MUTUALLY BENEFICIAL PRICE PROMOTIONS FOR MANUFACTURER AND RETAILER

(1)

1

MUTUALLY BENEFICIAL PRICE PROMOTIONS FOR

MANUFACTURER AND RETAILER

Ruud Willem Buijgers

(2)

2

Master’s Thesis

University

of Groningen

Faculty of Economics and Business

Department of Marketing

July 2013

Author Ruud Buijgers Damsterdiep 86b 9713 EK Groningen +31 (6) 239 015 29 Supervisor

(3)

3

SUMMARY

Short-term sales increases are often generated through price promotional activity within the product category. These price promotions are shaped through use of price discount with or without support of feature and/or display. The effectiveness of the different price promotional setups is assessed through use of regression models for own brand sales and category sales. This results in determining if price promotions are truly beneficial from a managerial perspective. Two parties are involved, the brand manufacturer and retailer, which both have different demands with respect to sales promotions. Where brand manufacturers are mainly interested in brand switching behavior, the retailer aims for category expansion.

The input for the regression models consists of scanner panel data of 28 supermarkets in the United States of America. The investigated product category is canned tuna. Most important finding is that a price discount supported by feature and display shows the strongest positive effect on short-term own brand and category sales. Furthermore, price discounts need to be implemented to create these sales increases, as use of feature and/or display without a price discount does not show substantial effects. Lastly, competitor use of price promotions affects own brand sales, which implies brand-switching effects being present. These effects are again strongest when a price discount is supported by feature and display. In addition, differences between brands are found in susceptibility to competitor price promotional activity.

Competitor susceptibility is also of interest to the retailer. Brands that experience immediate sales losses when competitors are on promotion are non-beneficial to retailers. The consumer switches to the promoted brand, which has lower margins. Retailers should induce category expansion by means of featuring, as there are mild indications for brand switching when display is used. Overall, retailers find significant category sales increases more often when feature advertising is used as opposed to the use of an in-store display.

Key words: Sales promotion; price promotion; effectiveness; regression models;

(6)

6

1. INTRODUCTION

When going grocery shopping nowadays, it is nearly impossible to miss the in-store advertising and price discounts. Brands are fighting for the attention of the customer by engaging in advertising and promotion attacks, with the main goal of increasing their sales. Both brand manufacturer and retailer engage in promotional activities. Brand manufacturers try to increase their demand for a single product through price reductions. These price reductions should in turn lead to additional sales, when consumers evaluate the lowered price worthy of switching products or increasing consumption. The relevant question then becomes whether the added revenues from additional sales are able to compensate for the margin loss instigated by the price reductions (Srinivasan et al. 2004). Brand manufacturers are able to evaluate this necessity aggregated over all retailers were the product is sold. Where brand manufacturers are mainly interested in these overall sales for a product within the market, retailers have additional concerns. Retailers can benefit from promotions through an increase in demand of the focal and complementary categories (Srinivasan et al. 2004). Moreover, low retail prices can be favorable to the retailer, as it might lead to new additional customers that increase sales in unrelated categories (Van Heerde et al. 2004). The interests of manufacturer and retailer are not necessarily aligned. Whereas manufacturers implement price reductions to induce brand switching and increased consumption, retailers reduce prices to create category expansion and store switching behavior. Question that remains is how effective the current price promotional strategies are. Furthermore, how should managers shape promotions to create beneficial outcomes for both parties?

As brand manufacturers and retailers pursue different goals with respect to implementing price reductions, both have different preferences on the accompanied support. In other words, brand manufacturers prefer a promotional setup that induces customers to switch brands, while retailers prefer a promotional setup that induces an increase in category sales or store switching behavior. Furthermore, retail managers have to engage in price promotions as the retail landscape is under high promotional pressure (Dekimpe et al. 2005). As other retailers are conducting price promotions, ceasing to do so could lead to reduced market share.

(7)

7 originated by Gupta (1988). They found that: “price promotions have a predominantly positive impact on manufacturer sales, revenues and category sales”. Furthermore, from a retailer perspective, a 100-unit increase for the promoted brand creates a potentially beneficial 35-unit increase for the retailer due to category expansion effects (Van Heerde et al. 2004). To summarize, although long-term effects are absent, in the short-term, both parties are able to benefit from price promotions from distinct sources of the sales promotion bump.

The contribution of this research is the determination of the effectiveness of price discounts, supported by feature and/or display advertising, and its effect on sales for manufacturer and retailer. Van Heerde et al. (2004) show that different promotional setups instigate different outcomes with respect to the sales promotion bump. This finding relates to an important additional aspect of this research. As retailers and manufacturers both strive for different forms of sales increase, price promotions should be shaped to create mutually beneficial outcomes. This implies the importance of brand switching and category expansion and the necessity to induce these demand effects. This leads to the following research question;

Research Question - What combination of marketing mix elements is most effective to brand manufacturers and retailers to induce brand switching and category expansion?

The main research question is answered by investigating multiple closely related problems. The effectiveness of price discounts with and without support is assessed, as well as the influence of competition on own brand sales. Furthermore, the effects of advertising and display are examined to estimate their influence on different parts of the sales promotion bump. This leads to the following sub-questions for answering the research question;

Sub-questions

How effective are price discounts in terms of short-term own brand volume sales increase? How is the short-term volume sales effect moderated by use of feature and/or display? What is the influence of the competition on own short-term brand volume sales?

(8)

8 Two econometric sales models are specified for investigating the research question. Both models are linear additive ordinary least squares (OLS) models, as this enables measurement of interaction effects of the marketing mix variables. Interaction effects can exist between a price reduction and the accompanied support by feature and/or display. This implies that sales can become increasingly higher when discounts and support increase. In other words, the occurrence of synergy between the mentioned constructs. Two models are built to measure two different outcomes of price promotions. The first model measures the effects of price promotions on brand switching, whereas the second model depicts category expansion effects. The second model is aggregated across all brands within the category.

The input for testing the hypotheses consists of scanner panel data from the canned tuna product category. Through regression, the effectiveness and impact of the available variables on sales is assessed. The results are elaborated upon in chapter four, while the most important findings are described here.

From the results it becomes apparent that price discounts become more effective when supported by feature, display or feature and display. Furthermore, competitors influence own brand sales within the category. The largest effect of competitor activity is found when feature and display is implemented along a price discount. In addition, when feature or display are used separately, display has stronger effects on own brand sales, as well as on brand switching and category expansion. In contrast, no substantial effects are found for use of feature and/or display without use of price discounts. Lastly, brand and store specifics codetermine price promotional effectiveness as the effects differ greatly between brands and stores.

