Measuring the success of marketing: Examining the effects of price
promotions on brand sales and how they differ between price/quality tiers
Master Thesis
MSc Marketing Intelligence & Marketing Management University of Groningen
Faculty of Economics and Business Department of Marketing
Lowie Hartjes S2396777 15 January 2017
9715 CP Groningen, The Netherlands +31 (0)6 53 45 90 37
l.hartjes@student.rug.nl
Management Summary
Marketing accountability is still considered to be an important subject in marketing literature as well as by managers. Numerous studies try to link marketing instruments to some sort of performance measure, but managers crave for simple measures such as sales. Therefore, this study examines the relationship between brand sales and one of the most popular marketing instruments in practice: price promotions. It provides an in-depth overview of the factors that influence this relationship in the short-run as well as the long-run, such as non-monetary promotions and competitive reactions. In addition, the effectiveness of price promotions between different price/quality tier brands is examined to find out if there are any asymmetric promotion effects. A VAR model is estimated to include dynamic effects and complex
relationships between three national brands and their corresponding price indices.
The results show that price promotions can greatly benefit brand sales in the short-run as well as in the long-run. When accompanied with non-monetary promotions these positive effects can last up to 10 weeks. However, the results do not show any permanent long-run results. Also, no asymmetric promotional effects were found. This means managers of high quality products have no promotional advantage over managers of cheap products or products of which the perceived quality is low and vice versa. In addition, price promotions do barely affect sales of other brands. This is probably due the maturity if the market in which brands already have established a strong position.
1. Introduction 3
1.2 Contribution to literature 5
1.3 Study outline 6
2. Theoretical framework 6
2.1 Direct effect of price promotions on sales (short- and long-run) 6
2.2 The influence of feature and/or display 9
2.3 The influence of the competitive reactions 10
2.4 Difference in promotion effectiveness between price tiers 11
2.5 Duration of the effects of price promotions 12
2.6 Conceptual model 13 3. Methodology 14 3.1 Data description 14 3.2 VAR model 17 3.3 Model estimation 18 3.3 Overview 19 3.4 Unit-root test 20
3.5 Lag length selection 20
3.6 Testing residual assumptions 21
3.7 Granger causality (GC) test 21
3.8 Impulse response functions (IRFs) 21
3.9 Generalized Forecast Error Variance Decomposition (GFEVD) 22
4. Estimation 22
4.1 Unit-root test 22
4.2 Lag length selection 23
4.3 Granger causality test 23
4.4 Testing residual assumptions 24
5. Findings 25
5.1 Impulse response functions (IRFs) 25
5.2 Generalized Forecast Error Variance Decomposition 29
6. Discussion 31
Table 12: Overview of hypotheses 31
6.1 Long-run effects of price promotions 31
6.2 Asymmetric promotion effects 32
7. Managerial implications 32
8. Limitations and future research directions 33
9. References 34
1. Introduction
Estimating the effectiveness of the marketing mix has been a popular topic and resulted in many interesting findings, which are adapted by retailers all over the world. However, retail managers are becoming more and more keen on optimizing the allocation of these marketing resources. Therefore, the accountability of marketing expenditures has a prominent spot in the marketing field. In fact, ‘Accountability and ROI of Marketing expenditures’ was the number 1 research priority at the Marketing Science Institute (MSI) from 2008 until 2010. However, there is not only evidence for the academic attention this subject received, also within firms people started discussing the position of the Marketing department.
According to Ambler (2003), executives only spend 10 percent of their meeting time on marketing, which shows that it receives little attention. In addition, Verhoef & Leeflang (2009) found that marketing has lost ground within the firm and state that decision areas such as pricing and distribution now is covered by other departments of the firm. Their results show that accountability serves as one of the main stimuli of regaining this influence. A Fortune 100 CFO stated in an interview, which Stewart (2008) conducted for his research, that marketeers should be able to demonstrate their value. Otherwise, marketing will remain merely some tactical activities of which the costs should be minimized and will it never have a prominent place within the organization. These findings again stress the importance of being able to link the marketing contributions to firm performance and the increasing interest in this subject.
But also in 2017, marketing accountability still is considered an important subject, since ‘Measuring and Communicating the Value of Marketing Activities and Investments’ is the main goal of the tier 2 priorities according to the MSI. The institute states that all
organizations focus on assuring that all money spent on marketing should count. This line of thinking received even more attention with the recent economic downturn.
awareness, did not turn out to be popular among the questioned CMOs. Obviously, these measures lack popularity since they are also hard to link to revenue.
As mentioned above, sales are generally considered to be a good indicator of the success of a marketing instrument. One of these main instruments that is used and has been used for a long time is price promotions. Since the beginning of the seventies, price
promotions increased in popularity as a marketing tool and take up the largest part of the marketing budget in most of the packaged goods categories (Currim and Schneider, 1991). Ever since, academic literature attained much attention to the effects of temporary price cuts. This led to a fairly good understanding of the short-run effects in sales. However, the exact long-run effects of price promotions are less clear. Findings in marketing literature
concerning this topic seem to differ from each other, which leads to some ambiguity. Factors such as stockpiling, forward buying, other types of promotions, and the reactions of
competitors can influence these long-run effects, which can make it hard to see how price promotions exactly affect sales over time.
The quality of a product also seems to influence the relationship between promotion and sales. Blattberg and Wisniewski (1989) speak of an asymmetric pattern, meaning that when a high quality product is promoted, it affects lower quality products differently than vice versa. This means that managers that want to promote products of lower quality would have a disadvantage compared to competing brands associated with higher quality.
To clarify the long-run effects of price promotions on sales, and to find out if asymmetric effects influence this relationship a model is constructed that includes three brands that each belong to a different price/quality tier. This allows for examination of the direct effect of a brand’s price promotions on its own sales, but also on brands that differ in price/quality. In addition, non-monetary promotions are also included to create a more complete model. The problem statement of this study can be summarized in the following research question:
This research question is divided into several sub-questions:
1. What is the long-term effect of price promotions on brand sales?
2. Do price promotions of brands from different price/quality tiers affect each other differently in terms of sales?
In order to answer these research questions a time-series analysis approach is used to examine a dataset that contains 124 weeks of deodorant (spray-cans only) sales across 5 supermarkets and eights different brands in the years 2003 to 2006. The effects of different types of
promotions are examined: display, feature and price promotions. Two type of effects are discussed in this study: direct effects and indirect effects. The direct effects reflect the immediate influence of price promotions on sales and the impact over time. The indirect effects reflect the possible effects of competitive reactions and other types of promotions. In order to include all these effects, a Vector AutoRegressive (VAR) model is used.
