Discount promotion and the demand for cinema in the Netherlands

(1)

Discount promotion and the demand for cinema in

the Netherlands

Paulina van der Doe

June 2018

Bachelor’s Thesis

Thesis supervisor: Dr. J.C.M van Ophem Student number: 10758178

(2)

This document is written by student Paulina van der Doe who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

1 Introduction

The demand for cinema in the Netherlands has increased over the last ten years, accord-ing to the 2016 annual report from De Nederlandse Verenigaccord-ing Filmtheater en Bioscopen (De NVFB, p.10). However, this demand varies greatly from month to month and from one day of the week to another. As with most leisure activities, demand is higher on Fridays and at the weekends. The report also shows that the demand is lowest during June and from the end of September until halfway through October.

In order to increase demand in this last period, a majority of Dutch cinemas collec-tively participate in a special discount promotion. Consumers can buy cinema tickets at a discount and redeem them at any of the participating cinemas. This promotion first started in 1995, but the details of how it is organised has changed over the years. Since 2016, consumers can avail of the promotion by buying two tickets for 10 euros at the Dutch drugstore de Kruidvat. Depending on the cinema they redeem their coupons at, this gives consumers a discount of approximately 50%. The tickets can only be used on Mondays, Tuesdays, Wednesdays and Thursdays.

Stichting Filmonderzoek, a non-commercial organization that researches the demand for cinema in the Netherlands, has been measuring the effects of this discount promo-tion. Their main focus is to measure the increase in demand for cinema tickets and the revenues lost due to market cannibalization. Market cannibalization happens when consumers who were already planning on visiting the cinema and willing to pay the full price make use of the discount promotion. The model used by Stichting Filmonderzoek has changed several times, but since 2011 the same model has been used every year. However, in 2016 and 2017, the results were significantly different from previous years, and it seems that the model has become outdated. One explanation for this might be that the partner store where the coupons can be bought changed. Another possible

(5)

explanation is that there may be other factors, unobserved by Stichting Filmonderzoek, influencing the demand for cinema.

Measuring the exact effects of a discount promotion on revenues is complicated as the demand depends on various variables, some of which are hard to quantify. According to Blanco and Pino (1997), the demand for cinema tickets is highly elastic and, therefore, revenues are expected to increase when price drops. This is also the conclusion of a study conducted by De Roos and McKenzie (2014). In their research, they evaluated a discount promotion for cinema tickets in Sydney, Australia and found that the revenues did increase. However, it is not only price but also the variety and quality of movies that determine the demand for cinema attendance (Dewenter Westermann, 2005). In order to measure the effect of price, a control variable for quality should be considered. Furthermore, research by Gemser, Oostrum and Leenders (2007) concludes that there is a difference in consumer preferences between those that visit commercial cinemas compared to visitors to so called art house cinemas. Therefore, it might be wise to evaluate these consumers separately.

The goal of this research is to measure the effect of the Dutch cinema discount promotion in its current form on revenues. To achieve this, a linear panel data model is used. The data used in this research were collected from Dutch cinemas for the years 2014 until 2017. In section 2, the theoretical framework is discussed. This framework is based on existing theory, previous empirical research and the models developed by Stichting Filmonderzoek. In section 3, the data, methods and the model used in this research are addressed. In section 4, the results are presented. The last section consists of a conclusion based on the results.

(6)

2 Theoretical Framework

The theoretical framework for this research is built on three foundations. The first and second subsections address theories and previous empirical studies of what influences cinema demand on macroeconomic and microeconomic levels, respectively. In the third subsection, the discount promotion on which this paper focuses is explained and previ-ous market research by Stichting Filmonderzoek is evaluated. The reason for discussing the findings from Stichting Filmonderzoek separately is that these findings are based on surveys rather than on econometric models. Nevertheless, these findings can provide some valuable insights into consumer behaviour in the Netherlands.

2.1 Marcoeconomic research

Relatively little research has been carried out on what influences the demand for enter-tainment, such as the cinema, in the Netherlands on a macroeconomic level. Macroe-conomic research focuses on the aggregated demand in an entire economy rather than on the box office performance of a particular cinema or movie and usually deploys time series data. The box office performance of a cinema or a movie is used to describe either the admissions or generated revenues. In this paper, box office performance refers to generated revenue, which is the most common use of the term.

One of the first that researched the demand for cinema using an econometric model was Cameron (1990). In a pooled cross-sectional time-series study, Cameron finds the demand for cinema to be relatively price and income elastic. This study used data from from the United Kingdom for the years 1966 until 1988 and a fixed effects estimation. It also included control variables: regions, number of cinemas, population and a dummy trend were included.