(9)

9

2. LITERATURE REVIEW

As price is the most central marketing mix element and it is directly related to demand, many retailers are engaged in price promotions to boost sales. In the Netherlands, about 24 percent of food retail sales are sold under some form of promotional support. These figures are comparable to the United Kingdom and Spain, while in the United States the promotional pressure even approaches 40 percent (Dekimpe et al. 2005). These figures imply that a large part of the products is sold on promotion, as retailers found that promotional discounts could drastically increase sales in the short-term. Although short-term sales increase, every single product on promotion is sold below its average price. Retailers and brand manufacturers thus indirectly invest in the additional short-term sales that are induced by price promotions. This investment needs to be beneficial to both parties.

Competitive price promotional strategies have evolved over time. Most important was the shift in marketing budgets from advertising expenditures to price promotions, as the short-term effectiveness is generally strong for the category demand (Tellis 1998; Nijs et al. 2001). However, the long-term effects of price promotions tend to be much weaker. Price promotional effects die out within approximately 10 weeks a period referred to as ‘dust settling’ and their long-term impact converges to zero in 98 percent of all cases (Nijs et al. 2001). These weak long-term impacts were also found for market shares (Srinivasan et al. 2000), as well as for purchase incidence, brand choice, and purchase quantity (Pauwels et al. 2002). Price promotions are thus mainly a tactical short-term tool to generate an immediate market response. Price reductions can be supported by use of feature and/or display to positively influence this market response. These marketing tools interact and create sales promotion bumps that consist of primary and secondary demand effects. The primary demand effects are accounted for by purchase and timing acceleration, while the secondary demand effects are defined as brand switching (van Heerde et al. 2003).

(10)

category-10 expansion effect, while the manufacturer also benefits from brand switching (van Heerde et al. 2004). Sun et al. (2003) underline this statement by indicating that promotions cause brand switching which contributes to profit.

The different instruments that shape price promotions, the short- and long-term effects, the different effects on the parts of the sales bump, and competitor effects, are further explained in the upcoming paragraphs.

2.1 Price Promotional Setup

(11)

11

2.2 Short and Long-term Effects

The short-term effects of price promotions are generally strong, with an average elasticity of 2.21, although they rarely exhibit persistent effects (Nijs et al. 2001). In their study they investigated 560 product categories over 4 years of market-level data. The large data set allowed for empirical generalizations on the promotional effectiveness in the short- and long-term. As stated, while the short-term effects are strong, the effect converges to zero in the long-term for 98 percent of the investigated product categories (see table 1). The category demand stabilizes after a so-called ‘dust-settling’ period, which arises after the immediate short-term sales increase and lasts for around 10 weeks. The absence of a long-short-term increase in category demand according to Nijs et al. (2001) is due to consumers not increasing their consumption or buying behavior with respect to promotional activity.

Table 1 – Price Promotion Effects on Category Sales

Effect Short-term Long-term

Positive 58% 2%

Negative 5% 0%

Zero 37% 98%

Pauwels, Hanssens and Siddarth (2002) acknowledge this finding. Their research investigated the long-term effects on a brand level and found that the effects were zero for the investigated products. The research depicted the investigation of a perishable and a non-perishable brand of respectively canned soup and yoghurt. The largest price promotional effect was found in the short-term and was due to brand switching within the category.

(12)

12 that price promotions actually generate store traffic, this was only true in 15 percent of the cases. In 85 percent of the cases, no significant effect was found of promotional impact on store traffic. Overall, conducting price promotions is more financially viable to the manufacturer than to the retailer. Srinivasan et al. (2004) propose that “manufacturer side payments are needed to offset retailer losses. However, only in a small fraction of the cases is there sufficient manufacturer surplus to allow for such side payments without making the combined channel impact negative”. The setup of price promotions should thus be shaped that these side payments are not a necessity, instead both parties must directly benefit from price promotion.

As all research points in the same direction where short-term effects are mainly positive, while long-term volume sales increases are absent on both category and brand level. These effects are not permanent and dye out over time. Price promotions should thus be accounted for on the short-term, implying their immediate effects added by the effects measured during the dust-settling period. These short-term effects can be split into different effects. According to Van Heerde, Leeflang and Wittink (2004), the sales promotion bump can be decomposed into three parts: (1) cross-brand effects (secondary demand), (2) cross-period effects (primary demand borrowed from other time periods), and (3) category-expansion effects (remaining primary demand). These three parts each roughly account for one third of the sales promotion bump. Though, the parts differ in attractiveness to the manufacturer and retailer.

2.3 Manufacturer and Retailer

(13)

13 category expansion effect” (Mela, Jedidi and Bowman, 1998). An overview of the different demand effects according to Van Heerde et al. (2003) is displayed in figure 1.

Figure 1 – Overview of Average Effects when Decomposing the Sales Bump

(14)

14 retailers can lower promotional costs without consequences for their market share. Jedidi et al. (1999) underlines this prisoners’ dilemma combined with the promotion sensitive consumer. Their research argues that in the long-term, price promotions make it more difficult to increase regular prices and increasingly greater discounts need to be offered to have the same effect on consumers’ choice. Furthermore, Nijs et al. (2001) argue that manufacturers would like to reduce their reliance on price promotions but are reluctant to do so, being afraid they lose the support of retailers who still appreciate the market expansive power of price promotions. Altogether, these insights are the drivers of high promotional pressure and add to the fact that price promotions seem to have become unavoidable within the retailing market.

Table 2 – The Prisoners’ Dilemma of Promotions Retailer B Lowers Promotional

Pressure

Retailer B Does Not Lower Promotional Pressure Retailer A Lowers Promotional

Pressure

New Status Quo, Lower

Promotional Costs (+) Retailer B Wins Market Share

Retailer A Does Not Lower

Promotional Pressure Retailer A Wins Market Share Status Quo

2.4 Shaping Price Promotions

Most desirable effects of price promotions are cross-brand and category-expansion effects, where the latter is simultaneously beneficial for both retailer and manufacturer. But can the unavoidable price promotions be shaped in such a way that both retailer and manufacturer experience these desired effects?

