The influence of official sports events sponsorship on own and competitor marketing elasticities and sales in the FMCG sector.

The influence of official sports events sponsorship

on own and competitor marketing elasticities and

sales in the FMCG sector.

Master Thesis | Amelie Schuler

Marketing Intelligence and Marketing Management

The influence of official sports events sponsorship

on own and competitor marketing elasticities and

sales in the FMCG sector

.

By

Amelie Schuler

S3306119

amelieschuler@gmail.com

MSc Marketing Intelligence & Marketing Management (Candidate)

Final Thesis, June 2018

At

University of Groningen

Faculty of Economics and Business

Department of Marketing

PO Box 800, 9700 AV Groningen (NL)

Nassaulaan 46A

9717 CL Groningen

+31 (0)6 46074839

Abstract

Major sports events attract millions of viewers every year and receive high media attention. Firms try to benefit from that by increasing their marketing activities around such events, or even engage in expensive official sponsorships. By doing so, they are trying to escape clutter due to increased marketing actions around events. But is sponsorship really able to lift the decreasing effect of events on marketing elasticities during event times? Also, companies want to avoid competitors profiting from “free-riding”, so another question is: In what way do these event-related marketing activities influence competitor sales? To give answers on these questions, this work investigates on the effects of advertising and pricing under the influence of major sports events and sponsorship on sales. Using four years of scanner data from three leading supermarkets in the Netherlands, short- and long run effects on own and competitor sales are estimated. The results show that sponsorship is indeed increasing own sales in the short run, and, different to advertising, decreases competitor sales. However, only in combination with price changes, sponsorship is able to increase marketing

Preface

Dear reader,

The following 58 pages will guide you through the wonderful world of modelling marketing effectiveness… Okay, to be honest, at this very moment, I have a hard time remembering, why I choose this Master programme in Marketing Intelligence and added the Management track to it. But this is due to a severe lack of sleep and the slowly releasing pressure of finalising this paper.

However, looking back over the past one and a half years, I do not regret choosing this Master for a second. This programme and the University of Groningen were the best end for my higher education, of that I am sure. I am not saying this, because I feel obliged to do so, but because I truly feel it. Of course, writing this thesis has cost me a lot of nerves and had me constantly facing new challenges. But five months later I am happy and proud of the results, and have probably learned as much about data analysis and marketing effects as in the two semesters before combined (and that was not too little). Even more I probably learned about myself, and I now fully understand what my supervisor, Maarten Gijsenberg, meant in the beginning of this when he said “It is also about the process.”. Before you can continue diving into the world of sports events, sponsorship, and marketing

effectiveness, I want to take the opportunity and thank several persons that have been a great help while writing the paper. First of all, I want to thank my parents and my brother Benedikt for

supporting me over all this years and being my home wherever I lived. Also, thank you, Dion, for everything you did for me in the last months and for keeping up with countless “ik ben zo moe”. Also, thanks to my thesis and real-life buddy David, for sharing insights, discussing R issues, and coffee breaks in the UB.

Last, but definitely not least, I want to thank my supervisor Maarten for the guidance through this process, his valuable insights, the nice talks, and for encouraging me in challenging times to not give up on my initial research concept. Thank you.

I hope you enjoy now reading through this paper and can take some interesting findings with you. If you want to discuss something, my email address can be found on the second page.

Table of Content

2.1. The influence of marketing instruments on brand sales ... 11

2.2. The influence of official sponsorships and major events on brands sales ... 13

2.3. The influence of official sponsorships and major events on marketing ... 15

2.4. Conceptual Model ... 18

3. Methodology ... 19

3.1. Error correction model ... 19

3.2. Data Description ... 20

3.3. Assessing stationarity and normality ... 22

3.4. Transformation of variables ... 24

4. Model Specification ... 25

5. Results ... 27

5.1. Results outside of the model ... 27

5.2. Model Diagnostics ... 29

5.3. Effects on own sales – Model 1 ... 32

5.4. Effects on competitor sales – Model 2 ... 35

5.5. Simulation ... 38

6. Discussion ... 39

6.1. Answers to the posed research questions... 40

6.2. Summary ... 43

6.3. Managerial Implications ... 44

7. Limitations and Implications for Future Research ... 46

References ... 48

List of Abbreviations

B2B - Business-to-Business (commercial business relations) B2C - Business-to-Consumer (business to private customers) CB - Competitor Brand

ECM - Error Correction Model

e.g. - Abbreviation for Latin term exemplī grātiā (“for the sake of an example”) FB - Focal Brand

FMCG - Fast Moving Consumer Goods

i.e. - Abbreviation of Latin term id est (“that is”)

IPS - Im, Pesaran and Shin test (Panel-data unit-root test)

IRI - Information Resources, Incorporate (American market research company) KLO - “Kijk- en Luisteronderzoek” (Dutch Television and Radio Audience Rating) LLC - Levin, Lin and Chu test (Panel-data unit-root test)

LR - Long run

OS - Official Sponsorship

PP - Phillips-Perron (Unit Root Test)

SKO - “Stichting Kijkonderzoek” (Dutch Audience Rating Foundation) SR - Short run

List of Figures

Figure 1.The sponsorship effects process. (Meenaghan, 2001, p. 115) ... 14

Figure 2. Conceptual model (Model 1). Figure 3. Conceptual model (Model 2). ... 18

Figure 4. Means of sums of price, advertising, and sales under the different conditions. ... 27

Figure 5 and 6. Sales of Mars Brands in Week 1 to 60 under the influence of two official sponsorships. ... 29

Figure 7. Overview of the total of the different short-run elasticities of advertising. ... 34

Figure 8. Overview of the total of the different short run elasticities of price. ... 34

Figure 9. Overview of the total of the different short-run elasticities of advertising. ... 37

Figure 10. Overview of the total of the different short-run elasticities of price. ... 38

Figure 11. Top-10 most viewed TV programmes in the Netherlands, 1994. (NOS - KLO, 1996) ... 52

Figure 12. Top-10 most viewed TV programmes in the Netherlands, 1995. (NOS - KLO, 1996) ... 52

Figure 13. Top-10 most viewed TV programmes in the Netherlands, 1996. (NOS - KLO, 1997) ... 52

Figure 14. Top-10 most viewed TV programmes in the Netherlands, 1998. (NOS - KLO, 1999) ... 52

List of Tables

Table 2. Summary of the included major sports events with their respective official sponsors. ... 22

Table 3. Phillips-Perron Unit Root Tests for Sales and Price. ... 23

Table 4. Levin-Lin-Chu Unit-Root Test & Im-Pesaran-Shin Unit-Root Test. ... 23

Table 5. Shapiro-Wilk normality tests for sales, price, and advertising. ... 24

Table 6. Correlation matrix including the response variable and most important predictive variables. ... 28

Table 7. Output of an analysis on variance inflation factors. ... 29

Table 8. Model quality and fit. ... 30

Table 9. Summary of model fit over all models performed (n=63). ... 31

Table 10. Results of the ECM model on competitor sales... 32

Table 12. SR and LR Simulation for FB increase in advertising spending. ... 38

Table 13. SR and LR Simulation for FB decrease in price. ... 39

Table 14. Extensive summary of all included major sports events incl. audience rating numbers and sources. ... 53

Table 15. Extensive results Phillips-Perron Unit Root Test over all brands in the data set. ... 54

Table 16. Extensive results Shapiro-Wilk-test for Normality over all brands in the data set... 55

Table 17. Extensive explanation of the model specification for Model 1. ... 57

Table 18. Individual brand-level estimates for AIC, BIC, and adjusted R-Squared (quality criteria). ... 58

Table 19. Individual brand-level results for model significance and explained variance (Model 1). ... 59

Table 20. Individual category-level results for model significance and explained variance (Model 2)... 59