1.2 Contribution to literature
The academic contribution of this study is to provide an overview of the findings of the existing literature concerning the relationship between price promotions and brand sales and testing the generalizability of these findings in a different product category. In addition, although the short-term effects of price promotions are extensively covered by previous literature and quite clear, findings concerning the long-run effects still are not as
straightforward and do differ from each other. This leads to some confusion. Therefore, these differences in findings are discussed in the theoretical framework
section and, eventually, this
study clarifies the long-run effects of price promotions. Attention is also paid to how non-monetary promotions and competitive promotions influence this relationship. In
addition, these effects of price promotions are examined more in depth by looking how they differ between different price/quality tiers. Whereas other literature mostly studied how the quality gap between store brand and national brands influences price promotions
1.3 Study outline
This report is set up as follows: First, the underlying theoretical framework of the research is discussed. Then the measurement methods is introduced as well as the data used. The results are then summarized and discussed. The paper ends with conclusions concerning the
findings, managerial implications, limitations and recommendations for further research.
2. Theoretical framework
The main focus of this study is on the effectiveness of price promotions on long-run sales within one product category and how this effectiveness differs per (high/medium/low) price/quality tiers. Before covering the literature concerning these tiers, first the relationship between price promotions and sales is covered in-depth as well as several important factors that can have an influence on it. The short-run effects of price promotions are shortly discussed and the (potential) effects of the use of other types of promotional instruments, such as feature and display are also examined, which might moderate the relationship between price promotion and sales. Whether or not the competition within the category uses any type of promotions are also discussed as a potential moderator. The literature review also covers the duration of price promotions and what factors can influence it.
The following sections are dedicated to describing previous findings of research concerning the topics mentioned above.
2.1 Direct effect of price promotions on sales (short- and long-run)
Short-run
Many studies already showed that price promotions have a positive impact on short-term brand sales. Walters (1991) and Nijs, Dekimpe, Steenkamp and Hanssens (2001) and
specific brand in store B. However, Walters (1991) found that, effectively, this cannibalization effect did not offset the positive effect of the price promotion.
To give a more detailed picture, Van Heerde, Gupta and Wittink (2003) decomposed the short-term effects. They concluded that these effects are equally divided between three main components: Brand switching, forward buying, and increased consumption. Brand switching means buying the price promoted product instead of the competitors within the category, but a part of it can also be a result of the earlier mentioned cannibalisation. Forward buying entails increased consumption in the period immediately following the price
promotion and relatively less purchases in later periods, which is discussed in the long-run section. Lastly, Van Heerde et al. (2003) refer to increased consumption as the expansion of the product category as well as the growth of the market as a whole. A similar study by van Heerde, Leeflang and Wittink (2004) confirms these findings.
The vast majority of literature concerning this topic found a positive direct relationship between temporary discounts and sales in the short-run, therefore this study does not dedicate much attention to these short-run effects.
Long-run
Although the usage of price promotion has been proven to be lucrative in the short-run, the long-run effects differ in the literature covering this topic.
Nijs et al. (2001) examined the duration of the effects of price promotions, which they call the dust-settling period. They found that in the 58% of the cases where price promotions had a significant short-run effect on demand, this effect could last 10 weeks on average. However, they also found that in almost 40% of the cases the effects are negated in following periods. Pauwels, Hanssens, Siddarth (2002) found that the duration of the effects of price promotions are at most 8 weeks and on average 2 weeks. A reason for this low average according to Pauwels et al. (2002) is that in mature markets, brand choices are in some kind of equilibrium. This means that consumers will return to buying their ‘regular’ brand after trying another brand because of its price promotion. The main reason for this is that, in
mature markets, the effects are limited and quickly offset by the reactions of competitors. The exact effects of competitive reactions is discussed in the section The influence of the
competitive reactions
. They do however state that other types of promotions are not taken
Pauwels et al. (2002) also mention that the total amount of category sales does increase in the long-run. This means that over time, people that did not buy from a certain category might buy from that category due to price promotions, which can be an explanation of the increase in brand sales.
Other reasons why price-promotions are more effective in the short-run than in the long-run are stockpiling and forward buying, also classified as quantity and timing
acceleration (van Heerde, Leeflang and Wittink, 2000). This means that when there is a price promotion, people start buying more products than they need at that moment or even buy the product while it is not needed at the moment. Consequently, these consumers will not need to buy that product for a while and therefore, the long-run sales will decrease. Zeelenberg and van Putten (2005) state that besides stockpiling and forward buying, there is another
explanation for the post-promotion dip. They found that people who regularly buy a product of a certain brand, but miss out on a price promotion on that product, sometimes do not buy the product at all anymore or even switch to another brand. This type of post-promotion dip is called inaction inertia; if people feel like they missed out on an attractive opportunity, a less attractive one is not taken. Zeelenberg et al. (2015) even call this a ‘dark side’ of price promotions, because it may result in a bigger decrease in sales in the long-run than an increase in sales in the short-run.
Nijs et al. (2001) add that the frequency of price promotions also plays an important part. They conclude that if a store uses more price promotions, they increasingly become a greater driver of the motivation for consumers to buy the product. This, in turn, results in consumers relying on future price promotions when they purchase. Van Heerde et al. (2000) state that this can lead to pre-promotion dips.
Thus, the literature covering this topic do not agree on the long-run effects of price promotions. The findings range from positive long-run effects, to a fast negation of the short-term effects, to even negative effects. However, the majority of the studies conclude that temporary price cuts do not have any effect in the long-run. Therefore the second hypothesis is:
2.2 The influence of feature and/or display
Besides price promotions, marketing literature also shows that other types of promotional instruments, such as feature and display, can influence sales. Therefore, this section discusses the findings of previous research concerning these two instruments and how they can
influence the price promotion effectiveness.
Direct effect on brand sales
Kumar and Leone (1988) studied whether price promotion, featuring or display has the largest effect on brand sales. They found that after price promotion, featuring explained 12% of the increase in brand sales and displays explained 4%, when not considering the
interaction effects. The main explanation of the positive direct effect of non-monetary promotion (such as features and displays) on brand sales is that it differentiates a brand from the other ones in the product category (Nijs et al. 2001). As well as the short-term effect price promotions, both featuring and displaying have been extensively proven to increase brand sales in the literature. Therefore, this study does not dedicate much attention to the
explanation of these direct effects.