2.4.1 Main Effect Price Promotions

(15)

15 elasticity is the lowest for unsupported price cuts and the highest for price cuts with feature and display”. In addition, Wittink et al. (1988) found a 25 percent increase in sales when a retailer jointly used feature and display (as opposed to lower results for separate support). These findings imply that the effects of price discounts increase when feature and display are put into place to support the price discount. Neslin (2002, p. 324) depicted an overview of average sales increases when a price reduction of 15 percent was implemented (see table 3). From the results it can be concluded that a price discount supported by feature and display yields the highest sales increase.

Table 3 – Sales Increase due to Price Reductions

Promotion Type Mean Percent Increase in Sales

15% TPRa _35%

Feature + 15% TPR 173%

Display + 15% TPR 279%

Feature + Display + 15% TPR 545%

a_{Temporary Price Reduction}

Lastly, “low demand is often a motivation for remedial action, and (temporary) price reductions offer a quick fix to boost sales and meet performance quotas” (Neslin 2002). Earlier research underlines the connection between price, feature and display, and its positive effect on short-term sales, therefore;

Hypothesis 1 - Price discounts (a) have a positive effect on short-term own brand sales. The effect increases when (b) feature, (c) display, or (d) feature and display support the price discount.

2.4.2 Effects of Feature and Display

(16)

16 storage costs. Neslin (2002) even states that stockpiling is a fundamental consequence of sales promotions. To avert this negative side effect of promotions, Sun et al. (2003) propose a maximum number of units purchased on promotion. This can increase the promotion effect of attracting new purchases switched from other brands or other stores, but can limit stockpiling behavior. Another variable that influences cross-period sales is promotional frequency. For instance, when the promotional frequency within a category is high, the necessity to stockpile is low, as consumers know the price discount will be available in future periods (Bijmolt et al. 2005). In contrast, a side effect of high promotional activity is consumers becoming more price conscious. Therefore, they tend to purchase products on deals and may develop a habit of stockpiling (Fok et al. 2005). Altogether, Van Heerde et al. (2003) state that stockpiling and/or category-expansion together constitute the dominant sources of sales effects due to temporary price cuts. Van Heerde et al. (2004) state that a cross-period effect is predominantly induced by feature-only supported price discounts, as consumers then apparently plan an inventory increase. As completely averting cross-period sales seems unrealistic due to the various causes mentioned, the marketing-mix variables should be implemented to at least minimize the effect of cross-period sales, while maximizing category-expansion and cross-item sales.

Category-expansion and Cross-item. Supporting price discounts by means of

advertising and display inevitably creates cross-period sales (Van Heerde et al. 2004). These cross-period sales can be countered by use of display, opposed to feature advertising, as displays attract consumers considering a product category purchase; resulting in brand switching, while feature advertisements tend to be seen by brand-loyal consumers (Wittink et al. 1988). Use of display without advertising thus minimizes the cross-period effect and results mainly in category-expansion and cross-item sales. Apparently, the use of display causes impulse buying and in-store decision-making, which does not induce stockpiling. Fok et al. (2005) underline this finding by stating that a display attracts the attention of the buyer when making the actual brand choice. However, supporting price cuts with display-only support is not a desirable outcome, as only ‘no support’ yields a lower price promotional sensitivity (Van Heerde et al. 2004).

(17)

17 course, is not a viable outcome as the use of feature and display has proven to impact short-term sales positively (own-brand promotional elasticity of 1.82, compared to 0.44 for ‘no support’). Furthermore, in relation to cross-store effects, the brand must be communicated to the consumers. Without use of feature and/or display, market expansion will be the only effect, as consumers have no reason to switch stores.

Store Switching. Van Heerde et al. (2004) also investigated the cross-store effects

within the canned tuna category. The results implied that using display and feature jointly had the largest effect on cross-store sales. On average, 81 percent of the additional sales due to the category-expansion where accounted for by these cross-store sales. However, these results are “quite unreliable”, as only four out of 16 cross-store effects are significant. Nijs, Srinivasan, and Pauwels (2007) state that “empirical evidence on the link between retail prices an store traffic/store switching is mixed.”

(18)

18 Overall, the link between retail prices and store switching is stated to be weak at best. This implies that pursuing the generation of store traffic through price promotions should not be a retail manager’s main goal. Category expansion is a plausible outcome, where generating store traffic yields insignificant results. “Perhaps managers should think of promotion as a tool for growing the category rather than only as a market share weapon” (Ailawadi and Neslin, 1998). The focus for retailers should thus be on category expansion through promotional activity.

Altogether, these findings imply different relations of feature and display towards the desired effects of brand switching and market expansion. It can be concluded that use of feature and display to support price discounts yields the largest response in short-term sales increase, while the effect of feature or display has different implications for the separate parts of the short-term sales bump. Therefore, the following is hypothesized;

Hypothesis 2 – (a) Category feature advertising has a positive effect on market expansion, where (b) use of competitor display has a negative effect on own brand sales due to brand switching.

2.5 Competition within the Category

More and more firms currently are engaged in determining and quantifying competitive effects (Kopalle et al. 2009). According to Nijs et al. (2001) the competitive structure is an important factor affecting the price-promotion effectiveness among competing firms. They state that the more brands are present within a category, the more competitive the category is. Subsequently, it can be concluded that price-promotion effectiveness will be higher (lower) in less (more) competitive environments. As stated earlier, price promotions only produce temporal benefits for the brands in the category. These temporal effects are induced because brand choices are in equilibrium in mature markets according to Pauwels et al. (2002). In addition; this equilibrium implies that price promotions are easily offset by competitive activity.

(19)

19 information spreading, which in turn increases consumer price sensitivity. Information theory underlines this argument by stating that providing information to consumers about the available alternatives will make price elasticities more negative (Ataman et al. 2010).

As most brands within the category often use price reductions and advertising to increase own market share, these brands also influence market shares of competitors and vice versa (Naik et al. 2005). Moreover, Naik et al. (2005) found that advertising and promotion not only affect own and competitor brand shares, but also exerts interaction effects. In summary, on a category level price promotions have their impact on competing brands, which implies that competing brands have their effect on own brand sales. In addition, this effect is moderated by use of advertising and/or display. Earlier studies also show that promotions have a large short-term effect on consumer brand choice (Mela et al. 1997). Consumers thus respond to competitive activity, which can be translated to a short-term sales increase (or decrease). Therefore;

Hypothesis 3 - Price discounts of competitors (a) have a negative effect on short-term own

brand sales. The effect increases when (b) feature, (c) display, or (d) feature and display support the competitor price discount.