Table 21. Individual brand-level estimates for Model 1. ... 60

1. Introduction

Sixty-five point eight billion dollar. This enormous amount of money is projected to be spent globally on sponsorship in 2018, 70% thereof on sports sponsorship (ESP Properties, LLC, 2018). Today’s global sponsorship spending is twice as high as only thirteen years ago, when the global spending was only at $30.5bn (IEG LLC, 2009). This trend towards investing marketing budget in official sponsorship, has also been discussed by research: Meenaghan (2013) finds sponsorship, after digital and social media, to be the most rapidly growing marketing instrument over several decades. Firms, however, do not only sponsor sports and sports events out of beneficence, they hope the investment will result in a sales increase. As official sponsors, the companies are aiming at transferring the positive emotions of popular events on their brand, and at benefiting through increased audience attention and media coverage. (Cornwell, Weeks, & Roy, 2005) This is possible as world championships, the Olympic Games, or soccer tournament finals regularly attract masses of viewers all over the globe. In 2018, the globally most viewed sports event, the Super Bowl, attracted only in the US over 100 million viewers for the ninth straight year (Adgate, 2018). However, also in the Netherlands millions of people follow sports events from home. Only in 2014, a new all-time-record was established: Over 9 million viewers watched the semi-final match of the soccer world-cup, Netherlands vs. Argentina, which amounts to an audience rating of 58.5% and an massive market share of 88.0%. (SKO, 2015)

But not only firms that are official sponsors use the public attention and increased audience around major sports events to communicate their brand to a broader mass. A common and simple tactic for every company is to simply increase marketing activities, such as TV advertising or price promotions, around events. A popular strategy that shows in aggregated numbers: Keller, W., Deleersnyder, and Gedenk (2018) find that during event times, price promotion activity to be 21.5% higher compared to non-event times. For advertising, Gijsenberg (2014b) finds that 25% more money is spent on

advertising than during non-event times in the UK. This is a considerable number considering that in 2013 the total advertising spending in the UK hit the £20bn mark. Taking a weekly average of £385mn, in an event week a total of £96.25mn would be spent additionally on advertising. This is however not only due to a higher number of marketing actions. Knowing of the popularity for marketers, media channels such as TV increase advertising prices heavily during event times. For a single sports event like the 2018 Super Bowl, for example, advertisers were investing over $5mn for a 30-second commercial (Adgate, 2018).

leads to lowered advertising effectiveness. (Gijsenberg, 2014b; Hammer, Riebe, & Kennedy, 2009; Keller, K. L., 1993) This results in a sales increase with regard to advertising spending that is relatively lower than in non-event times; the so-called advertising elasticities are smaller than outside of event periods. This is where sponsorship comes again into play. In order to avoid the undesired effects of decreased marketing effectiveness around events, firms invest in official sponsorship to leverage their marketing instruments. By doing so, they put themselves in an exclusive position, hoping to stand out from the multitude of firms increasing their marketing activity around major events (Pitt, Parent, Berthon, & Steyn, 2010). However, in contrast to traditional marketing instruments, effects of sports sponsorships on marketing and their influence on sales, are not well researched. (Keller, W. et al., 2018) In particular, their effect on competitor sales and their long-term development has not been analysed in academic literature. Thus, placing managerial decisions on research can be

challenging in this field. For marketers, this is problematic, as the above discussion on the enormous budgets spent on marketing around event-times and sponsorship, highlight how (sports) events-related marketing is a billion-dollar business. Obviously, firms want to make sure they invest their budget most efficiently and base their strategic marketing on reliable empiric insights. The lack of research was also commented on by M. Gijsenberg (Gijsenberg, 2014a) and B.T. Cornwell (Cornwell et al., 2005). Both researchers state that in-depth knowledge of marketing effectiveness and dynamics under the influence of major sports events is rather limited.

„One of the conclusions (…) is the fact that in-depth knowledge about marketing effectiveness and – especially – marketing dynamics over time around these major sports events is still relatively limited.

As such, they can be considered a call for further research….” (Gijsenberg, 2014a, p. 32)

Fortunately, lately, two academic papers have focused on assessing marketing effectiveness around major sports events. Firstly, Gijsenberg (2014b) assesses the influence on advertising own- and cross-elasticities around major sports events and finds them to be weaker during event times. However, this does not take official sponsorship into account. Another, very recent, paper by Keller, W. et al. (2018) adds to this and does integrate official sponsorship in their assessment of price promotion effectiveness around major events. They find them to be slightly less effective during event times and identify official sponsorship and category event-fit as drivers of promotion effectiveness. (Keller, W. et al., 2018) However, in contrast to Gijsenberg (2014b), the work does not analyse effects on advertising elasticities, nor does it consider cross-elasticities on competitor sales.

studies have shown how advertising does not only positively influence the sales of the own brand, but also of the competitor brands due to potentially increased category-demand (Schultz & Wittink, 1976; van Heerde, H. J., Gijsenberg, Dekimpe, & Steenkamp, 2013).

On a practical perspective, there is no academic basis yet for firms to asses if it even pays off to invest massive amounts into official sponsorship for improving their marketing, or if the competitor eventually profits more by “free-riding” and benefiting from category-demand effects as it has been shown for advertising (Schultz & Wittink, 1976; Sethuraman, Tellis, & Briesch, 2011). Hence, creating knowledge here is crucial in order to deliver guidelines and implications for marketers and to shed more light for them on how to invest their marketing budget most efficiently.

This research will focus on deriving such insight by answering the following research questions with a short and long-term perspective.

- How do major sports events influence own and competitor sales? - How does official sponsorship influence own and competitor sales?

- How are own- and cross-elasticities of advertising and price influenced by major sports events? - How are own- and cross-elasticities of advertising and price influenced by official sponsorship? - What is the added effect of official sponsorship to the mere events effect on advertising and price

elasticities?

In order to answer these questions, a large-scale empirical study will be conducted assessing the changes in price and advertising elasticities under the influence of major sports events and official sponsorship. Firstly, prior research and findings will be discussed, and research gaps will be exposed to create a solid academic fundament for this work. The literature review is followed by an

elaboration on the methodology used in the data analysis and a specification of the applied models for own and competitor sales. After reporting the outcomes of both models, they found elasticities will be used for a simulation of total effects on sales. Subsequently, the results will be discussed thoroughly and compared to prior findings. For the effects with a comprehensive academic backup, the findings can add to the generalisation of marketing effects, and for the less researched

phenomena, such as the cross-effects of official sponsorships on sales, the insights can be seen as new additions to the research field. Finally, in the last step of this research, managerial

2. Prior Research

In the field of sports event-related marketing, research has identified three main strategies to improve marketing efficiency around major sports events (Gijsenberg, 2014b).

Firstly, brands can simply take advantage of larger audiences and the positive atmosphere of such events. In this way, they can improve their marketing efficiency by benefiting from high media coverage and thus higher audience exposure (Zajonc, 1968). Additionally, they aim to transfer the positive attitude towards the event on their own brand. (Grohs, Wagner, & Vsetecka, 2004). This, in turn, results in increased attention and enthusiasm of the consumers for the marketed brands and a higher ranking in the relevant choice set, which can lead to greater purchase likelihoods (e.g., Gijsenberg, 2014b, Mazodier & Quester, 2014, Morris, Woo, Geason, & Kim, 2002).

A second option is to engage in the so-called ambush-marketing, which intentionally uses misleading communication for brands to be perceived as an official sponsor(e.g., Mazodier, Quester, & Chandon, 2012, Pitt et al., 2010). Firms do this to stand out from the marketing clutter during event times, similarly to actual sponsors. At the 1996 Olympics Nike, for example, heavily invested in ambush-marketing activities, e.g. plastering the city with billboards, or distributing branded merchandise at the competitions, whereas its rival Reebok spent $50mn on official sponsorship. (Pitt et al., 2010). This also leads to the third strategy for firms to profit from event times: establishing an official sponsorship relation with the event (e.g., Mazodier & Quester, 2014). Such a commitment has been shown to be an efficient and well-working marketing tool (Cho, Lee, Yoon, & Rhodes, 2011). But, official sponsorship is cost-extensive and limited to a few brands per event to maintain a certain exclusiveness, and even though research on the general effect of sponsorship is plenty, academic work on possible cross-effects on competing brands’ sales is limited. However, this is particularly of interest as firms want to ensure their marketing pays off the most for their own brand.