Moderating effect on relation between price promotion and sales
Van Heerde et al. (2000) tested 4 different price promotion scenarios for tuna and tissues. One scenario in which there was only a price promotion, one with price promotions combined with feature, one combined with display and one scenario including all types of promotions. They found that all combinations worked better than solely a price promotion in both categories. In 2005, Horvath et al. found that the unit sales impact of supported price promotions can be double as high as the impact of price promotions not accompanied with non-price promotions.
Although several studies have found that supported price promotions have a larger effect on sales than unsupported price promotions, many do not provide concrete
a non-price advertising message, since it can increase consumers’ price sensitivity. Hence, they will react more favourably to price reductions.
To summarize, price promotions supported by feature and/or display are expected to be more effective since these non-monetary promotions will consumers inform about these onsale products, where to find them and can increase price sensitivity.
2.3 The influence of the competitive reactions
As mentioned previously, competitive reactions can also influence sales and price promotion effectiveness. Leeflang and Wittink (1992) refer to a basic competitive reaction when a competitor reacts with the same marketing instrument as the initiating firm used. For
instance, if brand A were to decrease its price temporary, brand B would follow by doing the same. In addition, Leeflang et al. (1992) also distinct a multiple competitive reaction, which in this case can mean that brand B would not solely react with a price decrease, but also with a non-monetary promotion. Horvath et al. (2005) add to these types of reactions that
managers closely watch sales and if they decrease it can also lead to marketing instrument reactions. However, they also mention that managers follow competitors’ performance measures such as sales. If these managers observe that competing brands do well in terms of these measures they might interpret this as a threat and thus it also can cause reactions. Therefore, Horvath et al (2005) include these types of reactions as feedback effects in their model.
Nijs et al. (2001) did not find significant evidence for effects of the reactions of competitions on promotion effectiveness, since the most dominant form of reactivity in their research was no reaction at all. When this topic was revisited by Steenkamp et al. (2005), similar results were found: Competitive responses are often passive. However, when a competitor does react, it usually uses the same type of promotion that was used by the initiator. For instance, price promotions are countered by price promotions. Eventually, they found that the effects of promotions generally are not influenced by competitive reactions. However, Pauwels (2007) did find a small drop in the effectiveness of price promotions in terms of sales. They conclude that it will decrease with ten percent in the long-run when competitors do react with price promotions.
advertisements will increase the relative importance of non-price buying motivations. As mentioned before, it can be used to differentiate a brand from others and will make
consumers less sensitive to price promotions. This is confirmed by Chandon, Wansink and Laurent (2000). They split up why consumers are attracted to promotions into two types of benefits; hedonic and utilitarian benefits. According to them, monetary promotions (such as price reductions) result in relatively more utilitarian benefits than hedonic benefits.
Non-monetary promotions will make consumers focus more on hedonic benefits than utilitarian benefits, and therefore decrease the effectiveness of price promotions.
To summarize, if competitors do react with price promotions, it is expected to
negatively influence the effects of price promotions. In addition, when competition within the category uses unon-monetary promotions, consumers will focus more on non-price benefits. Therefore, it is expected that the combination of competitive price promotions and
competitive non-monetary promotion negatively influence the effect of price promotions. Although the direct effect of competitor reactions is not considered in the model used in this study, it is important to take their indirect influence into consideration when interpreting the findings.
2.4 Difference in promotion effectiveness between price tiers
How the reactions of competitors can influence the relationship between own price promotions and own brand sales has now been discussed. However, within a product category with multiple brands, prices can differ drastically. This leads to different
price/quality tiers. Therefore, this section discusses literature concerning these price/quality tier brands and if they affect each other’s sales differently when using price promotions.
perfectly with the quality perceived by the consumers. This is also something to take in consideration in the interpretation of the results in this study, since in this research price also serves as a proxy for quality.
A possible reason for the difference in this asymmetric pattern of price competition is that high price (and high quality brands) can temporarily lower their price while keeping its high perceived quality image, whereas a low-tier brand cannot increase its perceived quality, but only decrease its price. Bronnenberg and Wathieu (1996) provide the following
explanation: if a low quality brand uses a price promotion, customers of similar or lower price brands will increase their purchases of that specific brand. However, customers of high quality brands do not switch to the promoted low quality brand because they were
quality-sensitive enough to buy a high quality brand in the first place. In other words, high quality customers do not settle for lower quality products even if they are even cheaper during a promotion, because these products’ quality still do not match their standards.
In contrast to Blattberg and Wisniewski (1989), Bronnenberg and Wathieu (1996) found that a higher quality/price brand does not automatically have an promotion advance over lower quality/price brands. They mention that it only is the case if the quality gap between the brands is large enough relative to the price gap. This means that if a highly priced brand is considered to be overpriced compared to the standards achieved by other brands in the category, its promotions effectiveness decreases. If a brand is underpriced, this effectiveness increases. This means promotion asymmetry is not unconditional and this should be taken into account when interpreting the results.
Essentially, promotion asymmetry can give high quality brands a promotion advantage over lower quality brands and limits brands on the lower side of the quality spectrum. Based on these findings the last three hypotheses are:
H2:High-tier price promotions lead to a decrease in mid- and low-tier brand sales
H3:Mid-tier price promotions decrease only low-tier brand sales and do not affect high-tier brand sales
2.5 Duration of the effects of price promotions
In addition to the focus on long-run effects and difference between price/quality tiers, this study also examines the duration of the effects of price promotions on sales. As mentioned before, Pauwels et al. (2002) and Nijs et al. (2001) found that this period can be up to 8 and 10 weeks respectively. Two earlier studies also examined the duration of promotions. Mela, Gupta, and Lehmann (1997) found that the effect could last for 33 weeks and Mela, Jedidi, and Bowman (1998) found 21 weeks. Both studies examined nonfood products, just like in this study, which might be a reason for these relatively long periods. Pauwels et al. (2002) examined how long the effects lasted for both a perishable product (yoghurt) and a storable product (canned soup) and compared these durations. They confirm that storable products are longer affected by price promotions than perishables are. Their explanation for this difference in duration is related to the earlier discussed stockpiling. Perishable products need to be consumed relatively quickly compared to storable products, therefore they show stockpiling limitations.
What factors determine the exact duration of a price promotion has not received much attention in the literature. However, following the findings of previous literature presented in the sections discussed before, factors such as competitive reactions and non-monetary promotions are expected to be important drivers of this duration.