(20)

20

Table 4 – Overview Hypotheses

Hypothesis Relationship Description

Own Brand Effects

H1a - Price discounts of the focal brand lead to higher own brand sales H1b + Feature increases the effect of price discounts

H1c + Display increases the effect of price discounts

H1d + Feature and display increase the effect of price discounts Marketing Effects

H2a + Category feature advertising has a positive effect on market expansion

H2b - Competitor display has a negative effect on own brand sales due to brand switching Competitor Effects

H3a + Price discounts of the competition lead to lower own brand sales H3b + Feature increases the effect of competitor price discounts H3c + Display increases the effect of competitor price discounts

H3d + Feature and display increase the effect of competitor price discounts

2.6 Conceptual Framework

Figure 2 depicts the conceptual framework that guides this research. The price index of the focal brand has a direct effect on sales. This effect is moderated by the use of feature and/or display when a price discount is supported. The effectiveness of the price promotion is also dependent on competitor presence within the category. The strength of the competitor effect is co-determined by the use of feature and display. Furthermore, it is hypothesized that the advertising and display setup induce different sources of sales increase; brand switching or market expansion. These effects are independent on the initiator when it comes to use of feature. Featuring is hypothesized having a positive effect on category expansion, regardless of the initiator brand. Lastly, if a competitor brand implements display usage, brand switching is hypothesized.

(21)

21

(22)

22

3. METHODOLOGY

First the available data is described, which leads to the specification of the models for testing main and interaction effects. The methodology consists of two econometric sales models to investigate the effects on brand switching behavior and market expansion. To estimate the parameters of the variables, linear additive ordinary least squares (OLS) regression is implemented (Leeflang et al. 2000). Use of a linear additive model allows for testing the interaction effects of the different types of price promotional support. Furthermore, included and excluded variables are mentioned and defined.

3.1 Data Description

(23)

23 mix variables are aggregated across brands. Averages are constructed for the independent variables to capture the price promotional activity of all brands and the effect on category sales.

Table 5 – Price Discount Percentage per Brand per Support Type

Price promotions without support

Price promotions with feature-only support

Price promotions with display-only support

Price promotions with feature-and-display support Brand # Avg % Min %a Max % # Avg % Min % Max % # Avg % Min % Max % # Avg % Min % Max % 1 592 24 5 51 126 29 5 47 98 25 6 44 393 30 5 63 2 419 22 5 52 172 26 6 60 53 23 7 37 254 30 7 60 3 455 22 5 41 103 26 7 45 112 28 7 45 219 32 7 42 4 627 18 5 34 77 17 7 34 76 17 5 34 113 21 5 38

a_{In this table a price promotion is defined as a price 5% or more below the regular price (Van Heerde et al. 2004)}

Table 5 shows that there is a lot of variation in the price promotional setup. As can be seen, all brands show great differences in the minimum and maximum amount of price reduction for all types of support. For example, tuna item 1 was price promoted in 592 store-weeks without support. The average discount was 24%, with a minimum discount of 5% and a maximum discount of 51%. These ranges fluctuate when the type of support changes. For feature-only support the discount range is 5% to 47%, with display-only 6% to 44%, and with feature-and-display support 5% to 63%. Thus, the amount of variation that is present within the price discounts for each type of support is substantial, which can help originate the sales peaks in the dataset. An example of the sales development of brand 1 is given in figure 3, where sales peaks are clearly visible over time (week 1 through 52).

Figure 3 – Sales Development Brand 1 in Store 1

By using OLS regression, these peaks in sales can be explained when making use of the available explanatory variables price index, feature and/or display.

(24)

24

3.2 Model Specification

As the aim of this research is the investigation of price promotional setups that are beneficial to brand manufacturer and retailer, two econometric models are specified. First, a model that accounts for competitor effects on own brand sales is constructed (1). Second, a model that depicts the effect of the price promotional setup on category level sales is constructed (2). Linear additive regression models are used to test the hypotheses due to the possibility of adding interaction effects and the simplicity of the method. A description of the variables in both models is shown in table 6.

(1)

j = 1,…,4, i = 1,…,28, t = 2,…,103

(2)

i = 1,…,28, t = 2,…,103

Model 1 and 2 differ in the aggregation level. For model 2 all variables are aggregated across brands to examine the effect of combined brand activity on category expansion. Furthermore, variables for the competitor effects in model 1 are aggregated as well. Estimating the effect of each competitor separately would drastically increase the number of variables that need to be estimated per store. Through aggregation, the number of observations per parameter estimate increases, which increases the reliability of those estimates.

Own brand effects

Competitor effects

Lag, lead, seasonality

Marketing effects

(25)

25

Table 6 – Description of Variables

Model 1

Variable Description

SIjit The sales index for brand j in store i in week t

β0ji Intercept of the regression model for brand j at store i PIjit Price index of brand j in store i in week t

PIFjit Price index supported by feature of brand j in store i in week t PIDjit Price index supported by display of brand j in store i in week t

PIFDjit Price index supported by feature & display of brand j in store i in week t FOnlyjit Dummy variable that indicates if brand j is featured in store i in week t DOnlyjit Dummy variable that indicates if brand j is displayed in store i in week t

FDOnlyjit Dummy variable that indicates if brand j is featured & displayed in store i in week t PICjit Price index of the competitors of brand j in store i in week t

PIFCjit Price index supported by feature of the competitors of brand j in store i in week t PIDCjit Price index supported by display of the competitors of brand j in store i in week t

PIFDCjit Price index supported by feature & display of the competitors of brand j in store i in week t FOnlyCjit Dummy variable that indicates if brand j is featured in store i in week t

DOnlyCjit Dummy variable that indicates if brand j is displayed in store i in week t

FDOnlyCjit Dummy variable that indicates if brand j is featured & displayed in store i in week t Sit-1jit Lagged sales for brand j in store i in week t

Sit+1jit Lead sales for focal brand j in store i in week t Qt Dummy variable that indicates the quarters Yt Dummy variable that indicates the years Ɛjit Disturbance term for brand j in store i in week t

Model 2

Variable Description

SIit The sales index of all brands in store i in week t

β0i Intercept of the regression model for all brands at store i PIit Price index of all brands in store i in week t

PIFit Price index supported by feature of all brands in store i in week t PIDit Price index supported by display of all brands in store i in week t

PIFDit Price index supported by feature & display of all brands in store i in week t FOnlyit Dummy variable that indicates if brands are featured in store i in week t DOnlyit Dummy variable that indicates if brands are displayed in store i in week t

FDOnlyit Dummy variable that indicates if brands are featured & displayed in store i in week t Sit-1jit Lagged sales for brand j in store i in week t

(26)

26 Model 1 accounts for the influence of pricing, advertising and display of both own brand and competitor brands on sales. These influences result in the effectiveness of the marketing mix variables as well as the influence of competitors within the category. The outcomes show under what circumstances consumers are apparently willing to switch brands, which is viable information for brand manufacturers. Model 2 is of greater interest to the retailers, whereas this descriptive model depicts the effectiveness of the marketing mix variables on category expansion, which is the criterion variable of interest to the retailers.