(7)

between 1950 and 2002 using time series data. While they found that demand decreased significantly during those years, it must be noted that most of the decline in sales was between 1950 and the late 1960s. This decline can be partly explained by the fact that the function of cinema changed during that period. As it became more common for households to own a television, cinema changed from being an important source of news to an entertainment medium. A factor that might have caused even more decline was that this form of entertainment was increasingly regarded as entertainment for the masses during the 1950s (Spraos, 1962). According to Spraos, cinema became an inferior good and as average income increased, demand decreased. After the 1960s, the decline in demand slowed down, and since 1990, demand has even increased. Dewenter and Westermann investigate several factors that influence demand, other than the changed function of cinema. Their most important findings are the long-term effects of price and income. Using both the two-stage least squares (TSLS) and the seemingly unrelated regression (SUR) methods, they found an income and price elasticity above unity and therefore conclude that cinema demand is both price and income elastic.

Similar results can be found in a study conducted by Blanco and Pino (1997), which researches the demand for cinema in Spain using data from the years 1969 until 1992. During those years the demand dropped considerably by 77%. The decline was strongest during the 1980s. Using co-integration analysis, the authors conclude that this decrease was caused by changes in consumer preferences, which saw consumers switch from cinema to television, and relatively large increases in ticket prices. They found a price elasticity of -3.0 and conclude that demand for cinema is price elastic. In addition, they found a income elasticity of 1.6 and therefore classify cinema as a luxury good. Interestingly, demand increased from the year 1988. No explanation for this is suggested, presumably due to the limited number of years in which the increase was observed.

(8)

These findings might indicate that the demand for cinema in the Netherlands is price elastic as well. Although, it is also possible that these findings are not applicable to today’s market since the time periods on which these studies are based ended over 25 years ago.

2.2 Microeconomic research

More studies address the demand for cinema on a microeconomic level than on a macroe-conomic level. A relevant study by De Roos and McKenzie (2014) evaluates the effects of a discount promotion in Sydney, Australia in 2007, whereby cinemas sold tickets at a discount on Tuesdays, in an effort to attract more customers. De Roos and McKenzie investigated whether the overall demand for cinema increased or whether the promotion simply caused consumers to switch the day of attendance. Their sample consisted of 50 cinemas and, to control for heterogeneity across these cinemas, included variables for the number of seats and screens. To control for heterogeneity across movies shown on a given day, they included variables for budget, advertising and reviews of the individual movies, as well as number of awards won. They also included a variable for weather to account for the fact that rainfall might have a significant positive effect on sales. Hav-ing controlled for these factors and used a dummy variable for days when there was a discount, they found that overall revenues increased and conclude that cinema demand is relatively price elastic. They also found a negative correlation between admissions and Oscar awards, which might be explained by the fact that by the time a movie wins an Oscar it has already been shown in the cinema for a considerable amount of time.

Wallentin (2016) found that the perceived quality of a movie, which is often deter-mined by reviews, is an important factor that influences consumer demand. In order to investigate the relationship between box office performance and reviews he used ordi-nary least squares (OLS) regressions. As there is a significant discrepancy between the

(9)

characteristics that audiences favour and those that professional critics value, Wallentin used a combination of professional reviews from Metacritic and peer-based reviews from Rottentomatoes and the Internet Movie Database (IMDb). He concludes that there is a positive relationship between review scores and ticket sales. However, the significance of this relationship varies for different types of movies.

Gemser, Van Oostrum and Leenders (2007) also researched differences in the rela-tionship between review scores and box office performance. In their study of the Dutch cinema market, they found a clear distinction between factors that influenced visitors to art house cinemas and those that influenced visitors to commercial cinemas. Their most important finding was that film review scores are significantly correlated with the box office performance of movies shown in art house cinemas but not with the box office performance of movies in commercial cinemas.

2.3 Discount promotion and Stichting Filmonderzoek

The discount promotion evaluated in in this research began in 1995, and since then Stiching Filmonderzoek has attempted to measure its effects. However, a systematic and scientific approach was only adopted in 2006. The main reason for requiring a deeper insight was to examine the effect on revenues of changing the promotion pe-riod from June to September. Stiching Filmonderzoek’s main focus is to calculate the percentage of market cannibalization during the promotion period. Market cannibal-ization happens when consumers who were already planning on visiting the cinema and willing to pay the full price make use of the discount promotion. Due to the fact that the discount promotion changed, only the most recent model, developed in 2011, is evaluated.

Stichting Filmonderzoek’s model is based on surveys distributed to participants of the discount promotion. One of the questions the respondents are asked is: ‘Would you

(10)

have visited the cinema if there was no discount promotion?’. If a consumer answers yes, they are seen as cannibalistic. The extra visitors that attend the cinema because of the discount promotion is calculated as follows:

Extra visitors = (1 − percentage of market cannibalization) ∗ 2 ∗ redeemed coupons (1) In an intern report from Stichting Filmonderzoek (2017), it is given the number of visitor that made use of a coupon1 was 89,015 in 2016 and in 2017 that number was 153,921. The percentage of consumers participating in the discount promotion classified as cannibalistic was 43% in 2016 and 35% in 2017.