In the following literature review, the first strategy, increasing advertising and price promotion activities, and the last one, official sponsorship, will be discussed with regard to academic findings for their effects on sales and marketing instruments. Ambush marketing is not discussed per se, as it is not a focus of this work. Finally, expectations are made regarding short and long-run effects on own and competitor sales of marketing, events, and sponsorship.

2.1. The influence of marketing instruments on brand sales

Influence on own sales

the term elasticity refers to the proportionate reaction in sales with regard to a percentage change in advertisement spending or price (Hanssens, 2009).

For advertising elasticities within the fast-moving consumer goods (FMCG) industry, meta-studies, summarising various research findings, and single studies state them to be between 0.00-0.20 (Hu, Lodish, & Krieger, 2007; Leone & Schultz, 1980; Sethuraman et al., 2011; Tellis, 2004).

In the long run (LR), research has proven advertising elasticities to become stronger instead of fading out. Van Heerde, H. J. et al. (2013) find a SR advertising elasticity of 0.002 compared to a LR elasticity of 0.013, in line with other studies (Ataman, van Heerde, & Mela, 2010; Sethuraman & Tellis, 1991). For price elasticities of consumer packaged goods, a meta-analysis by Bijmolt, T. H.A., van Heerde, and Pieters (2005) finds them to be at an average of -2.6 on brand-level. Following this, a 10% price cut would lead to a 2600% sales increase on average. The reasons for sales increase are mainly brand switching and temporary category growth. (Bell, Chiang, & Padmanabhan, 1999) Specifically for time series data, Sethuraman and Tellis (1991) find an average price elasticity of -1.63. In the long run, price elasticities often diminish and thus show a different pattern compared to the findings for advertising elasticities. van Heerde, H. J. et al. (2013) for example, find an average SR price elasticity

of -1.427 compared to a weaker LR elasticity of -0.838, which is in line with Ataman et al. (2010) and

Dekimpe, Hanssens, and Silva-Risso (1998). Other studies, however, find the price elasticity to be similar or even slightly higher in the long run (Gijsenberg, 2014b). A reason for the different findings might lie in the brand and product selection in the datasets, because consumers are more prone to price promotions for durables and longer-lasting FMCG products and invest or stock-pile

(Sethuraman, Srinivasan, & Kim, 1999). Also by other studies, dissimilarities in focus categories and products are stated as the reason differences in the LR price effect development (Dekimpe et al., 1998; van Heerde, H., Helsen, & Dekimpe, 2007).

Concluding, for own advertising elasticities a low positive value is expected for both, short and long run, with a slightly LR higher elasticity. For own price elasticities, clearly a negative direction is expected for both short and long-term. However, it is unclear based on academic literature if price elasticities will remain or diminish in the LR. Thus, the results can add more insight to research here.

Influence on competitor sales

H. J. et al. (2013) find positive advertising cross-elasticities, Clarke (1973) and Ailawadi et al. (2001) report a negative influence. Danaher, Bonfrer, and Dhar (2008), on the other hand, find a mixed pattern. The researchers, however, also find an explanation for the different directions of advertising cross-elasticities. They seem to be negative in situations when a large brand’s advertising affects a smaller focal brand and vice versa (Danaher et al., 2008). The papers taking the long run into account, find stronger positive advertising cross-elasticities compared to the SR, similar to the findings for own-advertising elasticities in the LR (Gijsenberg, 2014b; van Heerde, H. J. et al., 2013). With regard to the cross-elasticities of price promotions, the work of Sethuraman et al. (1999) establishes some empirical generalisations. They find asymmetric cross-price effects depending on how the prices of the brands relate to each other. If an A-brand cuts its price, elasticities on the store brand sales are 0.48, if a store brand’s price is cut, the cross-price elasticity is only 0.34. (Sethuraman et al., 1999). Also, van Heerde, H. J. et al. (2013) find a positive short-run cross-price elasticity of 0.48. Ailawadi et al. (2001) find a higher cross-elasticity of 1.7, however, this analysis focusses specifically on P&G brands, whereas the other studies take a wide variety of brands and categories into account. Thus, most likely, P&G brand sales react particularly sensitive to price changes.

In the long run, van Heerde, H. J. et al. (2013) find cross-price elasticities at almost the same level as for short run, 0.413. Gijsenberg (2014b) finds LR cross-elasticities of price to be even higher as for SR (0.327 vs. 0.151). It might be that, similar to the findings of Bell et al. (1999) for own LR price

elasticities, the LR development of cross-price elasticities is dependent on product and category. Concluding, based on previous research, the following directions for advertising and pricing cross-elasticities can be expected. For advertising cross-cross-elasticities, no clear direction can be expected based on academia as prior studies have found different results depending on category and brand position. Even though the focus brand (sponsoring brands) are often the category leaders, there are also cases were the competitor brands are. As the results will be presented aggregately, literature does not allow to give an estimation on the overall direction. However, for the LR, analogue to own elasticities cross-elasticities are expected to be stronger compared to the SR. For price promotion cross-elasticities in both, SR and LR, a positive direction is expected. The strength is expected to differ across categories depending on how closely priced to brands within the category are. The LR

elasticity is expected to be similarly strong or slightly increased compared to SR.

2.2. The influence of official sponsorships and major events on brands sales

Whereas a multitude of studies examines the influence of major sports events on the economy of the host city or country (Baade & Matheson, 2004; Preuß, 2006; Rose & Spiegel, 2011), research

events on brand sales. Keller, W. et al. (2018), on one side, find in their research on price promotions and popular events significant negative effects of -0.059 for ice skating events and for the winter Olympics. Gijsenberg (2014b), on the other side, finds a low positive effect of events on sales (0.010), in particular before the event (0.012). Thus, the direction of the effect remains unclear, and the results of this work can help to add more insights to the direct effect of major events on FMCG sales. Luckily, academic work on the direct influence of official sponsorship of such major sports events on brand sales is less scarce. Under the influence of the 2008 Olympics, Cho et al. (2011) analysed the

influence of sponsorship on actual household-level sales of soft drinks in the US.Their main results

confirm the existence of a sponsorship effect on consumer purchasing behaviour isolated from advertising effects, i.e. consumers buying a higher number of brands that are official sponsors, compared to brands that are not officially sponsoring. Further, they find this effect most likely to be limited to short-term, i.e., to the duration of the respective event itself, as they cannot find a significant LR effect for Olympic sponsorship (Cho et al., 2011).

Other academic research studying the influence of sponsorship on response variables other than sales can also shed more light on sponsorship effects. In order to transfer their findings to insights on sales influence, the hierarchic model of the sponsorship effects process by Meenaghan (2001) (Fig. 1) will be introduced. The model displays how sponsorship subsequently affects different response indicators, e.g. customer perception, before resulting in actual purchase behaviour.

Figure 1.The sponsorship effects process. (Meenaghan, 2001, p. 115)

Meenaghan’s model is also in line with the brand value chain (Keller, K. Lane & Lehmann, 2006), stating that an improved brand image and reputation will lead to a financial effect in the long run. This argumentation can be applied to the results of Olson and Thjømøe (2009), who find that official sponsorship has a similar positive effect on customer purchase intention (+29.93%) as classic

Additionally, several researchers analysed changes in firm market value based on stock price following a sponsorship announcement. (Mishra, Bobinski Jr., & Bhabra, 1997; Miyazaki & Morgan, 2001) As brokers have high expert knowledge about how business decisions can influence sales development, an increase in stock prices can be seen as an antecedent of rising sales, and thus can be used as an indicator for sponsorship-effect on sales. On average, the researchers find corporate sponsorship announcements to impact stock market prices positively, and that sponsorship is regarded as worthwhile investments by the financial community. Thus, official sponsorship

announcements lead market analysts to expect an increase in a company’s profitability, which then, in turn, affects the prices for stocks of that firm. (Mishra et al., 1997; Miyazaki & Morgan, 2001) This can be translated to an expected positive effect of on sales.