The criterion variable is defined as the sales index for store i in week t. This is done to obtain information on when category sales are over or under performing for a particular retailer. When the sales index is below 1, category sales are below average, where a sales index above 1 indicates category sales which are above average. Furthermore, the influence of price index, feature, display and feature & display is assessed, together with the interaction effects. These effects are brand and store specific. Lastly, lag, lead and seasonality effects are implemented to further increase model fit.

3.3 Omitted and Included Variables

As is elaborated upon in the literature review, promotional frequency is a variable that co-determines promotion sensitivity of the consumer (Bijmolt et al. 2005, Fok et al. 2005). Within this research the promotional frequency variable is omitted. This implies that heterogeneity between stores will exist, which is captured by the error term in the equation.

(27)

27 the time window can give rise to multicollinearity issues, which will bias the parameter estimates (Leeflang et al. 2000). Altogether, this narrow time window seems sufficient for the purpose of this research.

Lastly, a seasonality variable is included in the model. This variable is a dummy variable that indicates the quarter in which the product is sold. This variable captures (possible) presence of seasonality effects in sales. Ice cream sales for instance increase when temperature rises. These seasonality effects are accounted for by the quarterly dummy variable Qt. A year dummy is also added to account for yearly differences in sales. Together these dummies make up for seasonality effects.

(28)

28

4. RESULTS

In this section the results of the research are described. After estimation of the parameters for the model, several validation issues are assessed. First, the pooling decision is tested by means of the Chow-test. Second, existing multicollinearity issues are resolved. Finally, the modeling assumptions with respect to the disturbance term are solved, making sure the estimated model is not biased. The error term of the model is tested for the nonzero expectation, the normality assumption, autocorrelation and heteroscedasticity. Solving these modeling issues leads to the presentation of the results of the estimated parameters, which show the effects of the marketing efforts of each brand. Lastly, the results are summarized and an overview of accepted hypotheses is given.

4.1 Pooling Decision

In pooling, the disaggregate nature of the data is maintained while all data is used to estimate a common set of parameters (Leeflang et al. 2000, p. 281). An important advantage of pooling the data across stores is the higher number of observations for each parameter, which increases the reliability of the estimates. In this case, the difference between pooling en not pooling the data across stores is the difference between 2856 or 102 observations per parameter. The pooling decision is dependent on the heterogeneity between stores, which is tested by means of the Chow-test. If store differences are large, pooling the data is not allowed as the effects of the explanatory variables differ substantially. The Chow-test (3) is used to test the pooling the decision of which the input is displayed in table 7.

(3)

Table 7 – Input Chow-test Brand 1

SSR df

Unpooled 828.23 2325

Pooled 1183.96 2834

(29)

29 thus not equal and the null-hypothesis of allowance of pooling is rejected. To give a more graphical rendering of the difference between stores, the effects of a 10% price reduction with and without feature support of brand 1 are given per store in figure 4 and 5.

Figure 4 – Sales Increase per Store for a 10% Price Reduction

Figure 5 – Sales Increase per Store for a Featured 10% Price Reduction

Both figures show substantial differences between stores, which underlines the pooling decision. The stores need to be treated as different entities as the effects of marketing efforts differ over stores. Thus, each store requires a different regression model, which implies estimating (5*28) 140 models. The blanks in the graphs are due to non-significant parameter estimates for the promotional setup in that particular store.

(30)

30

4.2 Multicollinearity

Multicollinearity gives rise to unreliable parameter estimates as the variance and covariance are inflated. Multicollinearity is found when independent variables are highly correlated (Leeflang et al. 2000, p. 347). Both models within this research inhibit interaction effects for testing the increase in sales when price discounts are supported. As the original variables and interaction effects of those variables are both incorporated in the model, the effects correlate. Extremely high Variance Inflation Factors (VIFs) are the consequence of simultaneously modeling these effects. Interpreting the parameter estimates hereby becomes unreliable while the interaction effects are a necessity for testing the hypotheses.

To overcome the problem of multicollinearity, the marketing variables are recoded. Other solutions are combining or deletion of variables. As the interaction effects make up for an important part of this study, recoding of the variables is the best option. By recoding, the separate interaction effects are deleted from the model and price becomes the only changing variable for the different support types. The recoding of variables is displayed in table 8.

Table 8 – Variables Recoded Brand 1 Store 1

Standard Coding Variables Recoded

Week PI F D FD PI PIF PID PIFD Fonly Donly FDonly

34 0.69 1 0 0 1 0.69 1 1 0 0 0 35 0.71 0 0 0 0.71 1 1 1 0 0 0 36 1 0 1 0 1 1 1 1 0 1 0 37 0.78 0 0 1 1 1 1 0.78 0 0 0 38 0.85 0 1 0 1 1 0.85 1 0 0 0 39 1 1 0 0 1 1 1 1 1 0 0 40 1 0 0 1 1 1 1 1 0 0 1

(31)

31 Due to recoding of the variables because of multicollinearity issues, both models had to be re-specified. The re-specified models (1 and 2) are displayed in the model specification paragraph (3.2).

4.3 Assumptions of the Disturbance Term

To examine that predictor variables’ effects are linear, that predictor variables do not interact, that predictor variables are truly exogenous and that all relevant variables are included in the specification and measured without error, validity of assumptions of the disturbance term is assessed (Leeflang et al. 2000, p. 330). The following disturbance term assumptions are tested and described; nonzero expectation, normality assumption, first-error autocorrelation and heteroscedasticity.

4.3.1 Nonzero Expectation

Violation of the assumption E(Ɛjt) = 0 is the most serious one (Leeflang et al. 2000, p. 331). If the nonzero expectation holds, the consequence would be a biased parameter estimate. For both models 1 and 2 the mean of the unstandardized residuals for each store is 0, which implies the nonzero expectation does not hold and remedy is not necessary.