Apart from this question, respondents were asked several other questions about their consumer behaviour. An interesting result of the survey evaluations is that consumers who preferred art house cinemas are more positive towards the discount promotion than consumers who preferred commercial cinemas (Intern report, 2016). One reason for this could be that consumers who prefer art house movies may be less sensitive to price and that their main reason for visiting a cinema is to see the particular movie that is being shown at that cinema at the time. Based on this, there is a strong indication that the price elasticity of consumers who prefer art house cinemas differs from that of consumers who prefer commercial cinemas. Visitors to art house cinemas are therefore more likely to be cannibalistic if they have made use of the discount promotion since they would have bought a ticket regardless of the promotion if they wanted to see a specific movie.

One of the limitations of the model used by Stichting Filmonderzoek is that it might be too simplistic to accurately calculate the extra demand, due to the number of factors that influence the demand for cinema as discussed in sections 2.1 and 2.2. Another limitation is that the model is solely based on surveys. Survey-based research

(11)

is problematic in two ways. First, only the answers of respondents are considered so the research sample is not random. Second, there is often a discrepancy between the revealed preferences of consumers and those stated by the consumers. Stated prefer-ence is the choice that consumers themselves think they will make, whereas revealed preference is the actual choice made. An example of this is the suggestion in the re-port by Stichting Filmonderzoek that respondents might have perceived themselves as stingy if they had answered that they would have not gone to the movies without the discount. This may have resulted in some consumers stating that they would have gone to the cinema without the promotion, whereas in reality they would not. This may have resulted in an upward bias in the cannibalization percentage.

3 Data

Data used in this research was collected from 264 different cinemas in the Netherlands for the years 2014 until 2017. The primary focus of the research is the relationship between the discount promotion and admissions. In order to measure this effect as accurately as possible, other control variables are included.

3.1 Dependent variable

Information about cinema ticket sales in the Netherlands was obtained from Stichting Filmonderzoek, who collected the admission figures per movie per day for each cinema during the years 2014 until 2017, amounting to a total of approximately 2.8 million observations. For this research, admissions are the total admissions per cinema per day, which leaves 281,094 observations. The dependent variable is the logarithm of the admissions per cinema per day. On average, each cinema in the Netherlands sold 670 tickets per day during the years in which the data were collected.

(12)

Observations Mean Standard deviation Min Max Admissions 281,094 670 970 1 13991

Table 1: Admissions

When viewed as a histogram, the logarithm of admissions per cinema per day has a left-tailed distribution.

Figure 1: Histogram admissions

The data also show that the average demand is higher on Fridays and at the week-ends than on weekdays and that the demand varies throughout the year. Demand is relatively low in June and September and high in January, February and December.

(13)

Figure 2: Average admissions per weekday 2014 - 2017

Figure 3: Average admissions per month 2014 - 2017

In 2016, the promotion period was from September 6 until October 14. During that period, a total of 4,083,773 tickets were sold by cinemas participating in the promotion, of which 1,104,607 were sold on either a Monday, Tuesday, Wednesday or Thursday. In 2017, the promotion was held from September 4 to October 12, and the number of tickets sold was 4,426,093, of which 1,245,860 were sold on days when the discount applied. Both the total number tickets sold during the promotion period and the total number of tickets sold at participating cinemas on days when the promotion was valid

(14)

were approximately 10% higher in 2017 compared to 2016. This increase can partly be explained by an outlier in the total admissions on October 5, 2017. On that day the majority of primary school teachers were on strike, which gave primary school students a day off. For this reason, the descriptive statistics of the total admissions excluding October 5 are also provided.

Year Total Promotion Total* Promotion* 2016 4,084,773 1,104,607 -

-2017 4,426,093 1,245,860 4,211,776 1,105,476 *Excluding observations from October 5th, 2017

Table 2: Admissions during promotion period

3.2 Explanatory variables

The variables this paper focuses on are the dummy variables for the promotion in 2016 and 2017. The dummy variables are constructed as follows:

promotion = promotion period ∗ weekday ∗ participating (2) The promotion period is 1 when the date of the observation was between September 6, 2016 and October 14, 2016 or between September 4, 2017 and October 12, 2017. Weekday is defined here as Monday, Tuesday, Wednesday or Thursday. The term ‘participating’ denotes whether or not a coupon could be redeemed at the cinema.

Important control variables that are included are price and income since previous studies have shown that price and income are important factors influencing the demand for cinema. Due to lack of available data on weekly and daily ticket prices, the average price of a cinema ticket per year is used for the price variable. Quarterly growth of

(15)

gross domestic product (GDP) relative to the same period in the previous year is used as the variable for income. The data for these variables was collected by the Centraal Bureau voor Statistieken (CBS).