2.6 Conceptual model
Figure 1 visualizes the model of this study as well as the corresponding hypotheses. H1 reflects the expected indifferent long-run relationship between price promotions and long-run brand sales. The rest of the hypotheses all concern the different price/quality tiers. H2 entails the negative relationship between high-tier price promotions and both mid- and low-tier brand sales. H3 concerns the negative relation between mid-tier price promotions and
Figure 1: Conceptual Model
3. Methodology
3.1 Data description
In order to answer the hypotheses and eventually the research question, a data set is used which contains sales and sales-related data of nine deodorant brands at five different Dutch supermarkets. Table 1 provides an overview of the available deodorant brands and
supermarket chains. It includes weekly data in a period from week 46 in 2003 until week 12 in 2006, which is a total of 124 consecutive weeks.
Supermarket chains Brands
Albert Heijn Dove Vogue
Edah Fa 8X4
Super de Boer Niva Axe
Jumbo Rexona
C-1000 Sanex
Table 1: Available supermarket chains and brands
Volume of band sales
Figure 2: Evolution of total sales per brand (aggregate market level)
Each number on the x-axis represents a certain week and consists of, first the last two digits of the year, followed by the week number. Note that per brand the sales at each supermarket chain is summed up, as indicated on the y-axis. There does not seem to be a clear pattern over time. However, there are many spikes in the sales volumes, which might be explained by promotions and/or seasonality (not included in this study). The upcoming analyses try to explain these increases and decreases in figure 2.
Price variable
The available data set includes a regular price as well as a promotional price. Both price variables are combined into one price variable called price index (PI) by dividing the regular price by the actual price, such that PI equals one when there is no promotion and exceeds one when there is a promotion in that week. This new variable does show a number lower than one a few times. However, these frequency of these odd numbers are is rather small and across all the supermarket chains. Therefore, these numbers would not harm the results of the analyses.
determined which 3 three brands should be included in the final model. Table 2 shows the average actual price per brand per chain. As can be seen, Fa has the lowest price, Axe the highest and both Rexona and Vogue are in the middle in terms of price.
Chain Fa 8X4 Sanex Vogue Rexona Dove Nivea Axe
Albert Heijn 1,70 1,78 2,16 2,27 2,41 2,50 2,90 3,03 Edah 1,64 1,94 2,18 2,07 2,44 2,43 2,65 2,99 Super de Boer 1,80 2,01 2,32 2,28 2,55 2,63 2,99 3,16 Jumbo 1,53 1,71 2,03 1,99 2,19 2,22 2,45 2,80 C-1000 1,68 1,87 2,28 2,18 2,39 2,35 2,58 2,95
Table 2: Price per brand per chain
Price war
During the period the data was collected, Albert Heijn initiated a price war. The other
supermarkets included in the dataset all participated and lowered their prices. Figure 3 shows the evolution of the mean price of all brands. Around 50th week (out of 124 weeks) the prices dropped as a result of the the price war.
Although the price war is expected to influence sales, the model used in this study does not account for these possible effects, since it would increase the complexity of the final model greatly.
Feature and display (and combination)
The Price variable
section made clear which brands can be considered to be high (Axe),
middle (Rexona and Vogue) and low tier brands (Fa) in terms of price. However, this study also takes into consideration the effects of the following non-monetary promotions: Feature, display and the combination of both. Feature means special attention for the product outside of the store, whereas display provides attention for the product in the store. Both are dummy coded on the individual store level. Similarly, the simultaneous use of both advertising instruments is also dummy coded.
For these effects to be included in the model properly, the corresponding brands should in fact participate in these types of promotions. Therefore, graph 3 shows the number of different promotions per brand at an aggregate market level.
Figure 4: Number of different promotions per brand (aggregate market level)
To create an equal amount of brands per price tier, it is decided to choose Rexona to be the representative of the middle price tier due to the more frequent use of all types of promotions compared to Vogue.
Chain
In order to simplify the final model, one out of the five chains is selected. Using similar reasoning as for the selection of brands in the previous section, the chain that participates the most active in terms of promotions is included. Therefore, Edah is chosen. It fulfills this criterion for every type of promotion.
3.2 VAR model
Several studies have applied VAR modeling to examine the influence of price promotions on sales. In 2001, Nijs, Dekimpe, Steenkamp and Hanssens conducted a similar study
researching the effects of price promotions in terms of category sales. Horvath, Leeflang, Wieringa, Wittink (2005) used similar modeling to examine the competitive reaction effects, and Slotegraaf and Pauwels (2008) examined the long-run effectiveness of promotions also using VAR modeling.
VAR modeling usually starts by choosing the relevant variables and transformations, and deciding which variables to make endogenous and which ones to make exogenous (Horvath et al. 2005). Variables that are of particular interest commonly are chosen to be endogenous, therefore price index and sales are chosen to be endogenous in this case. The other variables then can be controlled for by considering them to be exogenous in a VARX model. Variables that are considered exogenous are the non-monetary variables: feature and/or display. The lagged endogenous variables (price indices and sales), allow for the dynamic effects of variables that are considered exogenous in this study. Of course, in the most ideal situation, all available variables would be included in the model as endogenous. However, this might result in an extreme case of overparameterization (Nijs et al. 2001).
3.3 Model estimation
Formula (1) represents the formula in its structural form (1) t = week p = number of lags
= 6x1 vector including the endogenous variables in natural logarithm form at t
Yt A
= 6x1 vector containing the intercepts for each of the equations
= matrix containing parameters for all included endogenous variable at t-p ϕ
= 9x1 vector with the exogenous variables β
= 6x1 vector containing the error terms Σt
Both sales and the price index are log-transformed variables, because this accommodates multiplicative interaction, it simplifies the estimation and it increases the variance of trending time series (Wieringa and Horvath, 2003)
Formula (2) represents the same formula, however including the full vectors and matrix.
