
Calibrated on Scanner Data," Marketing Science, 2 (Summer), 203-238.
Gupta, Sunil (1988), "Impact of Sales Promotions on When, What, and How Much to
Buy," Journal ofMarketing Research, 25 (November), 342-55.
- 303-
Gupta, Sunil (1991), "Stochastic Models of Interpurchase Time With Time-Dependent
Covariates," Journal ofMarketing Research, 28 (February), 1-15.
Gupta, Sunil and Donald G. Morrison (1991), "Estimating Heterogeneity in Consumers'

Purchase Rates," Marketing Science, 10 (Summer), 264-269.
Hanssens, Dominique M., Leonard J. Parsons, and Randall L. Schultz (1990), Market
Response Models: Econometric and Time Series Analysis, International Series in
Quantitative Marketing, BostonIDordrecht/London: Kluwer Academic Publishers.
Hardy, Kenneth (1986), "Key Success Factors for Manufacturer Sales Promotions in
Packaged Goods," Journal ofMarketing, 50 (July), 13-23.
Helsen, Kristiaan and David C. Schnlittlein (1992), "How Does a Product Market's Typical
Price-Promotion Pattern Affect the Timing of Households' Purchases? An Empirical
Study Using UPC Scanner Data," Journal ofRetailing, 68 (Fall), 316-340.
ReIsen, Kristiaan and David C. Schmittlein (1993), "Analyzing Duration Times in

Marketing: Evidence for the Effectiveness of Hazard Rate Models," Marketing
Science, 11 (Fall), 395-414.
Hemant, Mikael (1992), "Projekt aprilkvitto: Dokumentation 2," working paper, ICA
Retail, Vastenls.
Hernant, Mikael and Per-Goran Persson (1994), "In Search of the Deal Prone Consumer: A
study of consumers' usage of sales promotions introducing a new type of single
source data," conference paper presented at the 23rd EMAC Conference in
Maastricht 1994.
Hoch, Stephen J., Xavier Dreze, and Mary E. Purk (1994), "EDLP, Hi-Lo, and Margin
Arithmetic," Journal of Marketing, 58 (October), 16-27.
Hoofnagle, William S. (1965), "Experimental Designs in Measuring the effectiveness of

Promotion," Journal ofMarketing Research, 2 (May), 154-162.
Holmberg, Carina (1992), "Effects of Feature Advertising and other Promotions: A Study
of Swedish Grocery Retailing Using Scanner Data," Research Report, Stockholm
School of Economics, Stockholm: EFI.
Inman, J. Jeffrey and Leigh McAlister (1993), "A Retailer Promotion Policy Model
Considering Promotion Signal Sensitivity," Marketing Science, 12 (Fall), 339-356.
Inman, J. Jeffrey, Leigh McAlister, and Wayne D. Hoyer (1990), "Promotion Signal: Proxy
for a Price Cut?," Journal of Consumer Research, 17 (June), 74-81.
- 304-
Jacobson, Robert and Carl Obermiller (1990), "The Formation of Expected Future Price: A
Reference Price for Forward-Looking Consumers," Journal of Consumer Research,
16 (March), 420-432.
Jones, John Philip (1990), "The Double Jeopardy of Sales Promotions," Harvard Business
Review, Septenlber-October 1990, 145-152.
Julander, Claes-Robert (1984), Extrapriser och veckoannonser: Effekter av

dagligvaruhandelns SA/VA-aktiviteter, Stockholm: EFI.
Julander, Claes-Robert (1987), "Improved In-Store Marketing from Technology Sourced

Data," paper presented at the Spectra USEDATA '87 conference in Berkshire,
England.
Julander, Claes-Robert (1989), "What's In the Basket? Receipt Analysis for Marketing
Decisions: An Exploratory Study," working paper, Foundation for Distribution
Research, Stockholm School of Economics.
Julander, Claes-Robert (1992), "Basket Analysis: A New Way of Analysing Scanner Data,"
International Journal ofRetail & Distribution Management, 20 (7), 10-18.
Julander, Claes-Robert and Carina Holmberg (1987), "Marknadsinformation fran

datakassor: Projektets uppHiggning samt exempel pa resultat," Research Report,
Stockholm School of Economics, EFI.
Julander, Claes-Robert and Per-Goran Persson (1992), "Towards a Theory of Promotional

Effects in Grocery Retailing," conference paper presented at the IAREP/GEW joint
Conference in Frankfurt 1992, Research Paper #6498, Economic Research Institute,
Stockholm School of Economics.
Kahn, Barbara E. and Donald R. Lehmann (1991), "Modeling Choice Among

Assortments," Journal ofRetailing, 67 (Fall), 274-299.
Kalwani, Manohar U. and Chi Kin Yim (1992), "Consumer Price and Promotion
Expectations: An Empirical Study," Journal of Marketing Research, 29 (February),
90-100.
Keng, Kau Ah and A.S.C. Ehrenberg (1984), "Patterns of Store Choice," Journal of
Kennedy, Peter (1992), A Guide to Econometrics, 3rd ed., Oxford: Basil Blackwell Ltd.
Kleinbaum, David G., Lawrence L. Kupper, and Keith E. Muller (1988), Applied
Regression Analysis and Other Multivariable Methods," Boston, MA: PWS-KENT
Publishing Company.
- 305-
Kotler, Philip (1991), Marketing Management: Analysis, Planning, Implementation, and
Control, seventh edition, Englewood Cliffs, NJ: Prentice-Hall Inc.
Krishna, Aradhna (1991), "Effect of Dealing Patterns on Consumer Perceptions of Deal