In order to control for heterogeneity across cinemas, the following cinema-specific variables are included: size, cinema classification and a dummy variable if a cinema did not participate in the promotion. Only the cinemas belonging to the Dutch chain Pathé did not participate. For size, the logarithm of the average number of seats per screen is used in order to measure the relative effect on admissions. Since there is indication that the consumer behaviour of visitors to art house cinemas differs from that of visitors to commercial cinemas, estimations are made using both a full sample and sub samples. Cinemas in the Netherlands can be classified according to five types: commercial, independent, art house, small art house and open air. The cinemas exam-ined for this study are classified as either commercial/independent or art house/small art house. The reason for grouping commercial and independent cinemas together is that these types of cinema show similar movies. Because little data were available for open air cinemas, these are omitted. Cinema-specific data were obtained from an online publication produced by the NVBF (2017). Of the cinemas included in this study, 142 are classified as commercial/independent and 122 as art house.

In order to correct for the heterogeneity between movies, a proxy variable for quality is included. This proxy variable is the review score from the IMDb website, on which users from all over the world can rate movies on a scale of 0 to 10. Zero represents a movie of extremely poor quality and ten is a movie of exceptional quality. The reason why only the IMDb review score is included, is the fact that only these review scores were easily accessible. The IMDb sum of scores for all movies per cinema per day was divided by the number of movies shown that day to obtain the average quality of movies shown per cinema per day. When comparing the average IMDb scores for commercial

(16)

cinemas with those of art house cinemas, it can be seen that the scores for art house cinemas have a sightly higher average mean and spread.

Observations Mean Standard deviation Min Max IMDb Arthouse 97,708 6.88 0.5054 3.1 8.7 IMDb Commercial 183,464 6.56 0.3473 3.1 8.6

Table 3: Admissions

Other control variables used include the number of movies shown per cinema per day, weekdays, vacation periods, national holidays, average rainfall and temperature per day in the Netherlands and a dummy for October 5, 2017. Rainfall is measured in millimeters per square meter and temperature is measured in degrees Celsius. Vacation periods and national holidays were obtained from the official website of the Dutch government, Rijksoverheid.nl and weather-specific data were retrieved from the online database of the Royal Netherlands Meteorological Institute (KNMI). Although the data show that the demand for cinema differs greatly throughout the year, no variables for months are included. This exclusion is due to the fact that the fluctuations in demand for cinema between months can already be explained by other included variables such as vacation periods and weather. For clarity, an overview with a brief description of each variable is provided in table 4.

(17)

Variable Level Description Retrieved by

Admissions Day/Cinema Sold tickets Stichting Filmonderzoek Promotion Day/Cinema Dummy variable for promotion Stichting Filmonderzoek IMDb score Day/Cinema Average IMDb IMDb

Number of movies Day/Cinema Number of movies per day Stichting Filmonderzoek Pathé dummy Cinema Dummy for Pathé cinema De NVBF

Price Yearly Average price per ticket CBS

GDP Quarterly GDP growth CBS

Size Cinema Average number of seats De NVBF Cinema classification Cinema Commercial or Arthouse de NVBF Day Day Weekdays, excluding Fridays N/A

Holidays Day Official holidays Rijksoverheid Weather Day Rainfall and temperature KNMI

Strike Day Dummy for October 5, 2017 N/A

Period Day Promotion period Stichting Filmonderzoek

Table 4: Overview variables

4 Model

The relationship examined in this research can be represented by a two-way error com-ponent model, in which the variables differ over time, between cinemas or both.

yit= Xit0β + 0

Z_i0γ + W_t0δ + uit

uit= vi+ ηt+ it

(18)

Xit represents the variables that differ between cinemas as well as over time. These

are the dummy for the discount promotion, the number of movies shown per day and the average IMDb score of the movies shown per cinema per day. Xi contains information

about cinema specific characteristics. These are the number of screens and whether a cinema participates in the promotion. Wt contains the data that only differs over time:

price, GDP, weekdays, temperature, rainfall, vacation periods, holidays and promotion period. The two-way error component is represented by uit.

When working with panel data, an ordinary least square estimation (OLS) produces biased estimations. This can be explained by looking at equation (3), in which videnotes

the unobservable cinema-specific effect, while ηt denotes the unobservable time-specific

effect, and it can be thought of as the usual disturbance in the regression (Balgati, 2005, pp. 11-33). As vi depends on the individual cinema and ηtdepends on the specific

day, the error term uit is correlated with the explanatory variables Zi and Wt. This

violates the exogeniety assumption of an OLS estimation and results in a biased and inconsistent estimation of the coefficients.