(2)
lnS
= natural logarithm of the sales for each of the three brands
LnPI
= natural logarithm of the price index for each of the three brands
F
= feature
D
= display
FD = combination of feature and display = intercept
α
3.3 Overview
In order to perform a analysis using a VAR a number of steps are required. This section provides a short overview over these necessary steps. After that, each step is elaborated on in the sections that follow
These steps are:
(1) Unit-root testing is needed to determine if the endogenous variables are stationary or evolving
(2) In order to determine how many lags should be included in the analysis by examining lag length criteria
(3) The unrestricted VAR model as it is presented in the model estimation
section is
estimated
(4) A Granger-Causality test is performed to see if lagged values of endogenous variables explain other endogenous variables
(5) Residual assumptions are checked concerning: non-normality, heteroscedasticity and serial autocorrelation
(6) The model is estimated once again, including the possible restrictions
(7) Impulse response functions graphs are used to see how the different brands sales are affected by price promotions
(8) A Generalized Forecast Error Variance Decomposition is used to determine the dynamic explanatory value of the endogenous variables
3.4 Unit-root test
As already mentioned in the overview
, unit-root testing is necessary to determine if there is an
3.5 Lag length selection
Before estimation using the VAR models, it is important to select a proper length for the included lags. Hanssens (1980) states that this is of great essence, because including the wrong number of lags can result in no Granger causality among the endogenous variables. This type of causality is explained later on in this study. Several methods can be used to determine a suitable number of lags. Gredenhoff and Karlsson (1999) compared these methods and found that the AIC and HQ criteria estimate the correct number of lags better than the BIC since they are less parsimonious. Liew (2004) also compared a number of criteria and concluded AIC and FPE minimize the likelihood of underestimating and maximizing the chance of finding the true amount of lags, especially when dealing with a small sample. Therefore, determining the right lag length for the model used in this study mostly depends on the AIC, HQ and FPE.
3.6 Testing residual assumptions
Similar to an ordinary least square, a number of assumptions concerning the residuals need to be tested when using a VARX model. These assumptions are displayed in table 3.
Assumption Method used to test assumption Literature
1. No serial autocorrelation Lagrange Multiplier (LM) Breusch & Pagan (1980) 2. Normal distribution of residuals Jargue-Bera test Jargue and Bera (1980) 3. Presence of homoscedasticity White Heteroskedasticity test White (1980)
Table 3: Overview of residual assumptions and corresponding tests
3.7 Granger causality (GC) test
After testing the residual assumptions, it is necessary to perform the Granger causality test. This test serves to find out of if the dependent variables are granger-caused by the other variables in the model. (Granger, 1969) The essentially means that if variable X causes variable Y then lags of X should be significant in the equation for Y. Lutkepohl (2007) states that the Granger causality test effectively examines if the inclusion of the history of X has additional predictive power compared to solely predicting Y based on its own history. This can be summarized in the following formula:
(Y |Y , = (Y |Y )
So, the GC-test regresses every single variable in the model on its own lagged value as well as the lagged value of other endogenous variables. Rejecting the null hypothesis in this case means that there is significant evidence that X does Granger-cause Y.
3.8 Impulse response functions (IRFs)
In order to examine the effect if a temporary price discount on brand sales and to allow for dynamic effects, IRFs can be used (Horvath and Wieringa, 2003). This method proves to be considered common practice in marketing VAR modelling, since several authors applied this method. (e.g., Horvath et al. 2005, Pauwels et al. 2002, Nijs et al. 2001 and Srinivasan et al. 2010). The IRF utilizes the estimated parameters of the VARX model to determine the net effect of a certain shock on the variables of interest. This effect is relative to their mean value in absence of this shock. Thus, essentially, by using the IRF the effect of a temporary price reduction on long-run brand sales can be measured (Srinivasan et al. 2010).
In order to find the duration of the effect of price promotions the IRFs can also be used. The time it takes for the IRF to stabilize after the initial shock, which existing marketing literature refers to as the ‘dust-settling period’ (e.g. Nijs et al. 2001). Impulse response functions can converge to zero over time, however, it is possible this does not happen and that they remain constant at a non-zero level. The former representing no permanent long-run impact and the latter means there is a long-run persistent impact caused by the price promotion (Dekimpe and Hanssens 1999).
3.9 Generalized Forecast Error Variance Decomposition (GFEVD)
4. Estimation
4.1 Unit-root test
Table 4 shows the results of the augmented Dickey Fuller test. As can be seen, every endogenous variable proves to be stationary, meaning that they do not contain a unit root, since all are significant. This allows for continuing with the analysis with any transformation of the variables.
Variables LnSalesAxe LnSalesRexona LnSalesFa LnPIAxe LnPIRexona LnPIFa Intercept 0.0000** 0.0250* 0.0000** 0.0000** 0.0000* 0.0000**
Trend & intercept
0.0000** 0.0403* 0.0000** 0.0000** 0.0000* 0.0000** H0 = The variable has a unit root. Significance level: **p<0.01, *<0.05
Table 4: Unit-root test: Augmented Dickey Fuller
4.2 Lag length selection
In order to find the right lag length, several criteria are examined for a maximum of 10 lags, displayed in table 5. Both the FPE and the AIC indicate three lags, SC and HQ indicate only one lag and the LR shows 9 lags. As mentioned before, the HQ, AIC and FPE are proven to be proper lag length selection criteria in the context of this study. Since the VAR(3) model is favoured by two of these three criteria and the VAR(1) only by one of them, a model
including three lags is the most suitable option to continue the analyses with.
Lag LogL LR FPE AIC SC HQ
0 568.0497 NA 5.44e-12 -8.913152 -7.473048 -8.328694 1 677.4072 188.0182 1.51e-12 -10.20013 -7.895959* -9.264994* 2 710.9966 54.21457 1.61e-12 -10.15784 -6.989606 -8.872029 3 753.2074 63.68641 1.49e-12* -10.26680* -6.234504 -8.630315 4 769.2102 22.46005 2.23e-12 -9.915968 -5.019613 -7.928812 5 788.0352 24.43949 3.24e-12 -9.614653 -3.854235 -7.276822 6 812.9904 29.77111 4.37e-12 -9.420884 -2.796404 -6.732379 7 849.5153 39.72880 5.00e-12 -9.430092 -1.941549 -6.390912 8 889.7233 39.50266 5.63e-12 -9.503918 -1.151312 -6.114063 9 964.5941 65.67608* 3.67e-12 -10.18586 -0.969192 -6.445331 10 997.2377 25.19863 5.43e-12 -10.12698 -0.046247 -6.035774
4.3 Granger causality test
Table 6 shows which variables Granger cause which at a significance level of 0.05. The sales of FA do Granger cause the sales of Axe and Rexona. FA’s price index Granger causes Axe’ sales as well as Rexona’s price index, which is also Granger caused by the sales of Axe. The model does not include any bi-directional causality. None of the brands’ PI Granger cause its corresponding sales. This means the history of price promotions of a brand does not provide additional predictive power in predicting its sales. In order to find out if there is any
relationship between PI and sales, a correlation test is performed. This shows that there is in fact an moderately to strong correlation between them.