Frequency and Willingness to Pay," Journal of Marketing Research, 28 (November),
441-51.
Krishna, Aradhna (1994), "The Effect of Deal Knowledge on Consumer Purchase

Behavior," Journal ofMarketing Research, 31 (February), 76-91.
Krishna, Aradhna, Imran S. Currim, and Robert W. Shoemaker (1991), "COnSUTIler

Perceptions of Promotional Activity," Journal ofMarketing, 55 (April), 4-16.
Krishnamurthi, Lakshman and S.P. Raj (1988), "A Model of Brand Choice and Purchase
Quantity Price Sensitivities," Marketing Science, 7 (Winter), 1-20.
Krishnamurthi, Lakshman and S.P. Raj (1991), "An Empirical Analysis of the Relationship
between Brand Loyalty and Consumer Price Elasticity," Marketing Science, 10
(Spring),172-183.
Krishnamurthi, Lakshman, Tridib Mazumdar, and S.P. Raj (1992), "Asymmetric Response
to Price in Consumer Brand Choice and Purchase Quantity Decisions," Journal of
Consumer Research, 19 (December), 387-400.
Kumar, V. and Robert P. Leone (1988), "Measuring the Effect of Retail Store Promotions
an Brand and Store Substitution," Journal ofMarketing Research, 25 (May), 178-85.
Lal, Rajiv (1990a), "Manufacturer Trade Deals and Retail Price Promotions," Journal of
Lal, Rajiv (1990b), "Price Promotions: Limiting Competitive Encroachment," Marketing

Science, 9 (Summer), 247-62.
Lattin, James M. (1987), "A Model of Balanced Choice Behavior," Marketing Science, 6
(Winter),48-65.
Lattin, James M. and Randolph E. Bucklin (1989), "Reference Effects of Price and
Promotion on Brand Choice Behavior," Journal ofMarketing Research, 26 (August),
299-310.
Lazear, Edward P. (1986), "Retail Pricing and Clearance Sales," The American Economic
Review, 76 (March), 14-32.
Leeflang, Peter S.H. and Jan C. Reuyl (1984), "On the Predictive Power of Market Share
Attraction Models," Journal ofMarketing Research, 21 (May), 211-5.
- 306-
Lilien, Gary L. (1993), "Marketing Models: Past, Present and Future," in Laurent, Lilien,
and Pras (eds.), Research Traditions in Marketing, BostonIDordrecht/London:
Kluwer Academic Publishers, pp. 1-19.
Little, John D.C. (1975a), "BRANDAID: A Marketing-Mix Model, Part 2: Structure,"

Operations Research, 23 (July-August), 628-655.
Little, John D.C. (1975b), "BRANDAID: A Marketing-Mix Model, Part 2:

Implementation, Calibration, and Case Study," Operations Research, 23 (July-
August), 656-673.
Little, John D.C. and Leonard M. Lodish (1981), "Commentary on "Judgment Based
Marketing Decision Models"," Journal ofMarketing, 45 (Fall), 24-29.
Lodish, Leonard M. (1982), "A Marketing Decision Support System for Retailers,"
Marketing Science, 1 (Winter), 31-56.
Louviere, Jordan J. and Gary J. Gaeth (1987), "Decomposing the Determinants of Retail
Facility Choice Using the Method of Hierarchical Information Integration: A
Supermarket Illustration," Journal ofRetailing, 63 (Spring), 25-48.
Malhotra, Naresh K. (1986), "Modeling Store Choice Based on Censored Preference Data,"
Mason, Charlotte H. (1990), "New Product Entries and Product Class Demand," Marketing
Science, 9 (Winter), 58-73.
Mazursky, David and Jacob Jacoby (1986), "Exploring the Development of Store Images,"
McCann, John M. and John P. Gallagher (1991), Databases and Knowledge Systems in
Merchandising, New York, NY: Van Nostrand Reinhold.
McCann, John, Ali Tadlaoui, and John Gallagher (1990), "Knowledge Systems in
Merchandising: Advertising Design," Journal ofRetailing, 66 (Fall), 257-277.
McGuire, Timothy W. and Doyle L. Weiss (1976), "Logically Consistent Market Share
Models II," Journal ofMarketing Research, 13 (August), 296-302.
Meyer, Robert J. and Joao Assuncao (1990), "The Optimality of Consumer Stockpiling
Strategies," Marketing Science, 9 (Winter), 18-41.
Moore, David J. and Richard W. Olshavsky (1989), "Brand Choice and Deep Price
Discounts," Psychology & Marketing, 6 (3), 181-196.
- 307-
Mowen, John C., Joshua Wiener and Clifford Young (1990), "Consumer Store Choice and
Sales Taxes: Retailing, Public Policy, and Theoretical Implications," Journal of
Retailing, 66 (Summer), 222-242.
Mulhern, Francis J. and Robert P. Leone (1990), "Retail Promotional Advertising: Do the
Number of Deal Items and Size of Deal Discounts Affect Store Performance?,"
Journal ofBusiness Research, 21, 179-194.
Mulhern, Francis J. and Robert P. Leone (1991), "Implicit Price Bundling of Retail
Products: A Multiproduct Approach to Maximizing Store Profitability," Journal of
Marketing, 55 (October), 63-76.
Naert, Philippe A. and Alain Bultez (1973), "Logically Consistent Market Share Models,"
Journal ofMarketing Research, 10 (August), 334-40.
Naert, Philippe A. and Peter S.H. Leeflang (1978), Building Implementable Marketing
Models, Leiden and Boston: Martinus Nijhoff Social Sciences Division.
Naert, Philippe and Marcel Weverbergh (1981), "On the Prediction Power of Market Share
Attraction Models," Journal ofMarketing Research, 18 (May), 146-53.
Naert, Philippe and Marcel Weverbergh (1985), "Market Share Specification, Estimation,
and Validation: Toward Reconciling Seemingly Divergent Views," Journal of
Nakanishi, Masao and Lee G. Cooper (1974), "Parameter Estimation for a Multiplicative
Competitive Interaction Model - Least Squares Approach," Journal of Marketing
Research, 11 (August), 303-11.
Nakanishi, Masao, and Lee G. Cooper (1982), "Simplified Estimation Procedures for MCI
Models," MarketingScience, 1 (Summer), 314-322.
Narasimhan, Chakravarthi (1984), "A Price Discrimination Theory of Coupons,"