Concluding, for the SR and LR effects of major sports events on brand sales, no clear direction can be assumed as research has shown different directions. (Gijsenberg, 2014b; Keller, W. et al., 2018) For the effect of sponsorship on sales, a positive direction is expected. For the long run effect, a direction cannot be estimated based on literature.

With regard to cross-effects of official sponsorship, i.e. the influence on competitor sales, there was no literature to be found. However, Olson and Thjømøe (2009), have shown that the own effect of sponsorship in direction and size is almost identical to traditional TV advertising. Also, Meenaghan’s effect process of sponsorship model strongly reminds of the hierarchic effects of advertising, as in Lavidge's Hierarchy of Effects model or the DAGMAR model (Dutka & Colley, 1995; Lavidge & Steiner, 1961). Thus, based on the high similarities of sponsorship effects to advertising effects, the cross-effect of sponsorship might be similar to cross-advertising cross-effects as discussed in 2.1. However, there is no sufficient academic background to clearly expect a direction. The results from this study will add insights regarding the direction of cross-elasticities of sponsorship.

2.3. The influence of official sponsorships and major events on marketing

Influence on own sales

besides the direct effect of sponsorship on sales discussed in the preceding chapter, also an indirect effect shown in an increase in the effectiveness of a sponsor’s advertising during the sponsored event. With regard to the effects of sponsorship on price elasticities, Keller, W. et al. (2018) find significant negative interaction effects of sponsorship and price for the majority of the respected events. Thus, sponsorship is able to increase the effectiveness of price promotions. They also identify official sponsorship – besides category-event fit and demand expansion – as a significant driver of relative promotion effectiveness during event times. (Keller, W. et al., 2018)

Besides the effects of official sponsorship, also the happening major sports events alone can influence marketing instruments such as price and advertising. Yet, there are different potential effects: Whereas increased audience attention and media coverage can be beneficial for the marketing activities (Meenaghan, 2001; Zajonc, 1968), clutter effects can decrease the marketing’s effectiveness by up to 50%. (Danaher et al., 2008) Recently, two papers have focused on the changes in advertising and price elasticities under the influence of cultural, seasonal, or sports events.

Firstly, Gijsenberg (2014b) investigated on the changes in own and cross advertising elasticities around major sports events. He finds that own-advertising effectiveness is not improved before or during an event. (Gijsenberg, 2014b) However, he states that potential beneficial memory effects, as discussed in Keller, K. L. (1993), can offset this decreased advertising effectiveness. Keller, W. et al. (2018) add to this research by analysing the effect of events on price promotion effectiveness. They find different effects compared to advertising elasticities during event times: Elasticities of price promotions are slightly higher during event times compared to non-event times. (Keller, W. et al., 2018) This is in line with Gijsenberg (2014b) who also gives a small outlook on price promotions during event times in his study, and finds them to be twice as effective during event times. However, it is important to mention here, that Keller, W. et al. (2018) find high variance across brands and events and actual price elasticity to be depending on the right brand-event combination.

given based on research as none of the papers includes the LR effect of event on price elasticities. However, it is likely that the effect diminishes analogue to the main effect of price.

Lastly, besides the influence of events or sponsorship on marketing efficiency, the effect of both, events and sponsorship, on sales and price elasticities is of interest. Thus, the additional effect on marketing by official sponsorship during event times. With regard to advertising effectiveness, only the paper of Cho et al. (2011) takes a look at this three-way-interaction. The researchers find the sponsorship effect on advertising to be stronger during the Olympics compared to outside the event. Also, Keller, W. et al. (2018) analyse the effect of both, events and sponsorship, however, with regard to price promotion effectiveness and find a significant interaction of all three factors. The negative interaction term indicates price cuts are even more effective, when the brand is an official sponsor during the event times, compared to the single effects of sponsorship and events on price.

Following, for the effect of both, sponsorship and events, on advertising elasticities, a positive

direction is expected following Cho et al. (2011). For price elasticities, a negative elasticity is expected in line with Keller, W. et al. (2018). With regard to LR effects, there is no academic literature to be found. An estimation based on related effect as done for other effects seems to vague here as a complex three-way interaction is discussed. Thus, it remains to wait for the model results to shed more light on this matter.

Influence on competitor sales

estimations of expected effects. Thus, the effect of sponsorship, and event and sponsorship, on marketing cross-elasticities will be treated as an open issue that will be investigated in this paper. Concluding, with regard to SR and LR cross-elasticities, negative interaction effects for events with advertising cross-elasticities are expected, indicating weakened advertising effectiveness during event times following Gijsenberg (2014b). For price cross-elasticities under the influence of events, no estimations can be given as this field is a research gap in academia. The same holds for the influence of both, sponsorship and events, on marketing cross-elasticities.

Summarizing, the literature review has shown up that research on the direct own and cross

elasticities of advertising and price is plenty, even with regard to SR and LR effects. However, when the moderating effect of event and sponsorships are considered, multiple research gaps are

identified. Academic literature on cross-elasticities and LR elasticities is lacking in particular, as most present papers focus on own elasticities and current effects. This is very unfortunate considering the high resources spent on official sponsorship as mentioned in the introduction. It would be utterly beneficial for marketers to have insight on if the added effect for the own firm beyond traditional advertising or price cuts, and how long this effect lasts in order to assess if such an investment pay off. Also, with regard to competitor brands, it is of high interest for companies to know if their own effect is greater than the cross-effects through “free-riding”, i.e. a neighbour brand benefiting through increased category-demand and spill-over effects. This is why this work aims at adressing named research gaps and pays special attention to cross-elasticities.

As a result, this paper might be able to show if it does indeed pay off for firms to invest in marketing around events and if the high costs of official sponsorship enable to leverage price and advertising effects and stand out from the competition.

2.4. Conceptual Model

In order to address the above found research gaps, two conceptual models as displayed in Figure 2 and 3 were designed to serve as a base for the statistical analysis on the effects of major events and sponsorship on marketing variables and sales. The models portray an abstract representation of actual market influences and relations. (Leeflang, P. S.H., Wieringa, Bijmolt, & Pauwels, 2015)

Model 1 Model 2

On the lowest dimension, this research approach aims to investigate the direct effects of advertising, price, major sports event, and official sponsorship on sales as can be seen from the conceptual models. On the next dimension, the conceptual models take the interactions of each major sports events and, respectively, official sponsorship with advertising and pricing into account. Lastly, the models include the joint effect in the form of two three-way interactions of official sponsorship, major sports events, and advertising and price, respectively. All effects will be estimated for both short and long-term to enable an assessment of the durability of effects. This extensive analysis is required in order to answer the research questions posed in the introduction, and to draw a comprehensive picture of sponsorship and event effects on marketing instruments and sales.

3. Methodology

Looking at the research tasks of this study as discussed 2.4, several requirements are imposed on a suitable statistical model. Firstly, this works aims at gaining insight on how sales are influenced by variables of interest such as advertising, price, events, or sponsorship while accounting for the influence of different control variables such as holidays and yearly quarters. Following, the model needs to be able to estimate direct effects of several predictor variables on the continuous response variable sales. Secondly, as this study focusses on assessing a possible effect of events and

sponsorship on marketing instruments, the model should also be able to estimate the added effects of events and sponsorship in the form of two-way interactions, and for the joint effects in the form of three-way interactions. Fourthly, this study seeks to understand all these effects and interactions for various supermarket categories, and at the same time aims at generalising the findings for FMCG product sales in general. Thus, the chosen model should allow for individual estimates on brand-level and overall insights. Lastly, this study aims to assess the durability of effects, as well as the influence on own and competitor sales. Hence, the chosen model should differentiate between own and cross-elasticities, as well as short- and long-term effects.