Dependent variables Lags of
variables
LnAxeSales LnRexonaSales LnFaSales LnPIAxe LnPIRexona LnPIFa
LnAxeSales X LnRexonaSales LnFaSales X X LnPIAxe LnPIRexona LnPIFa X X
Table 6: Output Granger causality test (3 lags). X indicates a significant Granger causality
4.4 Testing residual assumptions
Serial autocorrelation
Table 7 displays the outcome of the Lagrange Multiplier test for the VAR(3) model. As can be seen, for none of the lags the LM-stat is significant. This means the null hypothesis can not be rejected, hence there is no autocorrelation among the residuals. Therefore, the estimation can continue with 3 lags.
Lags LM-Statistic Probability
1 33.90864 0.5684
2 33.38514 0.5936
3 28.34477 0.8147
Normal distribution of residuals
This residuals assumptions concerns whether or not the residuals are normally distributed. As mentioned before, the Jarque-Bera test can be used for this. Its output is shows in table 8
Variable Jarque-Bera Prob.
LnAxeSales 225.0352 0.0000 LnFaSales 103.1798 0.0000 LnRexonSales 37.39192 0.0000 LnPIAxe 9812.054 0.0000 LnPIFa 23.18468 0.0000 LnPIRexona 62.00774 0.0000 Joint 10262.85 0.0000
Table 8: Output of the Jarque-Bera test
Table 8 shows that the joint JB value is significant (<0.05) and that every individual component also is significant. Therefore, the null hypothesis that states that normality is present in the residuals is rejected. This means this assumption is violated, which should be taken in consideration when interpreting the results.
Testing for homoscedasticity
The output of the White Heteroskedasticity test in appendix A shows that the joint probability is significant (<0.05). This means the data suffers from heteroskedasticity. When looking at the individual components, 8 out of 21 are significant. Due to this presence of
heteroskedasticity, it is important to be careful when interpreting the final results.
5. Findings
5.1 Impulse response functions (IRFs)
In order to find out what the direction and timing is of the effects of one endogenous variable on the other the impulse response function can be used. These IPFs are based on the
Effects of own price promotions
H1 states that the positive effects of price promotions are negated in the long-run. In order to find out if this hypothesis can be accepted or rejected the impulse response functions of the three brands are examined. They show a response to a change in their corresponding price indices. These IPFs are shown in figure 5.
Figure 5: Response of the three brands to a change in their corresponding price indices.
The first graph of figure 5 shows that the sales of Axe respond favourable to an increase in the price index, which represents a price promotion, in the first week. Apart from this positive direct effect, long-run effects also are present. Just before week 3, an increase in LNAXEPI still positively influence LNAXESALES. This also is the case for both FA and Rexona, whereas the sales of FA still experience a positive effect until week 10 as response to an increase in the price index of FA. However, after week 10 the sales of FA remain just below zero.
The graphs in figure 5 indicate that price promotions do result in an increase in sales in the short-run, but also in the long-run. This effect is even positive in most cases.
Effects of price promotions of brands from other tiers Effects of price promotions of high tier brand
Figure 6: IRFs of FA sales and Rexona sales as a response to change in Axe price index
The left IRF in figure 6 shows that a price promotion of Axe results in a small bump in sales of FA in the second week followed by a larger decrease in sales in week 4. After that, FA sales gradually increase until they stabilize around zero in week 9 and in the following period.
The right graph in figure 6 indicates an immediate increase in sales of Rexona as a response of a change in the price index of Axe. In the following period these positive effects remain visible, but rather small.
The results show that the sales of the low-tier brand are negatively affected by a price promotion of the high-tier brand. However, the mid-tier brand is positively affected. This results in only partly accepting the second hypothesis.
Effects of price promotions of medium tier brand
The left IRF in figure 7 indicates that as a response to an increase in the price index of Rexona, the sales of Axe initially increase followed by a sharp drop and eventually stabilizing around zero.
Surprisingly, according to the right IRF in figure 7, the sales of the FA respond positively to a price promotion of Rexona. This effect even remains during a substantial amount of weeks after which it also drops to zero.
The results show that the sales of the high-tier brand are affected (mostly negatively) by an increase of the price index of the mid-tier brand. The sales of the low-tier brand is only affected positively by this change. Therefore the third hypothesis is not accepted.
Effects of price promotions of low tier brand
Figure 8: IRFs of Axe sales and Rexona sales as a response to change in FA price index
The left IRF in figure 8 shows that a change in the price index of FA causes the sales of Axe immediately to decrease followed by an even larger drop in the third week. However, in the following week sales increase sharply and the response remains positive for a long period.
According to the right IRF in figure 8 the sales of Rexona initially react positively to a price promotion of FA followed by an negative change in sales in week 2 and 3, a positive one in week 4 and eventually remaining negative in the following weeks.
5.2 Generalized Forecast Error Variance Decomposition
Sales of Axe
Period LNAXEPI LNAXESALES LNFAPI LNFASALES LNREXONAPI LNREXONASALES
1 24.65550 75.34450 0.000000 0.000000 0.000000 0.000000 2 24.12723 70.04203 0.860118 4.181603 0.023905 0.765118 3 22.18516 65.13505 6.003431 4.352441 1.575194 0.748720 4 19.08387 59.27149 14.41612 3.884613 2.403880 0.940022 5 17.94398 57.23430 15.38893 5.773904 2.734546 0.924340 6 17.86969 57.07960 15.41898 5.799131 2.910139 0.922466 7 17.33669 55.68277 16.93638 5.734826 2.968771 1.340560 8 17.07301 54.95852 17.25274 6.021203 2.940175 1.754349 9 16.93328 54.54446 17.44747 6.032238 2.922732 2.119824 10 16.75777 54.01227 17.66725 5.991420 2.897987 2.673297
Table 9: Variance decomposition of sales Axe
Sales of FA
Period LNAXEPI LNAXESALES LNFAPI LNFASALES LNREXONAPI LNREXONASALES
1 0.000721 0.169487 62.84811 36.98169 0.000000 0.000000 2 0.224811 0.215675 61.25893 38.20959 0.059379 0.031615 3 0.245351 0.207128 65.25039 33.91188 0.335464 0.049793 4 0.946889 0.189094 63.95055 32.83644 0.310459 1.766576 5 1.171052 0.218176 62.44895 33.12101 0.367815 2.672999 6 1.171715 0.294966 61.66850 32.46244 0.368094 4.034282 7 1.166017 0.293733 60.56044 31.93346 0.388148 5.658205 8 1.149320 0.308542 59.49911 31.56403 0.441469 7.037527 9 1.135170 0.333583 58.71036 31.16826 0.462005 8.190621 10 1.124003 0.337990 58.02877 30.80899 0.478330 9.221913
Table 10: Variance decomposition of sales FA
Sales of Rexona
Period LNAXEPI LNAXESALES LNFAPI LNFASALES LNREXONAPI LNREXONASALES
1 2.227620 0.646696 6.243534 4.744720 37.47640 48.66103 2 3.368184 0.714069 6.206500 4.605631 36.19467 48.91095 3 3.109310 1.599068 6.075509 8.368187 30.94804 49.89989 4 2.825835 1.771961 6.121028 7.541220 28.70785 53.03211 5 2.757017 1.947702 5.942035 7.227252 27.58195 54.54405 6 2.908195 2.202852 5.823959 8.838726 26.28316 53.94311 7 2.832488 2.234325 5.738809 9.463149 25.43654 54.29469 8 2.812933 2.288210 5.997829 9.903971 24.93056 54.06650 9 2.851423 2.375132 6.253879 10.85905 24.34741 53.31311 10 2.866146 2.397278 6.390262 11.68469 23.93244 52.72919
Table 11: Variance decomposition of sales Rexona
The variance in the sales of Rexona in table 11 are mostly explained by themselves, as well in the first periods as over time. Rexona’s PI also has a large share in explaining the variance. However, this share decreases over time along with the increase of the share of FA’s sales.