Marketing Science, 3 (Spring), 128-47.
Neslin, Scott A. (1990), "A Market Response Model for Coupon Promotions," Marketing
Science, 9 (Spring), 125-145.
Neslin, Scott A., Caroline Henderson, and John Quelch (1985), "Consumer Promotions and
the Acceleration of Product Purchases," Marketing Science, 4 (Spring), 147-65.
Neslin, Scott A. and Robert W. Shoemaker (1983), "A Model for Evaluating the
Profitability of Coupon Promotions, Marketing Science, 2 (Fall), 361-388.
It
- 308-
Ostrom, Charles W. Jr. (1978), "Tinle Series Analysis: Regression Techniques," Sage
University Paper series on Quantitative Applications in the Social Sciences, 07-009,
Beverly Hills and London: Sage Publications.
Parssons, Leonard J., Els Gijsbrechts, Peter S.H. Leeflang, and Dick R. Wittink (1993),
"Marketing Science, Econometrics, and Managerial Contributions," in Laurent,
Lilien, and Pras (eds.), Research Traditions in Marketing, BostonIDordrecht/London:
Kluwer Academic Publishers, pp. 52-78.
Pellegrini, Luca (1989), "Consumers' behaviour and producer-distributor relationships in

convenience goods markets," in Pellegrini, L. and S.K. Reddy, Retail and Marketing
Channels, Routledge.
Persson, Per-Goran (1992), "Basket Analysis: A New Way of Studying Short Term Effects
of Promotions in Grocery Retailing," Research Report #356, Economic Research
Institute, Stockholm School of Economics (licentiate dissertation).
Persson, Per-Goran (1993), "Promotional Response Variation across Product Categories,"

Research Paper # 6529, Economic Research Institute, Stockholm School of
Economics.
Persson, Per-Goran (1994), "Decomposing Promotional Elasticity," Research Paper #6530,

Economic Research Institute, Stockholm School of Economics.
Quelch, John A. (1983), "It's Time to make Trade Promotions more Productive," Harvard
Business Review, (May-June), 130-136.
Raju, Jagmohan (1992), "The Effect of Price Promotions on Variability in Product

Category Sales," Marketing Science, 11 (Summer), 207-220.
Rao, Vithala R. (1984), "Pricing Research in Marketing: The State of the Art," Journal of
Business, 57, s39-s60.
Ratneshwar, S. and Allan D. Shocker (1991), "Substitution in Use and the Role of Usage
Context in Product Category Structures," Journal of Marketing Research, 28
(August),281-95.
Reibstein, David J. and Hubert Gatignon (1984), "Optimal Product Line Pricing: The
Influence of Elasticities and Cross-Elasticities," Journal of Marketing Research, 21
(August),259-67.
Rossiter, John R. and Larry Percy (1987), Advertising & Promotion Management, New
York, NY: McGraw Hill International Editions.
- 309-
Salop, Steven and Joseph Stiglitz (1977), "Bargains and Ripoffs: A Model of
Monopolistically Competitive Price Dispersion," Review of Economic Studies, 44,
493-510.
Salop, Steven and Joseph E. Stiglitz (1982), "The Theory of Sales: A Simple Model of
Equilibrium Price Dispersion with Identical Agents," The American Economic
Review, 72 (5), 1121-1130.
Steenkamp, Jan-Benedict and Michel Wedel (1991), "Segmenting Retail Markets on Store
Image Using a Consunler-Based Methodology," Journal of Retailing, 67 (Fall), 300-
320.
Strang, Roger A (1976), "Sales Promotion-Fast Growth, Faulty Management," Harvard