3.1. Error correction model

Srinivasan, & Franses, 2007) For marketing models, it is very beneficial to consider dynamic effects according to Leeflang, P. S. H. et al. (2017), who find prior performance and past own and competitor marketing actions to profoundly influence present sales.

After careful selection among different time-series modelling approaches, the established error correction model (ECM) (e.g. used by Leeflang, P. S. H., Wieringa, Bijmolt, and Pauwels (2017) and Gijsenberg (2014b) was chosen. Other models such as the VARX model (vector-autoregressive models with exogenous variables) (Nijs, Dekimpe, Steenkamps, & Hanssens, 2001), could not be used for this analysis as including two- and three-way interactions is not established with this model. ECM, o the other hand, does allow for including interactions and estimations on brand-level at an

acceptable computing time, and preliminary inferences about short- and long-term effects of the different included predictor variables can be made. Hereby, the long-term effects can be defined as discrepancy between equilibrium sales after marketing actions, potentially in combination with events and sponsorship, and the equilibrium sales outside of such actions (Horvath & Franses, 2004). Following, a lasting effect of marketing variables exists if the long-term effect does not equal zero. (Horvath & Franses, 2004) In the ECM, continuous variables will be log-transformed so the outcomes can be interpreted as elasticities. (Gijsenberg, 2014b; Hanssens, 2009; Leeflang, P. S.H. et al., 2015)

3.2. Data Description

The ECM will be estimated on an extensive dataset consisting of weekly supermarket scanner data from the Netherlands over a 4-year time period. It was collected by market research company Information Resources, Incorporated (IRI) and covers 208 weeks from July 1994 to July 1998. The final data set used in the analysis comprises of 180 products, in the following referred to as brands, spread over 40 distinct supermarket categories, such as “Deodorant” or “Brew Pils”. Of all included brands, 35% (n=63) are so-called focus brands, thus brands that have been official sponsor of a major sports events within the time period captured by the retail data. The remaining 117 brands are the top-3 within-category competitors with regard to market share of the respective focal brands. As it occurs that several focus brands can be within category, there are less competitor brands included than three times the amount of focus brands.

Table 1. Overview of the focal brands included in the final data set.

However, in order to be able to analyse the relationship of advertising to sales and the influence of events and sponsorship on it, the retail scanner data required complementation with advertising numbers, as well as event and sponsorship information.

Firstly, advertising statistics from the BBC research agency in the Netherlands was added. The BBC data gives information on advertising spending and timing on a product-level for a broad range of media: newspapers, magazines, television, radio, cinema, and outdoor-advertising. As the considered time span in this analysis is before the merge of the internet, there is no need nor possibility to include online data in the analysis.

Secondly, to define major sports events, audience rating numbers by the Dutch audience rating foundations “Kijk- and Luisteronderzoek” (KLO) and “Stichting Kijkonderzoek” (SKO) were used. The respective yearly overviews can be found in Appendix A. As some ratings were missing due to the age of the data and to complement the present data, further audience rating numbers were collected from the journalistic research database LexisNexis. The platform was used to scan newspaper articles on audience rating information for broadcasted sports events. The total amount of events was limited to the 30 most viewed events in captured time span. To ensure variations in the data set, a quota on soccer events was set so that the total amount of soccer events was a maximum of 50% (most viewed matches). This was necessary as soccer is by far the most-viewed sports in the Netherlands and a selection merely to audience rating would have led to a data set almost

completely consisting of soccer events. The included events are spread over the three popular Dutch sports, Ice Skating, Soccer and Field Hockey, as well as the Olympic Winter and Summer games. Lastly, official sponsorship was assessed by identifying the official sponsors of the 30 selected sports events. For the global tournaments over several weeks, such as the Olympics or the FIFA world and euro cups, the official sponsors could be easily identified from online records. For smaller, yearly sports events such as the ice skating championships or champions league matches, the respective

NO. CATEGORIES EXAMPLE CATEGORY NO. BRANDS EXAMPLE BRAND

Amstel 3 Brew Pils 4 Amstel Pils

Bolletje 6 Chocolate Cookies 6 Bolletje Chonellys

Coca-Cola 1 Cola 2 Coca-Cola Light

Heineken 2 Brew Old Brown 2 Heineken Oudbruin

Magnum 2 Ice Cream Single Packs 5 Magnum White

Mars 3 Chocolate Toffees 3 Mars Bar

Philips 2 Batteries 3 Philips Super Plus

Remia 6 Sate Sauce 6 Remia Satésaus

Sanex 6 Deodorant 7 Dermo Protector

Snickers 6 Chocolate Bars 7 Snickers Fun Size Mini

Unox 13 Canned Sausages 17 Unox Big Franks

official sponsors were derived through analysis of video material on the broadcasting platform

YouTube. The following table gives an overview of the selected events and according sponsors, a

more extensive version including audience ratings and sources can be found in Appendix B. Table 2. Summary of the included major sports events with their respective official sponsors.

YEAR DATE EVENT OFFICIAL SPONSORS

1994

17. June - 17. July Fifa WC (USA) Coca-Cola; Philips; Snickers 02. November Ajax-Salzburg (Champions League) Amstel Bier; Philips 23. November AC Milan-Ajax (Champions League) Amstel Bier; Philips

23. Nov. - 4. Dec. WK hockey men, Sydney Philips; Coca-Cola; Magnum

1995

06. - 08. January Euro Cup Speed Skating Herenveen Mars; Bolletje; Remia 11. - 12. February World Cup Speed Skating Men, Baselga di Pinè Mars; Bolletje; Remia 05. April Bayern München-Ajax (Champ. League) Amstel Bier; Philips 19. April Ajax-Bayern München (Champ. League) Amstel Bier; Philips 24. April Ajax-AC Milan (Champions League) Amstel Bier; Philips 27. October EK hockey men, Dublin Lion; Coca-Cola 22. November ReaI Madrid-Ajax (Champions League) Amstel Bier; Philips 06. December Ajax-Ferencvaros (Champions League) Amstel Bier; Philips

1996

19. - 21. January Euro Cup Speed Skating Herenveen Sanex; Remia; Bolletje 2. - 4. February World Cup Speed Skating men + women, Inzell Remia; Sanex

06. March Dortmund-Ajax (Champions League) Amstel Bier; Philips 20. March Ajax-B.Dortmund (Champions League) Amstel Bier; Philips 22. May Ajax-Juventus (finale Champions League) Amstel Bier; Philips 8. - 30. June Fifa EC (England) Coca-Cola; Snickers 19. Jul. - 4. Aug. Summer Olympics (USA) Coca-Cola

1997

4. January Elfstedentocht Unox

10. - 12. January Euro Cup Speed Skating Herenveen Sanex; Remia; Bolletje 14. - 16. February World Cup Speed Skating men + women (Nagano) Coca-Cola; Sanex 19. March Atletico Madrid - Ajax Amstel Bier; Philips 17. September PSV-Dynamo Kiev Amstel Bier; Philips 25. November Ajax - Bochum Amstel Bier; Philips

1998

7. - 22. February Winter Olympics (Japan) Coca-Cola 13. - 15. March World Cup Speed Skating men + women Herenveen Sanex 20. - 31. May WK hockey women, Utrecht Heineken 20. June – 1. July WK hockey men, Utrecht Heineken

10. June - 12. July Fifa WC (France) Coca-Cola; Philips; Snickers

With regard to timing conditions, it was decided to consider both, the short-run as well as the long run effect of the marketing variables and the moderations thereof, and to define the week of the event taking place as an event-condition.

3.3. Assessing stationarity and normality

Stationarity can be defined as the tendency of a time series to return to its deterministic components (Leeflang, P. S. H. et al., 2017). In other words, a time series fulfils stationarity when a shift in time does not alter the distribution shape. A reason for non-stationarity are so-called unit roots. They should not occur in time series analysis as they signify a systematic and unpredictable pattern. If unit roots are found in the time series, the data has to be manipulated to make the time series stationary before the ECM can be applied, e.g. by differencing them. (Leeflang, P. S.H. et al., 2015) To test for stationarity, unit root tests are performed on time series data. Firstly, the data was tested on the individual brand-level using Phillips-Perron (PP) Unit Root Test (Phillips & Perron, 1988) on all 180 time series for both the sales and price variable. In the tests, an intercept was included and the trend served as the exogenous variable. The complete table including all individual results can be found in

Appendix C.