It is, therefore, common in econometrics to use either a fixed effects or a random effects model instead of an OLS estimation. In a fixed effects model, OLS is applied to a modification of the linear equation. The modification in this model is obtained by subtracting the time average of the original equation (4) from the original equation (3). Time dummies are also added to equation (5).

yit = Xit0 β + Z 0 iγ + W 0 tδ + vi+ ηt+ it (4) - y_i = ¯Xi 0 β + Z_i0γ + ¯Wt 0 δ + vi+ ¯η + ¯i (5) yit− yi = β 0 (xit− xi) + δ0(wt− ¯wt) − (ηt− ¯η) + (it− ¯i) (6)

A downside of the fixed effects model is the fact that it removes information about time-invariant characteristics (Heij, Boer, Kloek, Dijk and Franses, 2004, p. 695). A

(19)

random effects model preserves this information. However, a random effects estimation can only be carried out when it is assumed that the unobservable individual and time effects are uncorrelated with most of the regressors. As this assumption often does not hold and a improved model has been developed by Mundlak (1978), the fixed effects method is no further discussed or used in this research.

Mundlak studied both the fixed effects and random effects methods and developed a linear regression model in which the information about the time-constant variables is preserved and the endogeneity bias is resolved by explicitly modelling for this bias. The Mundlak approach adds a high-level mean for every time and cross section varying constant. This means that for every entity a time average for every two-dimensional variable is included ( ¯Xi). In addition to the high-level means, a dummy variable (Di)

for each entity is included, which results in the following model:

yit = Xit0 β + Z 0 iγ + W 0 tδ + ¯Xi 0 θ + D_i0ζ + (vi + ηt+ it) (7)

The validity of the estimations carried out using this model relies heavily on the assumption that there is a linear relationship between Zi and the error term vi

(Mund-lak, 1978). Since this relationship is often hard to prove, most research uses the fixed effects method. However, a unpublished paper by two researchers from the University of Amsterdam (2015), proves that the validity of the Mundlak model holds for any kind of panel data and that it is not necessary for the relationship between Zi and vi to

be either linear or estimated. Therefore, the Mundlak model is used in this research without identifying the relationship between Zi and vi.

(20)

5 Results

In order to the estimate the effects of the discount promotion, both a fixed effects model with dummies and a Mundlak model are used. The reason for reporting both Mundlak and a fixed effects estimation is to show that the Mundlak model is correctly specified. If the Mundlak model is correctly specified the estimations of the coefficients should be more or less the same as the estimations of the fixed effects model. This comparison is relevant as it validates the use of the Mundlak model specified in (7) as mathematical proof for this is not publicly available.

In this study, time (number of days), is much larger than entities (number of cine-mas), which is rare for panel data. Therefore, in order to get more accurate estimates, time and entities are reversed in the construction of the model. The results show that there is very little difference between the parameters estimated by the fixed effects model and those estimated by the Mundlak model. Because time and entities are reversed, the cinema-specific information is retained but the the time-dependent information is lost. The higher level cinema averages of the promotion for 2016 and 2017, the average IMDb score and the number of movies as well as the dummies for cinemas are excluded from the results as they do not give valuable information. All models are estimated using heteroskedasticity-robust standard errors.

Besides these two models, two Mundlak models are also used using observations solely from arthouse cinemas and commercial cinemas, respectively. No fixed effects model is used for the particular cinema types due to information loss and, more im-portantly, the fact that the fixed effects estimator cannot estimate the effects of the promotion for art house cinemas. Since all cinemas in the subgroup art house partic-ipate in the discount promotion, the variable promotion is changed from a time- and entity-dependent variable to a variable that is only time dependent.

(21)

Table 5: Regressions results

Variable Fixed effects Mundlak Variable Fixed effects Mundlak Promotion 2016 0.154*** 0.158*** Vacation Middle 0.211***

(0.0652) (0.0261) (0.00486)

Promotion 2017 -0.00230 -0.00350 Vacation South 0.0545***

(0.0509) (0.0196) (0.00462)

IMDb 0.113*** 0.109*** New Year’s eve -0.700***

(0.00797) (0.00489) (0.0228)

Price -0.729*** New Year’s day -0.507***

(0.0140) (0.0226)

GDP 0.0258*** Christmas 0.0434***

(0.00297) (0.0171)

Rain 0.0110*** Easter 0.922***

(0.000255) (0.0201)

Temperature -0.0378*** Good Friday 0.344***

(0.000231) (0.0190)

Pathe Dummy 0.148*** 0.0840*** Kingsday -0.0297***

(0.0131) (0.0143) (0.0257)

LogSize 0.897*** 0.855*** Ascension day 0.513***

(0.00609) (0.00172) (0.0290)

Number of movies 0.103*** 0.111*** Whit Monday 0.631***

(0.00120) (0.00497) (0.208) Monday -0.548*** Strike 0.329*** (0.00414) (0.0477) Tuesday -0.315*** Period 2016 0.254*** (0.00392) (0.0107) Wednesday -0.267*** Period 2017 0.0928*** (0.00392) (0.00979) Thursday -0.321*** Constant -0.357*** 8.279*** (0.00392) (0.106) (0.168) Vacation North 0.148*** (0.00410) Fixed effects Mundlak Observations 281,094 281,094 Adjusted R2 _0.822 _0.810