4.3.2 Normality Assumption

The normality assumption assumes that the disturbance term is normally distributed. This assumption needs to hold for the standard test statistics for hypothesis testing and confidence intervals to be applicable (Leeflang et al. 2000, p. 343). For testing this assumption the Kolmogorov-Smirnov test shows if the unstandardized residuals deviate significantly from being normally distributed. Furthermore, visual inspection can give a first insight into this matter. Figure 6 shows the distribution of the unstandardized residuals. Visual inspection already shows the residuals may not be normally distributed. When inspecting the output of the Kolmogorov-Smirnov test, it shows highly significant signs of p<.01 over all stores. This implies the residuals are not normally distributed and parameter estimates may be biased.

(32)

32 samples per regression model, over which the parameter estimates are aggregated. These estimates are then compared to the original model. When there are no changes in signs nor changes from significant to non-significance outcomes with respect to the parameters, the original model is assumed to give reliable estimates.

Figure 6 – Normality Plot of Unstandardized Residuals Brand 1 Store 1

4.3.3 First-error Autocorrelation

(33)

33

(4)

The equation (4) shows the GLS-estimator . The β-values for the criterion variable and all the explanatory variables are corrected by means of GLS, which solves the autocorrelation problem for all stores that show violation of the assumption Cov(Ɛjt , Ɛjt’) ≠ 0, t ≠ t’. An example of store 14, which experiences first-error autocorrelation, can be found in table 9.

Table 9 – First-error Autocorrelation Store 14 Brand 1

Standard OLS GLS-Estimator

Ɛt 1 -.199* 1 -.145

Ɛt-1 -.199* 1 -.145 1

* p<.05

As can be seen from table 8, standard OLS estimation leads to significant correlation of Ɛt and Ɛt-1. After correction of variables by means of GLS, the first-error autocorrelation problem is solved. Figure 7 shows the autocorrelation plot for store 14. As can be seen from the graph, only lag number 4 and 9 show significant results as these correlation figures exceed the upper and lower confidence interval respectively. Therefore, after GLS correction, it can be assumed that first-error autocorrelation no longer forms biased parameter estimates.

(34)

34

4.3.4 Heteroscedasticity

The last modeling assumption that is examined with respect to the error term is homoscedasticity, where violation of this assumption implies heteroscedasticity. The assumption is violated when Var(Ɛjt) ≠ σ2. (Leeflang et al. 2000, p. 338) The variance found within the residuals of the model should thus be equal across periods. Within this research, heteroscedasticity is assessed by comparing the variance of year 1 with the variance of year 2 for each brand over all 28 stores. Heteroscedasticity is tested by means of the Levene statistic for homogeneity of variance. The results were non-significant for most stores. In some stores however, heteroscedasticity is an issue.

Brand 1 in store 4 is an example were heteroscedasticity is an issue. To solve the issue, all variables are corrected by means of GLS. The correction is done by dividing the variables by the accompanied standard deviation of the residuals of year 1 and 2 (.411597 and .179362 respectively). As is displayed in table 9, the Levene statistic is no longer significant and equal variances are assumed. Unfortunately, correcting the variables has a side effect in this particular case due to the nature of the recoded variables. Also displayed in table 10 are the VIFs per variable, which show immense increase for the continuous variables after applying GLS. The continuous variables namely display a ‘1’ when there is no price discount. This figure changes when divided by the standard deviation of the residuals, implying there is an effect that moves equal over all variables originally displaying a ‘1’. This allows multicollinearity to arise.

Table 10 – Heteroscedasticity Brand 1 Store 4

(35)

35 According to Leeflang et al. (2000), the error term not being homoscedastic is not nearly as critical as the nonzero expectation. They add; “If only the homoscedasticity assumption is violated, the least-squares estimator is (usually) unbiased but does not have minimum variance.” As the multicollinearity issue directly implies biased and unreliable parameter estimates, heteroscedasticity that is present for some stores is ignored due to the tradeoff decision of violation of one of both error-term assumptions.

4.4 Overview Modeling Assumptions

An overview of all modeling assumptions is displayed in table 11. Due to the differences in stores, none of the Chow-test results indicated the possibility of pooling the data. The recoding of the variables solves the multicollinearity issue. None of the variables show VIFs of above four. With respect to the error term assumptions, the non-normal distributed error terms are not posing an estimation problem as bootstrapping gives similar estimates. Signs and significance levels do not change. Furthermore, only a few stores experienced autocorrelation within the residuals for the model. First-error autocorrelation is corrected for all stores where the issue is present. Lastly, heteroscedasticity formed a problem especially for brand 2 and 4 where 9 and 14 stores respectively show differences in variance between year 1 and 2. As homoscedasticity is the only assumption being violated, the estimated models are assumed to be reliable.

Table 11 – Summary Modeling Assumptions

Model 1 Model 2

Brand 1 Brand 2 Brand 3 Brand 4 All Brands Pooling Decision

Chow-test p-value 0.000 0.000 0.000 0.000 0.000

Multicollinearity

# Variables with VIF>4 0 0 0 0 0

Error Term Assumptions

# Stores now normally distributed 0 0 0 0 0

After bootstrapping; difference in sign and significance no no no no no # Stores with first-error autocorrelation (corrected with GLS) 2 3 1 3 2

# Stores with heteroscedasticity (ignored) 2 9 2 14 6

(36)

36

4.5 Model Fit

After assessing the modeling assumptions, the fit for the separate models is inspected. Table 12 gives a summary of the model fit, where all models were significant at p<0.01. Furthermore, the R-square and adjusted R-square measures (averaged over all stores) are displayed, which are all above 0.73. This figure implies that on average the model explains more than 73% of the variance in the dependent variable (Leeflang et al. 2000, p. 348). Although the averages are satisfactory, the standard error of the R-square measures show there are some store models that have less explanatory power. Moreover, the adjusted R-square measure penalizes for adding variables that do not explain much of the variance within the dependent variable. Although these figures are by definition lower than the standard R-square figure, the deviation is not overly large. This implies that all added variables play their part in explaining the modeled phenomenon of sales increase. An overview of all R-square measures can be found in appendix 1.