6. Discussion
Table 12 gives an overview of the hypotheses and if they are (partly) accepted or not
Hypothesis Support Description
H1: The positive short-run effects of price promotions on sales are negated in the long-run
No
H2:High-tier price promotions lead to a decrease in mid- and low-tier brand sales
Partly High-tier price promotions only decrease sales of low-tier brands (but not substantially)
H3:Mid-tier price promotions decrease only low-tier brand sales and do not affect high-tier brand sales
No
H4: Low-tier price promotions do not affect mid-tier nor high-tier brand sales
No
6.1 Long-run effects of price promotions
The IRFs show that sales are affected by price promotions in the long-run. However, this seems to contradict the outcome of the Granger Causality test. This test indicates that the history of a brand’s price index does not significantly influence its sales. A possible
explanation for this seemingly contradiction might be that although the past values of FA’s PI do not influence FA’s sales, a change in FA’s PI might directly affect other factors positively, which in turn results in an higher level of sales in a number of subsequent periods. As
mentioned in the literature review, a non-monetary promotion can be one of these factors since it extends the positive effect on sales initially caused by price promotions. Consumers who tried a price promoted product might be reminded to keep buying this product in the long-run due to these non-monetary price promotions.
Although the price promotions of a brand cause short- and long-run increases in sales of that specific brand, the results show that these effects are not permanent. The IRFs show that the sales stabilize around zero after maximum 10 weeks. Possible reasons for the lack of permanent effects are also discussed in the literature review. The deodorant market can be considered a mature market, in which brand choices eventually return to an equilibrium. Hence, no permanent deviation of the initial market shares is likely in the long-run. Another reason for negation of promotion effects over time can be the earlier discussed competitive reactions. Once one brand is promoted, competitors can react by matching it or using other types of promotions. As found in the literature, these reactions negatively influence the initial price promotion and can eventually cancel out its effect.
The generalized forecast error variance decomposition also provides some important insights. Generally, the largest part of the variance in sales of a brand is explained by the past values of its own sales and own price index especially in first weeks. A reason for this can be that brands have taken an quite strong position in the market. This makes it hard for factors, other than the history of own sales and price index, to influence these brands and move them from their established position. This again ties back the probable equilibrium in the mature market of deodorants. Hence, it is hard to distort this equilibrium.
because they know that it is likely that there also will be a price promotion for their preferred brand in the near future. Therefore, they will hold off on purchasing until this happens.
The results do not show any signs of post-promotion dips. However, according to the literature discussed, these dips often do occur after stockpiling and forward buying. This lack of negative effects in the long-run might give reasons to believe that the increase in sales are not a result of quantity/timing acceleration. In addition, the results also do not provide evidence for brand switching as an explanation for the positive effect of price promotions. The decomposition of price promotion effects of Van Heerde et al. (2003) provides a third and most plausible reason for the solely positive effect on sales: increased consumption. This means that sales are boosted due to a growth of the category as well as the market as a whole. When looking at the IRFs, this would mean that after approximately 10 weeks that follow the price promotion, the category/market returned to its original size. Hence, the IRFs stabilize around zero after this period. Therefore, the results show that increased consumption is the most probable explanation for positive effect of price promotions and the absence of post-promotion dips.
6.2 Asymmetric promotion effects
achieved in the deodorant market have become reasonably high. So high that the more expensive products do not exceed these standards enough in terms of quality for asymmetric promotion effects to be present.
7. Managerial implications
Managers are keen on measuring how effective marketing actually is for their firm. This study provides an overview of what factors can influence this effectiveness of price
promotions in terms of brand sales. The results show that temporary lowering the price of a product can have great benefits for that sales of that product in the short-run but also in the long-run. However, managers should note that, in the long-run, price promotions do not directly affect sales, but do so indirectly through other factors. Therefore, it is important to accompany price promotions with other marketing instruments that positively influence they impact. Such instruments can be non-monetary promotions, since they can extend the effect of price promotions. These temporary price cuts do not result in permanent sales effects. The duration of the positive effects of price promotions can be up to 10 weeks. However, it should be noted that this study examined storable products. The duration of these effects can be significantly shorter for perishable products.
This study also shows that asymmetric promotion effects are not present in the context of only national brands. This means that there is no competitive difference in the price promotion effectiveness between different price/quality tiers. Therefore, managers of high quality products have no promotional advantage over managers of cheap products or products of which the perceived quality is low and vice versa. This knowledge provides managers with a better understanding of what impact price promotions can have.
Managers should also note that in mature markets, price promotions do not affect the sales of other brands greatly. This is do to the strong position which brands have established over time. Consumers are used to a certain brand and are not easily persuaded to switch brands. Managers will need to do more than solely promote their brand in order to accomplish this.
now. The most probable reason for this is that price promotions cause the category and market to expand, rather than stealing away sales from other brands or quantity/timing acceleration.
8. Limitations and future research directions
There are a few limitations concerning this study. In the period when the data was collected a price war started. As already mentioned in the data description
section, this price war was not
accommodated in the model, because it can make it overcomplicated. However, the prices of the deodorants dropped due to the price war. This might have influenced the effectiveness of price promotions on sales.