F-test 2767 4086

Prob > F 0 0

(22)

Table 6: Regressions results

Variable Commercial Art house Variable Commercial Art house Promotion 2016 0.149*** -0.0468*** Vacation Middle 0.254*** 0.142***

(0.0267) (0.0265) (0.00575) (0.00887) Promotion 2017 -0.0529** 0.0108* Vacation South 0.0597*** 0.033***

(0.0211) (0.0239) (0.00548) (0.00829) IMDb 0.0777** 0.126*** New Year’s eve -0.742*** -0.696*** (0.00804) (0.00631) (0.0252) (0.0463) Price -0.897*** -0.313*** New Year’s day -0.552*** -0.481***

(0.0166) (0.0249) (0.0259) (0.0460) GDP 0.0668*** -0.0612*** Christmas 0.0527*** -0.0325**

(0.00349) (0.00533) (0.0200) (0.0336) Rain 0.0120*** 0.00935*** Easter 1.111*** 0.571***

(0.000305) (0.000456) (0.0210) (0.0399) Temperature -0.0376*** -0.0366*** Good Friday 0.391*** 0.262***

(0.000268) (0.000439) (0.0227) (0.0352) Pathé Dummy 0.234*** Kingsday -0.0120 -0.0716

(0.0138) (0.0285) (0.0540)

LogSize 1.289*** 0.295*** Ascension day 0.389*** 0.177*** (0.0281) (0.0463) (0.0281) (0.0456) Number of movies 0.102*** 0.130*** Whit Monday 0.813*** 0.632***

(0.00856) (0.0009999) (0.0220) (0.0207) Monday -0.657*** -0.355** Strike 0.535*** -0.0489 (0.00492) (0.00731) (0.0511) (0.0475) Tuesday -0.382*** -0.198*** Period 2016 0.300*** 0.189*** (0.00467) (0.00694) (0.0121) (0.0203) Wednesday -0.304*** -0.187*** Period 2017 0.0974*** 0.0942*** (0.00432) (0.00675) (0.0110) (0.0186) Thursday -0.384*** -0.224*** Constant 8.063*** 5.508*** (0.00468) (0.00699) (0.209) (0.331) Vacation North 0.191*** 0.0778*** (0.00484) (0.00743) Commercial Art house Observations 183,461 97,633 Adjusted R2 _0.811 _0.748

F-test 4307 2154

Prob > F 0 0

(23)

6 Analysis

This research primary focus is the effect of the discount promotion. The estimations of the Mundlak model and the fixed effects model are roughly the same, which means that the Mundlak model removed most of the endogeneity between the cinema and time specific regressors and the error term. For this reason, only the Mundlak model is discussed, with exception of coefficients whereby the estimation results strongly differ. The effect of the 2016 and 2017 promotion are discussed first. In the second part of the analysis, the results of the other explanatory variables are discussed.

6.1 Effect of the discount promotion on admissions

According to the estimated Mundlak model, using the full sample, the effect of the discount promotion in 2016 is positively correlated with the number of sold tickets in 2016. According to the estimation, the promotion in 2016 has lead to an increase in sales of 100 ∗ (e0.154_{− 1) = 16.65 %. This implies that without this promotion, approximately}

1, 104, 607/116.65 ∗ 100 ≈ 950, 000 tickets would have been sold. This means that there were 1,104,607 - 950,000 ≈ 150,000 extra tickets sold as result of the promotion. This is remarkable as the number of sold tickets with a discount was only 810,015. This regression result can partly be explained by the success of the movie Bridget Jones’ Baby. According to the 2016 report from Stichting Filmonderzoek (2016), it is unusual that a movie released in September attracts so many visitors as Bridget Jones’ Baby did. The fact that the movie was released in the same period as the promotion most likely have caused an upward bias in the estimation of the coefficient of the promotion of 2016. The success of Bridget Jones’ Baby can also explain the difference between the estimated coefficients in the commercial and art house sub samples. The effect of the promotion of 2016 on the ticket sales of commercial cinemas was positive while the

(24)

effect of the promotion was negative for art house cinemas. Bridget Jones’ Baby was mostly played at commercial cinemas, possibly causing consumers who would otherwise have visited an art house cinema to switch to a commercial cinema.

The coefficient of the discount promotion in 2017 is not significant. This implies that the promotion was more successful in 2016 than 2017. This result is remarkable as well given that the number of coupons sold was about 77% higher in 2017 than 2016. An explanation for this might be that the dependent variable, the total admissions per day per cinema, was transformed by taking a logarithm. When a logarithmic transformation is taken of a variable, there is an automatic correction for outliers. Therefore, it is possible that the total coupons sold in 2017 was higher, but that the promotion only led to an strong increase in demand on certain days. According to the 2017 report from Stichting Filmonderzoek, there were three days on which the admissions strongly differed with the usual admissions. Two of the outliers were caused by cinemas holding special events. The other outlier was October 5th, when primary school teachers in the Netherlands went on a strike.