Business Review, July-August, 115-124.
The Economist (1992), "Procter's Gamble," July 25th, 59-60.
Uncles, M.D. and A.S.C. Ehrenberg (1990), "The Buying of Packaged Goods at US Retail
Chains," Journal ofRetailing, 66 (Fall), 278-296.
Vilcassim, Naufel J. and Dipak C. Jain (1991), "Modeling Purchase-Timing and Brand-
Switching Behavior Incorporating Explanatory Variables and Unobserved
Heterogeneity," Journal ofMarketing Research, 28 (February), 29-41.
Varian, Hal R. (1980), "A Model of Sales," The American Economic Review, 70
(September),651-659.
Varian, Hal R. (1984), Microeconomic Analysis, 2nd ed., New York, N.Y.: W.W. Norton
& Company Inc.
Wagner, Udo and Alfred Taudes (1986), "A Multivariate Polya Model of Brand Choice
and Purchase Incidence," Marketing Science, 5 (Summer), 219-44.
Walters, Rockney G. (1988), "Retail Promotions and Retail Store Performance: A Test of
Some Key Hypotheses," Journal ofRetailing, 64 (Summer), 153-180.
Walters, Rockney G. (1989),"An Empirical Investigation into Retailer Response to

Manufacturer Trade Promotions," Journal of Retailing, 65 (Summer), 253-272.
Walters, Rockney G. (1991), "Assessing the Impact of Retail Price Promotions on Product
Substitution, Complementary Purchase, and Interstore Sales Displacement," Journal
ofMarketing, 55 (April), 17-28.
Walters, Rockney G. and Scott B. MacKenzie (1988), "A Structural Equations Analysis of
the Impact of Price Promotions on Store Performance," Journal of Marketing
Research, 25 (February), 51-63.
- 310-
Walters, Rockney G. and Heikki J. Rinne (1986), "An Empirical Investigation into the
Impact of Price Promotions on Retail Store Pelformance," Journal of Retailing, 62
(Fall),237-266.
Wernerfelt, Birger (1991), "Brand Loyalty and Market Equilibrium," Marketing Science,
10 (Summer), 229-245.
Wheat, Rita D. and Donald G. Morrison (1990), "Assessing Purchase Timing Models:
Whether or Not Is Preferable to When," Marketing Science, 9 (Spring), 162-70.
Wilkinson, J.B., J. Barry Mason, and Christie H. Paksoy (1982), "Assessing the Impact of
Short-Term Supermarket Strategy Variables," Journal of Marketing Research, 19
(February), 72-86.
Wilkinson, J.B., Christie H. Paksoy, and J. Barry Mason (1981), "A Demand Analysis of
Newspaper Advertising and Changes in Space Allocation," Journal of Retailing, 57
(Summer), 30-48.
Winer, Russel S. (1986), "A Reference Price Model of Brand Choice for Frequently
Purchased Products," Journal of Consumer Research, 13 (September), 250-256.
Wittink, Dick R. (1987), "Causal Market Share Models in Marketing: Neither Forecasting
nor Understanding?," International Journal ofForecasting, 3, 445-448.
Wittink, Dick R., Michael J. Addona, William J. Hawkes, and John C. Porter (1987),
"SCAN*PRO©: A Model to Measure Short-Term Effects of Promotional Activities
on Brand Sales Based on Store Level Scanner Data," Working paper, Johnson
Graduate School of Management, Cornell University, Ithaca, N.Y.
Zenor, Michael J. (1994), "The Profit Benefits of Category Management," Journal of

Marketing Research, 31 (May), 202-213.
- 311 -
The Foundation for Distribution Research
The Foundation for Distribution Research at the Stockholm School of Economics was
founded in 1981 to address the long-term needs of students, academicians, industry and
society for research and education concerning issues of trade and distribution. Specifically,
the Foundation for Distribution Research covers these areas:
* Research programs dealing with issues of trade and distribution to offer

post-graduate students exciting research opportunities.
* Research on issues of vital importance to the distribution of goods and
services.
* Courses at the Stockholm School of Economics related to issues of trade
and distribution.
* Dissen1ination of results from research with significance for the distribution
of goods and services.
* Continuous exchange of research ideas and results with researchers in
Sweden, other European countries and the u.S.
The Foundation for Distribution Research is connected to the Economic Research Institute
(EFI) at the Stockholm School of Economics. The Foundation is supported by leading
Swedish corporations engaged in the distribution of goods and services, manufacturers as
well as retailers and wholesalers. Sponsors are invited semiannually to share results from
the Foundation's research, teaching and other activities. In this way the Foundation
manages to conduct an ongoing stimulating dialogue with the participating companies,
activities that also help gathering ideas for future projects.
The sponsors of the Foundation for Distribution Research are:

ADA AB Kooperativa Forbundet
Ahlsell AB Kungsornen AB
Apoteksbolaget AB Procordia Food & Beverages AB
ARLA ekonomisk forening Svenska Nestle AB
DAGAB AB, Svenska Tobaks AB
DLF, Dagligvaruleverantorers Forbund Svenska Unilever AB
ICAAB
Foundation for Distribution Research
THE ECONOMIC RESEARCH INSTITUTE AT THE
STOCKHOLM SCHOOL OF ECONOMICS
POBox 6501, 5-113 83 STOCKHOLM. Tel +46-8-7369000 • Fax +46-8-339489
Publications from the Foundation for

Distribution Research, April 1995
Systembolaget och dess kunder - ett ekonomisk psykologi om hushallens

diskussionsunderlag for att bedoma finansiella beteenden.
konsekvenser vid en eventuell avreglering R Wahlund & J Gunnarsson.
av detaljhandeln med alkoholhaltiga Foundation for Distribution Research
drycker. Working Paper Series, Stockholm, 1994.
P Hakansson & M Soderlund
EFI Research Report, Stockholm, 1995.
Den nyckfulle Svensson - reflektioner efter
ett seminarium om strategier for handeln.
Hushallens finansiella strategier - en N-E Wirsall & PG Holmlov
explorativ studie. Foundation for Distribution Research,
R Wahlund & J Gunnarsson. Stockholm, 1994.
Promotional Response, Sales Development

In Search of the Deal Prone Consumer - a and Purchasing Patterns - in the
study of consumers' usage of sales ,confectionery category.
promotions introducing a new type of C Holmberg & U Sanden.
single source data. Foundation for Distribution Research
M Hemant & P-G Persson Working Paper Series, Stockholm, 1994.
Foundation for Distribution Research,
Working Paper Series, Stockholm.
An Empirical Study of Consumers'
Reference Prices in a Promotion Intensive
Medievalsprocessen och dess aktorer - en Product Category.
deskriptiv studie av reklamkampanjer, M Hemant, C Holmberg & P-G Persson.
medieval och samarbete. Foundation for Distribution Research
U Sanden Hakansson. Working Paper Series, Stockholm, 1994.
Working Paper Series, Stockholm, 1994. Prissattning av dagligvaror - en explorativ
intervjuundersokning med beslutsfattare.
Hushallens finansiella strategier - nagra LFossum.
resultat fran en empirisk studie i
Publications from the Foundation for Distribution Research

Working Paper Series, Stockholm, 1994. Toward a Theory of Promotional Effect in
Grocery Retailing.
Upplevd kvalitet, produktivitet och C-R Julander & P-G Persson.
lonsamhet i dagligvaruhandeln. EFI Research Paper, Stockholm, 1993.
A Magi & C-R Julander.
Foundation for Distribution Research A Causal Model of Labor Productivity - an
Working Paper Series, Stockholm, 1994. empirical test using PLS.
C-R Julander & M Eliasson.
Cannibalisation Among Consumer Goods Paper presented at EMAc Conference,
- an empirical study of the introduction of Barcelona, Spain, May 25-28, 1993.
new products in the milk category.
P Hakansson & M Soderlund. Intelligent Food Distribution Networks in a
Foundation for Distribution Research Common Market Europe
Working Paper Series, Stockholm, 1994. o Fredriksson & P G Holmlov
Paper presented at the Executive and
Produktivitet och lonsamhet i Researcher Seminar "The Swedish Food
dagligvaruhandeln Industry and the European Market", Bro, 21-
A Magi, M Eliasson & C-R Julander. 23 October, 1992.
Working Paper Series, Stockholm, 1994. Basket Analysis: A New Way of Analysing
Scanner Data.
Do Strong Brands Make a Difference? C-R Julander.
P Hakansson. International Journal of Retail and
Foundation for Distribution Research Distribution Management, Vol 20, No.7,
Working Paper Series, Stockholm, 1994. 1992, pp 10-18.
Promotional Response Variation Across Basket Analysis: A New Way of

Product Categories Studying Short Term Promotional Effects
P-G Persson. in Grocery Retailing.
EFI Research Report, Stockholm, 1993. P-G Persson.
Marketing Mix and Market Share
- a descriptive study of advertising and Datorkommunikation i
promotion behavior in two consumer distributionssystem.
product categories. o Fredriksson,
U Sanden. EFI Research Report, Stockholm, 1992.
Working Paper Series. Stockholm, 1993.
Decomposing Promotional Elasticity

P-G Persson.
EFI Research Report, Stockholm, 1993. Effects of Feature Advertising and Other
Promotions.
Evolution of Distribution: Prerequisites for C Holmberg,.
and Outcomes of Inter-Organizational EFI Research Report, Stockholm, 1992.
Information Systems
o Fredriksson & M Vilgon Den liittrorliga detaljhandeln.
Paper presented at the European Science N-E Wirsall & C-R Julander.
Foundation Conference, Berlin, Germany, 6-7 EFI, Stockholm, 1991.
September, 1993.
Productivity in Swedish Grocery Retailing. Stockholm, 1986.

M Eliasson & C-R Julander.
EFI, Stockholm, 1991. Kabel-tv i Sverige.
L Hedman & P G Holmlov.
Framgangsrik anvandning av Stockholm, 1986.
ny informationsteknologi i distributionen
av varor and tjanster. Distribution i forandring
o Fredriksson & P G Holmlov (ed.). K-O Hammarkvist.
TELDOK, Stockholm, 1990. Stockholm, 1986.
Telefaxen och anvandarna. Tittande pa satellit-tv i Upplands Vasby.