Also, this study tries to examine whether there is a difference in price promotions effectiveness between three quality tiers. Which brand belongs to which quality tier was based on their price, so price serves as a proxy for quality. As discussed in the literature review, this relationship is a plausible one. However, price and quality are not found to be correlating perfectly. Future research can attempt to use alternative proxies for quality that show a higher correlation with quality itself.
Another limitation is that the dataset contains only one product category, which limits the generalization of the results. As mentioned before, price promotion effectiveness can differ between perishable and storable product categories. Therefore, future research should examine both types of product categories and look if price promotions effectiveness differs even more between quality tiers when comparing storable and perishables.
The results show that there are no asymmetric promotion effects in the deodorant market. As discussed, this might be due to high quality standards in this particular market, which have been established over time. Future research can examine if this also is the case in a relatively new product category in which the gaps between brands are higher in terms of quality.
Therefore, future research should focus on what the exact drivers are of long-run effectiveness of price promotions are.
Also, seasonality is not accommodated for in the model. However, different seasons might influence the sales of deodorant differently. Since they are often used to prevent perspiration, it might be that these products are sold more often in the summer compared to the winter. Therefore, seasonality might also influence how price promotions affect sales.
Lastly, two assumptions of the residuals are violated. The results show that they suffer from heteroskedasticity and nonnormality, which can have an impact on the rest of the
results.
9. References
Ataman, M. B., Van Heerde, H. J., & Mela, C. F. (2010). The long-run effect of marketing strategy on brand sales. Journal
of Marketing Research
, 47(5), 866-882.
Ambler, Tim (2003), Marketing and the Bottom Line, 2d ed. Edinburg: Pearson Education
Blattberg, R. C., & Wisniewski, K. J. (1989). Price-induced patterns of competition. Marketing science
, 8(4), 291-309.
Bronnenberg, B. J., & Wathieu, L. (1996). Asymmetric promotion effects and brand positioning. Marketing science
, 15(4),
379-394.
Chandon, P., Wansink, B., & Laurent, G. (2000). A benefit congruency framework of sales promotion effectiveness. Journal
of marketing
, 64(4), 65-81.
Currim, I. S., & Schneider, L. G. (1991). A taxonomy of consumer purchase strategies in a promotion intensive environment. Marketing Science
, 10(2), 91-110.
Dekimpe, M. G., & Hanssens, D. M. (1999). Sustained spending and persistent response: A new look at long-run marketing profitability. Journal of Marketing Research
, 397-412.
Gabor, A., & Granger, C. W. (1966). Price as an Indicator of Quality: Report on an Enquiry. Economica
, 43-70.
Granger, C. W. (1969). Investigating causal relations by econometric models and cross-spectral methods. Econometrica: Journal of the Econometric Society
, 424-438.
Gredenhoff, M., & Karlsson, S. (1999). Lag-length selection in VAR-models using equal and unequal lag-length procedures.
Computational Statistics
, 14(2), 171-187.
Hanssens, D. M. (1980). Market response, competitive behavior, and time series analysis. Journal of Marketing Research
,
470-485.
Horvath, C., Leeflang, P. S., Wieringa, J. E., & Wittink, D. R. (2005). Competitive reaction-and feedback effects based on VARX models of pooled store data. International Journal of Research in Marketing
, 22(4), 415-426.
Horváth, C., & Wieringa, J. E. (2003). Combining time series and cross sectional data for the analysis of dynamic marketing
systems
. University of Groningen.
Kumar, V., & Leone, R. P. (1988). Measuring the effect of retail store promotions on brand and store substitution. Journal of Marketing Research, 178-185.
Liew, V. K. S. (2004). Which lag length selection criteria should we employ?. Economics bulletin
, 3(33), 1-9.
Lütkepohl, H. (2007). General-to-specific or specific-to-general modelling? An opinion on current econometric terminology. Journal of Econometrics
, 136(1), 319-324.
Mela, C. F., Jedidi, K., & Bowman, D. (1998). The long-run impact of promotions on consumer stockpiling behavior. Journal of Marketing research, 250-262.
Mela, Carl F., Sunil Gupta, and Donald R. Lehmann (1997), “The long-run Impact of Promotion and Advertising on Consumer Brand Choice,” Journal of Marketing Research, 34 (2), 248–61
Nijs, V. R., Dekimpe, M. G., Steenkamps, J. B. E., & Hanssens, D. M. (2001). The category-demand effects of price promotions. Marketing science
, 20(1), 1-22.
Pauwels, K., Hanssens, D. M., & Siddarth, S. (2002). The long-run effects of price promotions on category incidence, brand choice, and purchase quantity. Journal of marketing research
, 39(4), 421-439.
Pauwels, K. (2007). How retailer and competitor decisions drive the long-run effectiveness of manufacturer promotions for fast moving consumer goods.Journal of Retailing
, 83(3), 297-308.
Slotegraaf, R. J., & Pauwels, K. (2008). The impact of brand equity and innovation on the long-run effectiveness of promotions. Journal of Marketing Research
, 45(3), 293-306.
Steenkamp, J. B. E., Nijs, V. R., Hanssens, D. M., & Dekimpe, M. G. (2005). Competitive reactions to advertising and promotion attacks. Marketing science
,24(1), 35-54.
Stewart, D. W. (2009). Marketing accountability: linking marketing actions to financial results. Journal of Business
Research
, 62(6), 636-643.
Van Heerde, H. J., Leeflang, P. S., & Wittink, D. R. (2000). The estimation of pre- and post-promotion dips with store-level scanner data. Journal of Marketing Research
, 37(3), 383-395.
Van Heerde, H. J., Leeflang, P. S., & Wittink, D. R. (2004). Decomposing the sales promotion bump with store data. Marketing Science
, 23(3), 317-334.
Van Heerde, H. J., Gupta, S., & Wittink, D. R. (2003). Is 75% of the sales promotion bump due to brand switching? No, only 33% is. Journal of Marketing Research, 40(4), 481-491.
Verhoef, Peter C., and Peter SH Leeflang. "Understanding the marketing department's influence within the firm." Journal of
marketing
73.2 (2009): 14-37.
Walters, R. G. (1991). Assessing the impact of retail price promotions on product substitution, complementary purchase, and interstore sales displacement. The Journal of Marketing
, 17-28.
Wieringa, J. E., & Horvath, C. (2005). Computing level-impulse responses of log-specified VAR systems. International
Journal of Forecasting
, 21(2), 279-289.
Zeelenberg, M., & Van Putten, M. (2005). The dark side of discounts: An inaction inertia perspective on the post-promotion dip. Psychology & Marketing