Table 3. Phillips-Perron Unit Root Tests for Sales and Price.

SALES PRICE

Number time series sign. 172 151

% of time series sign. 96 % 84 %

alternative hypothesis: stationary

The summarized outcomes of the PP test (Table 3) show that 96% (sales), 84% (price) respectively, of the tested time series do not have a unit root and fulfil the stationarity criterium as can be seen from the significant test results. The non-stationarity of a few time series, however, is not problematic per so as recent studies have shown individual series unit root test to lack power compared to based test. (Gijsenberg, 2014b) Thus, the dataset was also tested for stationarity using two panel-based unit root tests, namely the Levin-Lin-Chu (LLC) Root Test & Im-Pesaran-Shin (IPS) Unit-Root Test. (Im, Pesaran, & Shin, 2003; Levin, Lin, & James Chu, 2002)

Table 4. Levin-Lin-Chu Unit-Root Test & Im-Pesaran-Shin Unit-Root Test.

alternative hypothesis: stationary alternative hypothesis: stationary

The outcomes of the LLC and IPS tests (Table 4) show p-values of smaller than 0.01 for all tested variables sales, price, and advertising. Hence, the null hypothesis of the existence of a unit root can be rejected and the data can be considered significantly (trend) stationary. This allows interpreting the long-term parameters of the model as the cumulative effects of temporary changes in the marketing instruments, besides the interpretation as permanent effects of permanent changes (Gijsenberg, 2014b).

LLC Test Z-VALUE P-VALUE IPS Test Z-VALUE P-VALUE

Sales -5.7277 < 0.01 *** -9.3805 < 0.01 ***

Price -4.096 < 0.01 *** -8.4016 < 0.01 ***

Table 5. Shapiro-Wilk normality tests for sales, price, and advertising.

SALES PRICE ADVERT.

Number time series sign. 179 179 179

% of time series sign. 99.4 % 99.4 % 99.4 %

alternative hypothesis: non-normal distribution

Furthermore, the individual time series were tested for normality regarding the distribution of prices, advertising, and sales using the established Shapiro-Wilk-test. Assessing normal distributions before performing statistical models is a crucial assumption to be fulfilled for drawing conclusion on the significance based on the p-value. (Leeflang, P. S.H. et al., 2015) Overall, 99.4% of all time series on brand-level show a normal distribution with regard to the three continuous variables (Table 5). Thus, the assumption of normality is fulfilled and conclusions of significant estimates can be on the p-values estimated in the following statistical models. A table containing the extensive results for all brand-level Shapiro-Wilk tests can be found in Appendix D.

3.4. Transformation of variables

After assuring the data set was fulfilling the stationarity and normality criteria, it was further prepared for analysis by transforming given variables into required forms.

Natural logarithms

Firstly, continuous variables, such as sales or advertising, were transformed to natural logarithms. By using natural logarithms, regression outcomes can be interpreted as elasticities, which have the advantage of being time- and dimensionless. Thus, they can be generalised for comparisons between points in time and different measurements. (Leeflang, P. S.H. et al., 2015) As the dataset used in this work is covering a four-year period and uses different units for e.g. sales, which can be liters or kilogram, elasticities allow to summarise the results over all brands, and to compare them across the brands. Using natural logarithms in marketing models is an established method used in a broad range of papers in the field. (Bijmolt, T. H.A. et al., 2005; Leone & Schultz, 1980; Sethuraman et al., 1999)

Competitor marketing variables

Secondly, in order to control for marketing cross influences, competitor advertising and price

possible (Little, 1970). For advertising, the sum was chosen as it represents the total of marketing that the consumer is exposed to in the corresponding week.

For Model 2, which investigates influences on competitor sales, competitor marketing variables were substituted by focal brand marketing variables. The own marketing variables from Model 1 were accordingly substituted by the respective competitor brand’s price and variable. All three within-category competitor brands will be respected in the model by being stacked on each other. The regression will accordingly give outcomes on category-level while two dummy variables will control for the position within the category of the competitor brand, i.e. if it is the top-1 or top-2 competitor.

4. Model Specification

In the following both statistical models, for own and competitor sales, are specified as error correction model. Generally, the dependent variables will be own sales or sales of the focal brand (FB), respectively sales of the competitor brand (CB), as the aim is assessing how official sponsorship influences advertising and price promotion elasticities and corresponding cross-elasticities. As predictor variables, advertising spending and price promotions are the main focus; however, also other independent variables will be included in the model to increase its predictability, such as seasonal effects and national holidays. (Leeflang, P. S.H. et al., 2015)

Model 1. DV: Sales of Focal Brands

A detailed explanation of the single elements of the model specification can be found in Appendix E.

where

∆ First difference operator

𝑆𝑎𝑙𝑒𝑠𝐹,𝑡 Volume sales of focal brand F in week t.

𝐸𝑖𝑡 Vector containing dummy variable for the two conditions of event times i (no event (i=0), event (i=1)) in week t.

𝐻𝑗𝑡 Vector containing dummy variable for the two conditions of holidays j (no holiday (j=0), holiday (j=1) in week t.

𝑄_𝑘𝑡 Vectors containing dummy variables for the two conditions (yes/no) of three yearly quarters k (Q2, Q2,3 Q4) in week t. Intercept is Q1.

𝑂𝑆𝐹𝐵,𝑡 Vector containing dummy variable for the two conditions of official sponsorship (OS) of focal brand (no official

sponsor (l=0), official sponsor (l=1)) in week t. 𝐴𝑑𝑣𝐹,𝑡 Deflated advertising by focal brand F in week t.

𝑃𝐹,𝑡 Deflated price of focal brand F in week t.

𝐴𝑑𝑣𝐶,𝑡 Total deflated advertising by top-3 within-category competitor brands C in week t.

𝑃𝐶,𝑡 Average deflated price top-3 within-category competitor brands in week t.

𝛽0𝐹 Intercept per focal brand F.

𝛽_{𝑖 𝐹} The main effects of the different conditions of the dummy variables on sales of the focal brand F. ∝𝐹 SR and LR advertising elasticities of focus brand F w.r.t. sales of F.

𝛾𝐹 SR and LR price elasticities of focus brand F w.r.t. sales of F.

∏𝐹 Adjustment effect for focal brand F.