C Lindberg, P G Holmlov & K-E Warneryd. P G Holmlov, L Hedman & M Thomas.
TELDOK. Stockholm, 1989. Uppsala Universitet, Uppsala, 1985.
What's in the basket? Reklam i satellit-tv alIt narmare.

Receipt analysis for marketing decisions: P G Holmlov & C-R Julander.
An exploratory study. Stockholm. 1984.
C-R Julander, Working Paper, 1989.
Finpappersgrosshandelns utveckling sedan
Debatt, forskning, fornyelse. 1950.
Nodvandiga inslag i handelns utveckling. A-M Lindstedt.
N-E Wirsall. EFI Research Paper 6278.
Torsten and Ragnar Soderbergs Stiftelser, Stockholm, 1984.
Goteborg och Stockholm, 1988.
Svensk farghandel i utveckling.
Marknadsinformation fran datakassor. B Engman.
C-R Julander & C Holmberg. Stockholm, 1984.
EFI Research Report. Stockholm, 1987.
Reklam i satellit-tv.
Distribution av varor och tjanster i Konsekvenser for svensk industri.
informationssamhallet. En forstudie som P G Holmlov & C-R Julander.
utgangspunkt for fortsatt forskning. Stockholm, 1984.
o Fredriksson, P G Holm1ov & C-R Julander.
Stockholm, 1987. Extrapriser och veckoannonser. Effekter
av dagligvaruhandelns SAIVA-aktiviteter.
C-R Julander.
Stockholm, 1984.
Improved In-Store Marketing from
Technology Sourced Data. Kopa via datorer. Handel utan butiker och
C-R Julander. grossister via teledata.
Paper presented at Spectra Usedata -87 T Falk.
Conference. Stockholm, 1983.
Projekt 2000. Current Trends in Distribution Research.

Foundation for Distribution Research, International Joumal of Physical Distribution
Stockholm, 1987. and Materials Management. Vol 13, No 5/6,
1983.
Effektivisering av golvbranschen. En Red av T Falk & C-R Julander.
studie av struktur och lonsamhet samt
nagra ideer till forandringar. Kopcentrum i stadskarnan - inverkan pa
L Albert & 0 Fredriksson. detaljhandel och konsumenter.
T Falk & C-R Julander