6.2 Control variables

Looking at the estimated coefficients of the control variables, some interesting results can be observed. One of them is that size has significant positive influence on admis-sions. Interestingly, this effect differs greatly when the sample is split into the two sub samples. Size has a coefficient of 1.289 in estimation based on the commercial sub sample 0.295 in the art house sub sample. This means that an increase in size of about 1% would increase the demand for commercial cinemas by 1.3% while this increase is only 0.3% for art house cinemas. A possible explanation might be that visitors of com-mercial cinemas favor larger cinemas as they tend to have more facilities, for example more food and beverage options or special events. According to the survey

(25)

evalua-tions from Stichting Filmonderzoek (2017), the main reason why consumers visit an art house cinema, is the perceived quality of the movie. Visitors of art house cinemas might, therefore, value such extra facilities less.

The influence of rain and temperature is roughly the same for the full sample and both the sub samples. The estimation of all of the coefficients are highly significant. The influence of weather can easily be explained by the fact that visiting a cinema is an indoors activity. When the outside temperature is higher, consumers prefer to be outside. When temperature is low, consumers prefer to be inside. The opposite is true for amount of rainfall. Although the relationship between weather and the demand for cinema might seem obvious, it is still worth mentioning as this relationship has rarely been discussed in previous research, with exception from the research by De Roos and McKenzie (2014).

IMDb score has a significant positive effect on demand. The effect of IMDb score is greater for art house movie theaters than for commercial cinemas. An increase of the average IMDb score per day caused a increase in demand of approximately 12.6% and 7.7%, respectively. This result can be explained by the difference in consumer behaviour and preference and is in line with previous empirical studies from Gemser, Oostrum and Leenders (2007) and Wallentin (2016) and the survey results of Stichting Filmonderzoek (2017). This result might also be an an indication that movie review scores in general have a significant effect on the demand for cinema. However, it has to be noted that IMDb score is a far from perfect measurement of review scores in general. Many other review score sources exist that were not taken in account in this research and possibly differ in review scores from the IMDb scores for the individual movies. Other review scores should be included as well in order to provide a conclusion about the effect of reviews in general.

(26)

misleading to conclude a price elasticity based on these results as only four different prices (average price for 2014, 2015, 2016 and 2017) were taken into account. Daily ticket prices per day per cinema should have been included in order to calculate a more accurate effect of price.

GDP growth has a positive effect on admissions in general. However, when the sample is split, the coefficients show an opposite sign. The effect of GDP is positive for commercial and negative for art house. Based on the economic theory and the data, no explanation can be given for this. Possibly, there were other unobserved factors correlated with GDP growth that are not included.

As expected based on economic theory and descriptive statistics, weekdays have lower sales of tickets compared to weekends. Of all the weekdays, Wednesdays had the smallest negative effect, meaning the average admissions is higher than other weekdays. This can be explained by the fact that traditionally most primary school in the Nether-lands are closed on Wednesday afternoons, giving children aged 4 until 12 a part of the day off. When comparing the two sub samples, the effect is greater for commercial cinemas. This is possibly due to the fact that movies targeting children are played more often in commercial cinemas than art house cinemas.

The fact that commercial cinemas play relatively more movies targeting children can also explain the difference in the estimated coefficients for the vacation periods, holidays and the dummy for October 5, 2017. The effects of all three of the vacation periods are significant and positive and greater for commercial cinemas than art house cinemas. The same applies to the coefficients for Good Friday, Easter, Ascension day and Whit Monday. The dummy variable for Christmas is positive for commercial but negative for art house. For the holidays New Year’s Eve, New Year’s Day the effects in both the sub sample were negative. However, the effect was smaller for commercial cinemas than art house cinemas. The strike on October 5 caused an increase in admissions of

(27)

100 ∗ (e0.535 − 1) = 70.74% for commercial cinemas while no significant change was observed for the art house cinemas. No significant result was obtained for Kingsday.

The Pathé dummy was positive for both the Mundlak model and the fixed effects model, although the effect differed by approximately 14.8 - 8.4 = 6.8 %. This difference is surprisingly as almost all the other coefficients are roughly the same. The reason for this is unknown. Nevertheless, it can be assumed that an positive effect is correct. Pathé is the market leader when it comes to chains of cinemas in the Netherlands and is well know. This caused the demand for tickets higher. The Pathé dummy is omitted in the regressions using only the art house sub sample as no Pathé cinema can be classified as art house.