𝜀𝐹,𝑡 Error term

Model 2. DV: Sales of Competitor Brands

∆𝑙𝑛𝑆𝑎𝑙𝑒𝑠𝐶𝐵𝑚,𝑡= 𝛽0𝐶𝐵+ 𝛽1,𝑖𝐶𝐵𝐸𝑖𝑡+ 𝛽2,𝑗 𝐶𝐵𝐻𝑗𝑡+ ∑ 𝛽3,𝑘𝐶𝐵𝑄𝑘𝑡 3 𝑘=1 + 𝛽_{4,𝑙𝐶𝐵}𝑂𝑆𝐹,𝑡+ ∑ 𝛽_{5,𝑛𝐶𝐵}𝑇𝐵𝐶𝐵𝑚+ 2 𝑛=1 ∝1𝐶𝐵∆ 𝑙𝑛𝐴𝑑𝑣𝐹,𝑡 + 𝛾1𝐶𝐵∆ 𝑙𝑛𝑃𝐹,𝑡+∝2𝐶𝐵∆𝑙𝑛𝐴𝑑𝑣𝐶𝐵𝑚,𝑡+ 𝛾2𝐶𝐵∆ 𝑙𝑛𝑃𝐶𝐵,𝑡+ (∝3𝐶𝐵𝐸𝑖𝑡∗ ∆𝑙𝑛𝐴𝑑𝑣𝐹,𝑡) + (𝛾3𝐶𝐵𝐸𝑖𝑡∗ ∆𝑙𝑛𝑃𝐹,𝑡) + (∝4𝐶𝐵𝑂𝑆𝐹,𝑡∗ ∆𝑙𝑛𝐴𝑑𝑣𝐹,𝑡) + (𝛾4𝐶𝐵𝑂𝑆𝐹,𝑡∗ ∆𝑙𝑛𝑃𝐹,𝑡) + (∝5𝐶𝐵𝑂𝑆𝐹,𝑡∗ 𝐸𝑖𝑡∗ ∆𝑙𝑛𝐴𝑑𝑣𝐹,𝑡) + (𝛾5𝐶𝐵𝑂𝑆𝐹,𝑡∗ 𝐸𝑖𝑡∗ ∆𝑙𝑛𝑃𝐹,𝑡) + ∏𝐶𝐵 [ 𝑙𝑛𝑆𝑎𝑙𝑒𝑠𝐶𝐵,𝑡−1− ( ∝6𝐶𝐵𝑙𝑛𝐴𝑑𝑣𝐹𝐵,𝑡−1+ 𝛾6𝐶𝐵𝑙𝑛𝑃𝐹,𝑡−1+∝7𝐶𝐵𝑙𝑛𝐴𝑑𝑣𝐶𝐵𝑚,𝑡−1+ 𝛾7𝐶𝐵𝑙𝑛𝑃𝐶𝐵𝑚,𝑡−1 + (∝8𝐶𝐵𝐸𝑖𝑡−1∗ 𝑙𝑛𝐴𝑑𝑣𝐹,𝑡−1) + (𝛾8𝐶𝐵𝐸𝑖𝑡−1∗ 𝑙𝑛𝑃𝐹,𝑡−1) + (∝9𝐶𝐵𝑂𝑆𝐹,𝑡−1∗ 𝑙𝑛𝐴𝑑𝑣𝐹,𝑡−1) + (𝛾9𝐶𝐵𝑂𝑆𝐹,𝑡−1∗ 𝑙𝑛𝑃𝐹,𝑡−1) +(∝10𝐶𝐵𝑂𝑆𝐹,𝑡−1∗ 𝐸𝑖𝑡−1∗ 𝑙𝑛𝐴𝑑𝑣𝐹,𝑡−1) + (𝛾10𝐶𝐵𝑂𝑆𝐹,𝑡−1∗ 𝐸𝑖𝑡−1∗ 𝑙𝑛𝑃𝐹,𝑡−1))] + 𝜀𝐹,𝑡 𝐹 = {1, … , 128} , 𝑚 = {1, 2, 3} 𝑡 = {1, … , 208} where

∆ First difference operator

𝑆𝑎𝑙𝑒𝑠𝐶𝐵𝑚,𝑡 Volume sales of respective top-3 within-category competitor brand CBm in week t.

𝐸𝑖𝑡 Vector containing dummy variable for the two conditions of event times i (no event (i=0), event (i=1)) in week t.

𝐻𝑗𝑡 Vector containing dummy variable for the two conditions of holidays j (no holiday (j=0), holiday (j=1) in week t.

𝑄_𝑘𝑡 Vectors containing dummy variables for the two conditions (yes/no) of three yearly quarters k (Q2, Q2,3 Q4) in week t. Intercept is Q1.

𝑂𝑆𝐹𝐵,𝑡 Vector containing dummy variable for the two conditions of official sponsorship (OS) of focal brand (no official

𝑇𝐵𝐶𝐵𝑚 Vectors containing dummy variables for the two conditions (yes/no) of two top brands n (Top-1, Top-2) in week t. Intercept is Top-3.

𝐴𝑑𝑣𝐹,𝑡 Deflated advertising by focal brand F in week t.

𝑃𝐹,𝑡 Deflated price of focal brand F in week t.

𝐴𝑑𝑣𝐶𝐵𝑚,𝑡 Deflated advertising by top-3 within-category competitor brand CBm in week t.

𝑃𝐶𝐵𝑚,𝑡 Deflated price of top-3 within-category competitor brand CBm in week t.

𝛽0𝐶𝑎𝑡 Intercept per category.

𝛽_{𝑖 𝐶𝐵} The main effects of the different conditions of the dummy variables on sales of top-3 within-category competitor brand CBm.

∝𝐶𝐵 SR and LR advertising elasticities of top-3 within-category competitor brand CBm sales of CBm.

𝛾𝐶𝐵 SR and LR price elasticities of top-3 within-category competitor brand CBm sales of CBm.

∏𝐶𝐵 Adjustment effect for focal brand CB.

𝜀𝐹,𝑡 Error term

5. Results

After carefully selecting the dataset, cleaning and preparing it, and specifying an appropriate statistical model, this chapter will focus on the results regarding the present research tasks. Firstly, descriptive findings and correlations will be discussed followed by the model diagnostics.

Subsequently, the ECM outcomes will be reported and compared to prior findings.

5.1. Results outside of the model

5.1.1. Descriptives

In order to get a first insight on how the dependent variables sales as well as the two marketing instruments advertising and price change under the influence of events and sponsorship, a basic descriptive analysis was performed. The outcomes are portrayed in Figure 3 and show the means of

the sums of sales, advertising and price over all categories, under the influence of events and sponsorship to give an overall image on all included FMCG categories.

As visible from the diagram, sales, advertising, and sales do not differ strongly during event times from non-event times - only prices increase slightly during event times. However, if the factor sponsorship is added, the mean sales increase strongly. Thus, sponsoring brands sell more on the average. However, advertising increases as well during sponsorship and price decreases. Thus, on the average sponsoring brands spend more on advertising and lower the prices. A development that stays also for sponsorship during event times, however, prices increase again slightly. Concluding, based on the descriptive analysis a sales increase under sponsorship and events is visible, but it cannot clearly be stated which predictor exactly increases sales. The ECM is expected to explain the causality behind the increased sales.

5.1.2. Correlations

In order to investigate further on the relations found between the variables based on the descriptive analysis, and to assess the quality of the data, a correlation matrix was created using Pearson correlation coefficient. In this way, unusual relations can be detected, or potential multicorrelations can be indicated. (Malhotra, 2004)

Table 6. Correlation matrix including the response variable and most important predictive variables.

Sales EventDummy FocalBrandDummy SponsorshipD.

Sales 0.11*** 0.08***

Price -0.20*** 0.19***

Advertising 0.55*** 0.01*** 0.11*** 0.06***

SponsorshipD. 0.17*** 0.13***

Note: For better visualization, it was decided to only depict results significant on a p=0.05 level.

Overall, the correlation matric shows an ordinary pattern. Firstly, the positive correlations of focal brands with sales, price, and advertising underline that the chosen brands are leading A-brands actively advertising and disposing of a price premium. Furthermore, focal brands show a positive correlation with sponsorship, which is reasonable as focal brands are the brands in the dataset acting as an official sponsor in certain weeks. Secondly, the correlations with events and sponsorship add to the findings for descriptives. The positive correlations with advertising indicate that promotion expenditures are slightly increased during events (0.01), and even stronger during official

sponsorship (0.06). Additionally, sponsorship seems to be able to increase sales as indicated by the significant positive correlation.

Figure 5 and 6. Sales of Mars Brands in Week 1 to 60 under the influence of two official sponsorships. Even though the correlation matrix and the graphical insights indicate that events and sponsorship can increase sales, more sophisticated analysis is required to statistically prove a relationship between these variables. Furthermore, these primary analyses give no insight on the moderation effect of events and sponsorships on marketing efficiency. Also, it is of interest how lasting these possible effects are. To derive these insights, the appropriate time series models as specified in 3.1

were performed and will be reported in the following after discussing the model diagnostics.

5.2. Model Diagnostics

5.2.1. Checking for multicollinearity – VIF scores

As a part of the model diagnostics, variables included in the model were first tested or

multicollinearity. By testing for this basic assumption of linear models, it can be assessed if analysing the interaction effects in the modelling approach is allowed. If variables highly intercorrelate and explain each other, only the separate effects can be assessed. (Leeflang et al. 2015)

Table 7. Output of an analysis on variance inflation factors.