Stockholm, 1983.
Handelns fornyelse.
N-E Wirsiill.
Vasteras, 1982, utokad upplaga 1986.
80-talets grosshandel.
Red av K-O Hammarkvist & N-E Wirsall.
Stockholm, 1980.
Attiotalets handel.
Red av A Thimerdal & N-E Wirsall.
Stockholm, 1979.
EFI The Economic Research Institute
Reports since 1991
Published in the language indicated by the title
1991
Bergholm, M., Hemdistribution med hjaIp av Wahlund, R., Skatter och ekonomiska
datorkommunikation. Research Report. beteenden. En studie i ekonomisk psykologi
Bojo, J. , The Economics of Land Degra- om framst skattefusk och sparande utifran
dation. Theory and Application to Lesotho. 1982 ars skatteomlaggning.
Brytting, T., Organizing in the Small Wahlund, R., Studenternas betalningsvilja for
Growing Firm. studier vid Handelshogskolan i Stockholm.
Edlund, P-O., S9gaard, H.T., Business Research Report.
Cycle Forecasting: Tracking Time-Varying Westelius, A., Westelius, A-S.,
Transfer Functions. Research Report. Decentraliserade informationssystem. Tva
Ericson, M., Iggesundsaffaren. Rationaliteter fallstudier inom ekonomistyrning. Research
i en strategisk forvarvsprocess. Report.
Hom af Rantzien, M., Endogenous Fertility Westlund, K., Affarsetik. En studie av etiska
and Old-Age Security. Research Report. bedomningar. Research Report.
Jennergren, L.P., Naslund, B., Options with Wirsan, N-E., Julander, C-R., Den lattrorliga
Stochastic Lives. Research Report. detaljhandeln.
Jonung, L., Gunnarsson, E., Economics the
Swedish Way 1889-1989. Research Report. 1992
Jungenfelt, K., An Analysis ofPay-as-You-Go
Pension Systems as Dynastic Clubs. Research Charpentier, C., Ekonomisk styrning av
Report. statliga affarsverk.
Lundgren, A., Technological Innovation and Edlund, P-O., Karlsson, S., Forecasting the
Industrial Evolution - The Emergence of Swedish Unemployment Rate: VAR vs.
Industrial Networks. Transfer Function Modelling. Research
Nilsson, A. G., Anskaffning av standardsystem Report.
for att utveckla verksamheter. Utveckling och Eldof, J.A., Varying Data Quality and Effects
provning av SIV-metoden. in Economic Analysis and Planning.
Nilsson, J. , Ekonomisk styrning i ett Eliasson, M., Julander, C-R., Productivity in
multinationellt foretag. Research Report. Swedish Grocery Retailing. - changes over
Normark, P., Swartz, E., Bolagisering av time and a causal model.
ekonomiska foreningar. Research Report. Ewing, P., Ekonomisk styrning av enheter
Rutstrom, E., The Political Economy of med inbordes verksamhetssamband.
Protectionisnl in Indonesia. A Computable Fredriksson, 0., Datorkommunikation i
General Equilibrium Analysis. Distributionssystem. Erfarenheter och effekter
Sjostrand, S-E., Institutions as Infrastructures vid inforandet av tva multilaterala
of Human Interaction. Research Report. interorganisatoriska informations-system -
Sooerqvist, T., Measuring the Value of exemplet BASCET Infolink AB. Research
Reduced Health Risks: The Hedonic Price Report.
Technique Applied on the Case of Radon Fredriksson, T., Policies Towards Small and
Radiation. Research Report. Medium Enterprises in Japan and Sweden.
Research Report.
Foretagsledning bortom etablerad teoti, 1993
red. Sjostrand, S-E., Holmberg, I.,
EFI/Studentlitteratur. Ekvall, N., Studies in Complex Financia
Gerdtham., U., Jonsson,B., Sjukvards- Instruments and their Valuation.
kostnader i de nordiska Uinderna. Flink, T., The Developnlent of High-Spee
Research Report. Trains - An Industrial Dynanlics Approach
Hedvall, M., Produktkvalitet som kon- Research Report.
kurrensmedel i producentvaruprodu- Gerdtham., U-G., Hertzman, P., Boman, G.
cerande industri. Research Report. Jonsson, B., Impact of Inhaled Corticosteroid .
Holmberg, C., Effects of Feature Advertising on Asthma Hospitalization in Sweden: ~
J
and Other Promotions Research Report. Pooled Regression Analysis. Research Report.
Jansson, D. , Spelet kring investeringskalkyler. Gun-Ho Cha, Reappraisal of Marke1
Onl den strategiska anvandningen av det for- Efficiency Tests Arising from Nonlineru
givet-tagna. Norstedts Forlag/EFI. Dependence, Fractals, and Dynamical System~
Jennergren, L.P., Naslund, B., Valuation of Theory.
Executive Stock Options. Research Report. Hertz, S., The Internationalization Processe~
Kokko, A., Foreign Direct Investment Host of Freight Transport Companies.
Country Characteristics and Spillovers. Nilsson, K., F6rhaIlanden som paVerkal
Ljung, A., Intressentstrategier. rapporteringstiden - En studie i tre koncerner.
- En longitudinell studie av utvecklingen i tva Research Report.
svenska foretag. Redovisningensroller, red. Samuelson, L.A.:
M3rtensson, P., Mahring, M., Information Ostman, L.
Support From Staff to Executives. - An Ruist, E. , Sambandsanalys med ordinal-
Explorative Study. Research Report. skalevariabler. Vilka statistiska metoder kar
Rabiee, M. , A Review of Demand for anvandas? Research Report.
Telecommunication: An Analytical Approach. Skogsvik, K. , Accounting for PensioI
Research Report. Obligations. Research Report.
Paalzow, A. , Public Debt Management. Sjostrand, S-E., Om foretagsformer.
Research Report. Stein, J., Strategy Formation and Manageria:
Persson, P-G., Basket Analysis. Agency. A Socio-Cognitive Perspective.
A New Way of Studying Short Term Effects SOderlund, M., Omvarldsmodeller ho~
of Promotions in Grocery Retailing. Research beslutsfattare i industriforetag - en studie a,
Report. svenska leverantorer till fordonsindustrin.
Sjostrand, S-E. , On Institutions and Sorensson, T., Swedish Convertible Bond~
Institutional Change. Research Report. and their Valuation.
SOdergren, B., Decentralisering. Forandring Tyrstrup, M., F6retagsledares arbete. Er
i foretag och arbetsliv. logitudinell studie av arbetet i en foretags-
Thodenius, B., Anvandningen av ledning.
ledningsinformationssystem i Sverige: Udden-Jondal, E., Wage Formation in (;
Lagesbild 1991. Research Report. Unionized Economy.
Tollgerdt-Andersson, I., Sjoberg, L., Intresse Whitelegg, J., Hulten, S., Flink, T., Higl
och kreativitet inom tjansteproducerande Speed Trains. Fast Tracks to the Future.
foretag. Research Report.
Wahl, A., Konsstrukturer i organisationer. 1994
Kvinnliga civilekonomers och civilingenjorers
karriarutveckling. Andersson, H., Ett industriforetags om·
vandling. En studie av Hagglunds for
andringsprocess 1922-81 med bas i foretaget:
produkter, relationer och resurser.
Andersson, H., En produkthistoria. (separat 1995
publicerad bilaga till ovanstaende)
Cowan, R., Hullen, S., Escaping Lock-in: Gunnarsson, J., Wahlund, R., Hushallens
The Case of the Electric Vehicle. Research finansiella strategier. En explorativ studie.
Report. Fonden for Handels- och Distributions-
Dahlfors, G., Jansson, P., Essays in forskning, EFI.
Financial Guarantees and Risky Debt. Sandberg, J., Statistisk llletod dess
Foretag och marknader i foriindring - dynamik vetenskapliga hemvist, grundlaggande
i natverk, red. Mattsson, L-G., Rulten, S. principer och mojligheter inom
EFI/Nerenius & Santerus Forlag. samhallsvetenskapen. Research Report.
Helgesson, C-F., Coordination and Change in SOderqvist, T., Benefit Estimation in the Case
Telecommunications. Research Report of Nonmarket Goods. Four Essays on
Lundstrom, T., Wijkstrom, F., The Swedish Reductions of Health Risks Due to Residential
Nonprofit Sector. Research Report. Radon Radiation.
Nessen, M., Essays on Exchange Rates and
International Finance.
Niemeyer, J., Essays on Market
Microstructure -Empirical Evidence from
Sonle Nordic Exchanges.
Nilsson, K., Rapportering i svenska
storforetag 1993. Delrapport 1. Faktiska tider
och paverkande faktorer for manatlig
resultatrapportering. Research Report.
Nilsson, K., Rapportering i svenska
storforetag 1993. Delrapport 2.
Rapporteringstider for kvartals/tertials-
rapporter. Research Report.
Nordenlow, L., Entreprenorskap i industriella
natverk. Research Report.
Noren, N-E., Om Flexibilitetspaverkande
beslut. Research Report.
Normark, P. , Medlemsagda foretag.
Organisering av strategiska forandringar.
Nyman, K., Studier av inkomstfordelningar
med hjalp av mikrosimuleringsnl0deller.
Research Report.
Th.odenius, B., Anvandningen av lednings-
informationssystem i Sverige: Lagesbild 1993.
Research Report.
Sjoholm, G., Redovisningsmatt vid
bolagisering. Utformning och effekter.
WijkstrOID, F., Det organiserade ideella
arbetet i Sverige. Research Report.
Wirs3.ll, N-E., Holm1ov, PG., Den nyckfulle
Svensson. Reflektioner efter ett senlinarium
om strategier for handeln. Fonden for
Handels- och Distributionsforskning vid EFI.