Table 12 – Summary Model Fit

Model 1 Model 2

Brand 1 Brand 2 Brand 3 Brand 4 All Brands Model Significance # Models Significant* 28 28 28 28 28 R-square Measurers Average R2 .839 .762 .850 .783 .737 Standard Error R2 (.052) (.075) (.051) (.065) (.062) Average Adjusted R2 .796 .715 .820 .740 .700

Standard Error Adjusted R2 (.063) (.091) (.061) (.078) (.070) * p<.01

4.6 Parameter Estimates

(37)

37

Table 13 – Overview Parameter Estimates per Brand

a

None significant b

Counterintuitive results are shown in bold

Own Brand Effects Competitor Effects

Brand 1 PI PIF PID PIFD FOnly DOnly FDOnly PI PIF PID PIFD FOnly DOnly FDOnly

Significant # Stores + 0 0 0 0 4 3 8 2 3 2 15 0 1 0

Significant # Stores - 24 20 22 27 0 1b 0 0 3 1 0 1 1 0

Average β -2.984 -4.973 -5.408 -12.271 1.437 0.323 1.857 3.563 -1.588 -0.374 3.398 -2.438 -0.093 NSa Standard Error (.807) (1.470) (1.157) (9.024) (.695) (1.013) (.580) (.637) (5.407) (8.487) (.769) (.000) (1.515) NS

Significant # Stores - 17 21 14 27 0 0 0 0 1 3 0 1 2 2

Average β -2.441 -3.538 -6.328 -8.766 0.397 0.784 1.241 NS 1.520 -0.718 4.962 -7.207 0.336 0.821 Standard Error (.736) (.956) (1.836) (4.084) (.146) (.004) (.291) NS (2.862) (5.825) (6.962) (.000) (1.581) (2.034)

Average β -4.401 -9.123 -8.345 -16.183 1.358 1.397 3.105 4.475 5.556 11.021 4.983 -5.419 -1.181 -1.677 Standard Error (3.752) (4.791) (3.120) (6.102) (.000) (.542) (.713) (1.025) (1.132) (2.707) (1.470) (1.598) (.000) (.289)

Average β -3.854 -6.652 -9.474 -10.515 1.064 0.823 1.685 2.140 2.991 4.682 3.096 -0.486 -1.025 0.174 Standard Error (1.076) (1.901) (3.129) (5.783) (.064) (.175) (.804) (.743) (.760) (2.658) (.763) (.000) (.500) (1.193)

All Brands PI PIF PID PIFD FOnly DOnly FDOnly

Significant # Stores + 0 0 0 0 2 2 9

Significant # Stores - 10 21 18 28 1 2 0

(38)

38

4.6.1 Own Brand Effects

All parameter signs for the own brand effects of brand 1 to 4 are as expected. All are negative, which implies a sales increase when the price is reduced. Furthermore, the effect becomes larger when feature and/or display are added to support the price reduction. At practically all stores the effects are significant. Especially when feature and display are both implemented to support the price discount. With this promotional setup, one store shows a non-significant impact at most. Moreover, the promotional effects are also largest for this setup, as the parameter estimates are more negative when a price discount is supported by feature and display. In addition, using a display yields better results than use of feature. Only brand 3 shows a larger sales impact of feature opposed to display, although this result is due one extreme outcome for store 10. Within this store, the β-value for a price discount supported by feature is -26.3, which is almost 3 times larger than the average β-value.

When further inspecting the brand specific size of the effects, brand 3 shows the largest average effect of price promotions on sales, as well as high standard errors. The effect of the price promotions of brand 3 thus fluctuates heavily between stores. This fluctuation is true for all brands that use feature and display simultaneously. The standard error is relatively large for this support type, which implies large differences of the effect when compared between stores.

With respect to the feature and/or display only variables, the results are less convincing. Two different causes underlie the results that are found. First, feature and/or display only are not often used without a price reduction. Second, if used, the effect is not nearly as large as when implemented in combination with a price reduction and thus often not significant. Altogether, only a minority of the stores show significant results when feature and/or display only is implemented as a marketing tool to boost sales. Feature and display only show the most substantial effects. Although still at a small number of stores, the effect is larger than feature or

display only use. This result implies that if brand managers plan to increase sales without use of a temporary price reduction, feature and display used together provides the best results.

(39)

39

4.6.2 Competitor Effects

With respect to the competitor effects, the results are mixed. These mixed results question the generalizability of the parameter estimates. Almost all price index variables (with or without support) are significant at less than half of the investigated stores. Only a price discount supported by feature and display for the competitors of brand 3 and 4 show a substantial number of significant effects (at 21 and 26 stores respectively). This result implies that brand switching only occurs genuinely for some brands, under certain support types. Brand 3 and 4 seem to be weaker brands due to the large number stores at which a negative effect on own brand sales is present when the competition is on promotion. Whereas for brand 1 and 2 the effects are smaller and significant at a smaller proportion of the total stores. What is interesting to find is the strength of the effects. Competitor use of price reduction supported by display has a larger negative effect on own brand sales compared to competitor use of feature or feature and display. This points in the direction of display having the largest effect on brand switching. Although the effect is not consistent over stores or brands, it does imply that some brands are more susceptible to competitor promotional behavior. In this case, brands 3 and 4 are more susceptible.

In contrast, brand 1 and 2 occasionally even benefit from competitor use of promotions. For brand 1, competitor use of display leads to additional own brand sales in 3 stores, and competitor use of feature in 1 store. Brand 2 experiences similar effects, competitor display increases own brand sales in 1 store, while competitor use of feature leads to an increase in 3 stores. Furthermore, when the competition implements feature and display along a price reduction, these brands experience a loss of sales at a smaller number of stores than do brand 3 and 4.

(40)

40

4.6.3 Category Expansion Effects

At the bottom of table 13 the aggregated effect of all brands on total category sales is displayed. Price reductions, with and without support, all show the expected negative signs at the stores were significant effects are found. Use of feature and display to support a price discount shows the largest effect on category expansion. When comparing the use of feature or

display as types of support, use of display shows a larger average effect.

Use of feature only or display only shows mixed results and significant effects are found for a minority of the stores. Simultaneous use of feature and display only, generates more effects that are significant. Additionally, at 9 out of 28 stores the relationship shows the expected sign. Although, as is stated in paragraph 4.6.1, this promotional setup is not used regularly, but when implemented, the effect is often non-significant.

Overall, the results show that use of feature is significant in more stores than use of display. Use of display shows stronger effects, but also shows mild indications of brand switching. As retailers do not benefit from brand switching, in a trade of situation, feature is most beneficial to the retailer. Concluding, feature is more often effective and switching effects are smaller. Category expansion should thus be induced by use of feature advertising.

4.7 Summary of Results

After describing the most important findings that follow from the results, a summary is shown in this paragraph. Additionally, all hypotheses and their support are displayed in table 14. As significant support of the hypotheses is frequently found for a limited number of stores, a column is added which depicts the average number of stores for which a hypothesis holds. First, in summary, the results suggest the following:

Use of price discounts increase short-term own brand sales, especially when the price discount is supported by feature and/or display.