Week Advert. Price Event_D Holiday_D OS_D Q2_D Q3_D Q4_D FocalBrand_D 1.050 1.027 1.049 1.077 1.125 1.051 1.620 1.542 1.534 1.070

The variance inflation factors (VIF) for all tested independent variables are moderate (VIF={1;4}). (Ailawadi et al., 2001; Hair, op. 1995; Malhotra, 2004) We can conclude, that there is no sign of multicollinearity, and interaction effects can be created and analysed in the ECM.

5.2.2. Pooling

Additionally, the used categories differ in seasonality (e.g. ice cream as a very seasonal category vs. cookies as a less seasonal category) and nature of purchase (e.g. batteries as somewhat planned purchase vs. single-packed chocolate bars as more of an impulse purchase), which leads to effects being expected to differ. Within the categories, slope differences are not expected to be especially high, which is why Model 2 was chosen to be pooled for categories, while accounting for the brand position (top-1 brand, top-2 brand, intercept: top-3 brand) of the three included brands with a dummy variable.

5.2.3. Model Fit

Before finally reporting the results of both models, the overall model quality and fit will shortly be assessed. This is a crucial step in predictive analysis to ensure that predictions do not systematically deviate from the actual values (Leeflang, P. S.H. et al., 2015).

Table 8. Model quality and fit.

MODEL A MODEL B MODEL C MODEL D

AIC Min -1068 -1067 -1079 -1078 Avg. -671 -670 -666 -668 Max. 0 -1 -3 -2 BIC Min -9259865 -8337681 -9933057 -9852236 Avg. -602376 -203985 -721082 -555076 Max. 7646292 8667075 6779451 7267175 Adj. R2 Min 0.09 0.09 0.08 0.13 Avg. 0.46 0.46 0.46 0.47 Max. 0.79 0.79 0.78 0.78

Firstly, a hierarchical model building approach was used in order to choose the model with highest the best fit and highest predictive quality. Thereby the quality criteria Akaike Information Criteria (AIC), Bayesian Information Criteria (BIC) were used to measure of estimating precision and parsimony in the parameterization, and adjusted R-Squared to assess explained variance and parsimony. The aim was to build a model with an as high as possible adjusted R-Squared while decreasing AIC and BIC. (Leeflang et al. 2015) An extensive overview over the model criteria for all four models on brand-level can be found in Appendix F.

again, as Table 8 only took Model 1 into account. The extensive results for all performed models, i.e. on brand- and category-level, can be found in Appendix G. Table 9 gives a summary over all

performed models.

Table 9. Summary of model fit over all models performed (n=63).

p-value (F-Test) Adjusted R2 RSE MSE RMSE

Model 1, n=63 0.000 0.466 0.223 0.073 0.269

Model 2, n=63 0.000 0.445 0.254 0.082 0.287

Firstly, all regressions performed on Model 1 and 2 show significant p-values for the F-Tests. This signifies that the equation as a whole explains the changes in the dependent variable, sales, better due to the inclusion of predictor variables, compared to what could be explained based on chance. (Leeflang, P. S.H. et al., 2015)

Secondly, the average explained variance in the dependent variables as shown by the adjusted R-Squared is good with 46.6% (Model 1) and 45.5% (Model 2). Naturally, certain products and categories are more prone to marketing instruments and their sales can be explained better, while the variance in sales of other categories can be explained less. However, the average R-Squared is satisfying compared with a similar time-series model by Gijsenberg (2014a) (0.395), and the median also shows that most models fit well.

Lastly, the predictive quality of the models is assessed by taking the Mean Squared Error (MSE) and Rooted Mean Squared Error (RMSE) into account. MSE uses squared terms which allows weighting large prediction errors more than smaller one. (Leeflang, P. S.H. et al., 2015) The MSEs of Model 1 and 2 with 0.073 and 0.082 are close to zero and thus reflect high predictive power. For an even better judgement comparing the MSEs of two split parts of the data set would give even more insight. However, this was not possible here, as the sponsorships and events are not similarly spread over the whole series. The rooted MSE, RSME, delivers a value in the unit of the dependent variable and should ideally be higher than the standard deviation of the residuals. (Leeflang, P. S.H. et al., 2015) The RSME is in both models higher than the average residual standard error (M1: 0.269 > 0.223; M2: 0.287 > 0.254) which further confirms the quality of both models.

As fit and quality of both models are satisfying, the following results can be said to have a statistically secured basis. The results depicted for Model 1 and 2 were summed up over all models using the added Z-score method (Rosenthal, 1991). By accounting for the heterogeneity between the

t-statistics and p-values of the separate models and by that weights significant results more heavily than insignificant ones. More information on the procedure can be found in Appendix I.

For the following model results, one-sided p-values will be used to assess significance of variables with a clear expected direction, and for variables with no expectations regarding direction due to a lack of prior research, two-sided p-values will be taken into account.

5.3. Effects on own sales – Model 1

To begin with, the results of Model 1 estimating effects on own sales will be reported. In some cases, the significant estimates will also shortly be compared to prior findings to create a base for the following discussion. The individual brand-level estimates of Model 1 can be found in Appendix H. Table 10. Results of the ECM model on competitor sales.

EXPECTED DIRECTION ESTIMATES ADDED Z-SCORE P-VALUE (1-SIDED) P-VALUE (2-SIDED) (Intercept) ≠ 0 5.2433 -24.07 0.00* 0.00* C ont rol v a r. SR HolidayDummy < 0 -0.0094 -4.82 0.00* 0.00* SR Q2_Dummy ≠ 0 -0.0003 -2.76 0.00* 0.01* SR Q3_Dummy ≠ 0 0.0140 -1.53 0.06 ´ 0.13* SR Q4_Dummy ≠ 0 -0.0106 -2.71 0.00* 0.01* Mai n ef fec ts SR EventDummy ≠ 0 -0.0368 -1.16 0.12* 0.25* SR SponsorshipDummy > 0 0.1938 -1.71 0.04* 0.09 ´ SR Advertisingln > 0 0.0032 -3.75 0.00* 0.00* SR Priceln < 0 -1.4201 -52.36 0.00* 0.00* SR SumAdvln ≠ 0 0.0008 -1.20 0.12* 0.23* SR AvePriceln ≠ 0 -0.1055 -5.47 0.00* 0.00* In te rac ti o n s SR EventXAdvln _{SR EventXPriceln} < 0 _{< 0} -0.0032 _0.0559 -1.48 _-2.05 0.07 ´ _0.02* 0.14_0.04* * SR OSXAdvln > 0 -0.0082 -0.73 0.23* 0.46* SR OSXPriceln < 0 -0.1754 -1.87 0.03* 0.06 ´ SR EventXOSXAdvln > 0 0.0210 -0.52 0.30* 0.61* SR EventXOSXPriceln < 0 -0.0395 -1.40 0.08 ´ 0.16* Lagged ma in ef fec ts & int er act. LR Advertisingln > 0 0.0034 -3.10 0.00* 0.00* LR Priceln ≠ 0 -0.0939 -24.08 0.00* 0.00* LR SumAdvln ≠ 0 0.0019 -2.67 0.00* 0.01* LR AvePriceln > 0 0.0275 -7.76 0.00* 0.00* LR EventXAdvln < 0 -0.0046 -0.24 0.41* 0.81* LR EventXPriceln ≠ 0 0.0182 -2.54 0.01* 0.01* LR OSXAdvln ≠ 0 0.0154 -0.09 0.46* 0.93* LR OSXPriceln ≠ 0 -0.0121 0.34 0.63* 1.27* LR EventXOSXAdvln ≠ 0 0.0117 1.50 0.93* 1.87* LR EventXOSXPriceln ≠ 0 0.0160 2.66 1.00* 1.99* LR Adjustment < 0 -0.2842 -47.80 0.00* 0.00*