Full Text 01

Uploaded by

Copyright:

Available Formats

Full Text 01

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Full Text 01

Uploaded by

Copyright:

Available Formats

Modeling the Impact of

Sales Promotion on Store Profits

Address: Sveavagen 65, Box 6501, S-113 83 Stockholm. Tel. +468-7369000

A Management and Organization

© EFI and the author 1995

Graphic Systems i Stockholm AB, 1995

This volume is submitted as a doctor's thesis at the Stockholm School of Economics. As

Stockholm, April 1995

Stockholm, April 1995

2. SALES PROMOTION AND THE CONSUMER

3. PROMOTIONAL SALES RESPONSE

6. SIMULATING PROMOTION PROFITS

7. EMPIRICAL DATA AND ESTIMATION METHOD

8. EMPIRICAL ILLUSTRATION 1: COFFEE

9. EMPIRICAL ILLUSTRATION 2: CEREALS

10. EMPIRICAL ILLUSTRATION 3: PASTA

1.1. The Problem

1.2. What is Sales Promotion?

1.2.1. Defining Sales Promotion

Sales promotion is the direct inducement or incentive to the sales force,

Sales promotion represents those marketing efforts that are

1.2.2. Types of Sales Promotion

Figure 1.1. Types ofsales promotion.

Retailer Promotions Trade Promotions Consumer Promotions

1.3. Why is there Sales Promotion?

1.3.1. Sales Promotion Objectives

1.3.2. Economic Rationale for Sales Promotion

1.4. Planning Sales Promotion

The retailer's promotion-planning process should be based on the progression from

1.4.2. Store Level Promotional Strategy

S = total retail sales over the planning period

They then define the average fraction total markup as:

Table 1.2. Pricing and promotion strategies for two supermarkets

1.4.3. Selecting Items to Promote

1.4.4. Deciding the Promotion Frequency

S = sales for a merchandise program during one week

During the promotional weeks, the gross margin is:

G2 (p,! ,/2 ) = (p - c)· S(p,po,/2 ,t)- h· 12 - vo (1.7)

c = retailer's marginal cost per unit

The average gross margin per week can be calculated as:

1.5. Importance and Growth of Sales Promotion

1.6. Complex Planning Problem

1.6.1. Unknown Profit Impact

1.6.2. Easy in Principle

Kotler (1991) describes the theory of effective marketing resource allocation, an

1.6.3. Impossible in Practice

1.7. Purpose and Delimitations

• Define how to maximize promotion profits, identify important determinants of the

• Demonstrate the difference between retailer-oriented promotion profit

1.7.3. Steps to Achieve the Purpose

) 1.8. Organization and Overview of the Report

Figure 1.2. Structure of the report.

2.3. Sales Promotion and the Store Choice

2. which is (are) inferred

3. either from a set of ongoing perceptions and/or memory inputs

4. and which represent(s) what that phenomenon signifies to an

2.4. Sales Promotion and the Category Choice

2.5. Sales Promotion and the Purchase Quantity Choice

2.6. Sales Pronlotion and the Brand Choice

2.7. Combining the DecisiollS