When feature and/or display are implemented without a price discount, the effects are small and often non-significant. This finding suggests that consumers usually only consider (additional) product buying when the brand is on discount.

(41)

41 No substantial effect is found for competitor use of feature and/or display without a price discount. This is in line with the trivial findings for own brand sales effect when feature and/or display only are implemented. Price discounts are considered being a necessity for consumers to switch brands.

With respect to category expansion, use of feature and display simultaneously has the most impact on increase in category sales. Moreover, significant effects are found at all 28 stores.

The effect of a price discount on category sales increases when supported by feature or display, where use of a display has a larger impact than featuring. In addition, use of feature advertising is more often significant.

Use of feature and/or display without price discount does not show substantial effects on category expansion. Only when feature and display are used together the expected positive impact is found at 9 stores.

(42)

42

Table 14 – Hypothesis Results

Hypothesis Relationship Description

Own Brand Effects

H1a - Price discounts of the focal brand lead to higher own brand sales H1b + Feature increases the effect of price discounts

H1c + Display increases the effect of price discounts

H1d + Feature and display increase the effect of price discounts Marketing Effects

H2a + Category feature advertising has a positive effect on market expansion

H2b - Competitor display has a negative effect on own brand sales due to brand switching Competitor Effects

H3a + Price discounts of the competition lead to lower own brand sales H3b + Feature increases the effect of competitor price discounts H3c + Display increases the effect of competitor price discounts

H3d + Feature and display increase the effect of competitor price discounts

Own Brand Effects Result Average # of stores

H1a - Partly supported 21

H1b + Supported 21

H1c + Supported 21

H1d + Supported 27

Marketing Effects Result Average # of stores

H2a + Partly supported 21

H2b - Not supported 6

Marketing Effects Result Average # of stores

H3a + Not supported 4

H3b + Not supported 5

H3c + Not supported 4

(43)

43

5. DISCUSSION

This section describes the conclusions with respect to the conducted research. The main research question and sub-questions are answered. Furthermore, the managerial implications are given. Lastly, limitations and directions for future research are formulated.

5.1 Conclusion

The main goal of this research is finding a promotional setup that benefits both brand manufacturer and retailer. To investigate this, the effects of price promotions are assessed in terms of brand switching and category expansion. Brand switching is beneficial to the brand manufacturer and category expansion is beneficial to both players. To measure brand-switching behavior, own brand and competitor effects are modeled. Category expansion is examined through the effect of price promotions on aggregated total sales across brands within the category.

Price promotions without support are less effective than with support, which is in line with Van Heerde et al. (2004). The effect increases with use of feature, display and feature & display (in that order), which corresponds with the findings of Van Heerde et al. (1999) and Wittink et al. (1988). In contrast, when feature and/or display are implemented without use of a temporary price reduction, the effects are not as clear. Only use of feature and display could help increase sales if brand managers decide not to lower their price. The effect is then largest, while only effective within some stores. With respect to own brand sales it can be concluded that price promotions are most effective if there is a price discount, which is supported by feature and display. Merely lowering the price does not maximize sales potential. Consumers need to be informed about the price reduction by use of advertising and display. This empirical finding matches the findings of Van Heerde et al. (2004).

(44)

44 sales. Concluding, consumers mainly switch brands when feature and display are used in combination with a price discount. However, indications in line with Wittink et al. (1988) and Fok et al. (2005) are found. Display can have large effects on brand switching behavior as a display attracts the attention of a consumer who is already considering a category purchase. In addition, the actual brand choice is still dependent on the brand that is on display.

As retailers do not benefit from brand switching, their main interest is expanding the total category sales. Most effective setup for increasing category sales is a price reduction supported by feature and display. Similar to own brand sales, the effectiveness of price promotions on total category sales increases when supported by feature, display and feature & display (in that order). Although the effect of display on category expansion is larger, featuring a product is more frequently effective.

In sum, for both brand manufacturer and retailer, supporting price discounts by use of feature and display simultaneously is most effective. The sales effects are stronger and mostly significant.

5.2 Managerial Implications

For brand managers to increase their own brand sales, use of price discounts supported by feature and display have the strongest effect. Moreover, this price promotional setup is most frequently effective. Naturally, reducing the price and implementing feature and display comes at a cost. Liquid assets can be insufficient for funding large-scale price promotions including feature and display. In case of a trade-off situation with respect to the support type, use of a display in combination with a price reduction has a stronger effect. What should not be neglected is the store effectiveness. When choosing between feature or display for support, store specifics play an important role in determining the effectiveness. Brand managers must always take this into account and should thus test the marketing effectiveness per store. By examining if an effect is found in a particular store, including the strength of the effect, store specific promotional campaigns that maximize sales can be developed.

(45)

45 Furthermore, brand managers are advised to closely monitor the marketing activity of competitors. Competitor activity can have great negative impact on own brand sales. Sales forecasts should thus be adapted to this competitor activity, as better forecasts can lead to cost savings. In addition, boosting own brand sales by means of price promotions is more effective when implementation is done by one brand only. Own brand promotional effects are less strong when multiple brands are on promotion (Nijs et al. 2001).

Retail managers do not directly profit from price promotions as the margin per sold product is often reduced when on promotion. Inducing a category expansion effects is thus desired, as non-promoted brands are sold with standard margins. As the category expansion effect is largest when feature and display are both implemented together with a price reduction, both retailer and brand manufacturer can benefit from this marketing strategy. However, retailers should manage their margins to ensure the combined effect does not become negative. In other words, the benefits of price promotions should induce a certain ‘minimum of additional sales’ to make up for the loss of the lower margins per product. As price promotional costs and effectiveness differ per brand, retailers should be aware of the consequences of brand differences with respect to their below the line results.

Lastly, in line with Van Heerde et al. (2004), display has a stronger effect on category expansion then does featuring. Although, in case of a trade-off situation, use of feature should be preferred due to display generating possible brand switching which is not beneficial to the retailer. In addition, feature is more often significant.

5.3 Limitations and Future Research

(46)

(47)

47

Acknowledgments

(48)

48

APPENDIX 1 – R-Square Measures

Brand 1 Brand 2 Brand 3 Brand 4 All Brands

R2 Adj. R2 R2 Adj. R2 R2 Adj. R2 R2 Adj. R2 R2 Adj. R2

MUTUALLY BENEFICIAL PRICE PROMOTIONS FOR MANUFACTURER AND RETAILER