7 Conclusion

In this paper the effect of a discount promotion on the demand for cinema in the Netherlands was researched. In order to answer this question an empirical research was conducted based on panel data from the years 2014 until 2017. The effects were estimated using an econometric model.

Based on the obtained results no conclusive answer about the effectiveness of this promotion can be provided. The results show that the promotion in 2016 would have attracted more extra visitor than the number of tickets that were sold with a discount. This would imply a market cannibalization rate of 0%. For the promotion of 2017 no significant change in admissions were observed. This would imply that the promo-tion did not increase admissions and that the cannibalizapromo-tion rate was equal to 100%. Both these results differ so greatly from economic theory and the market research from Stichting Filmonderzoek, that the conclusion is that the model and methods used in this research are not adequate enough to measure the exact effects of the discount

(28)

promotion.

Possible reasons for the inability of the model to measure the effects are model misspecification and lack of data. The model might have suffered from omitted variable bias by not including control variables for special events that might have caused outliers in the dependent variable. Heterogeneity across movies might have even caused more bias. To control for this heterogeneity, only review scores were taken into account. Possibly, there are various other factors that influence the popularity of a movie, such as advertising or budget. In addition to this there, there could be a lack of sufficient data. The promotion in its current form only exists for two years. This might not be enough to accurately measure the effect.

Despite the fact that the estimations obtained in this research did not provide a sufficient basis for a conclusive answer about the effectiveness of the promotion, interesting results about the control variables were found. The results indicate that there is a difference in consumer behaviour between visitors of commercial cinemas and art house cinemas. Especially when it comes to the effect of review scores and size on admissions. A higher IMDb score has a greater positive influence on demand for art house cinemas than it does for commercial cinemas. For size the opposite is true. Besides this, it can be concluded that schools being closed on a given day increases the demand for cinema tickets more for commercial cinemas than for art house cinemas.

Further research is required to determine the exact effect of the discount promotion of the demand for cinema. In further research more variables to control for the hetero-geneity across movies should taken into account. Furthermore, more data or another model might be needed to accurately model the demand for cinema.

(29)

References

Author 1, Author 2. (2005). Estimation of the impact of time-invariant explanatory variables in panel data: A reappraisal of Mundlak (1978). Unpublished.

Baltagi, B. H. (2005). Econometric analysis of panel data (Third edition). Chichester, UK: John Wiley Sons.

Blanco, V. and Banos Pino, J (1997). Cinema Demand in Spain: A cointegrating analysis. Journal of Cultural Economics, 21 (1):57-75.

Cameron, S. (1990). The demand for cinema in the United Kingdom. Journal of Cul-tural Economics, 14 (1):35-47

Dewenter, R. and Westermann, M. (2005). Cinema Demand In Germany. Journal of Cultural Economics 29 (3):213-231

Gemser, G and Oostrum, M van and Leenders, M. (2007). The impact of film reviews on the box office performance of art house versus mainstream motion pictures. Journal of Cultural Economic, 31 (1):43-63.

Heij, C. and Boer, P. d and Kloek, T. and Dijk, H.K. van and Franses, P.H. (2004). Econometric Methods with Applications in Business and Economics. Oxford University Press.

The Internet Movie Database. (n.d.). [Online database]. Retrieved May 27, 2018 from https://www.imdb.com/interfaces/

The Royal Netherlands Meteorological Institute. (2018). [Online database]. Retrieved May 12, 2018 from http://projects.knmi.nl/klimatologie/daggegevens/selectie.cgi.

(30)

Mundlak, Y. (1978). On the pooling of time series and cross section data. Economet-rica, 46 (1):69-85.

De Nederlandse Vereniging Filmtheater en Bioscopen. (2016). Annual Report. Retrieved May 5, 2018 from https://www.denvbf.nl/files/2016-jaarverslag-nvbf-online.pdf

De Nederlandse Vereniging Filmtheater en Bioscopen. (2017). Boekingsgroep, stoelen, zalen. Retrieved May 5, 2018 from https://www.denvbf.nl/files/20170703-leden-nvbf-op-boekingsgroep-stoelen-zalen.pdf

Roos, N. de and McKenzie, J. (2014). Cheap Tuesdays and the demand for cinema. International Journal of Industrial Organization, 33 (1) :93-109.

Spraos, J. (1962). The decline of the cinema: and economist’s report. Londen, United Kingdom: Allen Unwin.

Stichting Filmonderzoek (2016). Evaluation of the discount promotion. Intern rapport, not publicly available.

Stichting Filmonderzoek (2017). Evaluation of the discount promotion. Intern rapport, not publicly available.

The government of the Netherlands. (2017). Retrieved May 5, 2018 from https://www.rijksoverheid.nl/onderwerpen/schoolvakanties

Wallentin, E. (2016). Demand for cinema and diverging tastes of critics and audiences. Journal of Retailing and Consumer Services, 33 (3) :72-81.

Discount promotion and the demand for cinema in the Netherlands