An investigation in the price determinants of video games and their effect on player bases

(1)

Faculty of Economics and Business

MSc Marketing Intelligence

An investigation in the price determinants of video games and

their effect on player bases

Master thesis by Redmer Nijboer S2787474

Koeriersterweg 3 9727 AA, Groningen r.t.nijboer@student.rug.nl

(+31)6 839 031 15

Supervised: dr. K. Dehmamy, co-assessed by prof. dr. J. Wieringa k.dehmamy@rug.nl, j.e.wieringa@rug.nl

MSc Thesis Marketing Intelligence 17th of June 2019

Department of Marketing PO Box 800

(2)

2

Abstract

The video game industry has been growing massively over recent years, becoming the most valuable industry within the entertainment sector. Nevertheless, little academic research efforts have been spent on trying to understand its workings. In this paper, the prices of the most grossing video games of 2016 will decomposed by means of a dynamic hierarchical factor model. In turn, with the derived factors of the model, the number of players for different genres of games will be estimated via vector autoregressive models. From these analyses it is found that common market wide, genre level and country level dynamics exist, each having a different impact on the player base. Additionally, evidence is found for the presence of network effects between players and a substitution good for video game consumption. The application of the dynamic hierarchical factor model and the resulting findings in this paper are deemed to be relevant for both the academic literature as well as the practical side of marketing.

(3)

3

Preface

In the year 2014 I started out as a student of Economics & Business Economics. In these three years of being a BSc student I became familiarized with economic thinking and developed an interest in the international side of economics. As a result, the track International Economics & Business was chosen. Upon finishing my Bachelor’s I felt I was not done studying and wished to extend my period as a student. Over the years I also became fascinated by mathematics and physics and even considered switching career paths. During the information sessions of these programs I discovered that such a move was not for me and thus I continued with the MSc International Economics & Business. Late last year, it was time to write the thesis for this program but this project felt daunting for me, thus I extended my time as a student even further and picked the MSc Marketing Intelligence.

Simply put, this may have been the best choice in my student life as the program strikes, in my opinion, a perfect balance between business, economics and data analytics. Over the last couple of months I enjoyed the courses provided and it feels like I have expanded my ‘toolkit’ substantially. Now, with this final piece of work (i.e. MSc thesis) I hope to show that the efforts of the last year prove to be fruitful.

All of this would not have been possible without the lectures and tutorials that have been given by the professors of the Marketing department at the RUG. I wish to express my gratitude for your enthusiasm during this academic year and hope that you will continue to push future students to a higher level. A special thanks goes out to dr. Keyvan Dehmamy, who has been a constant factor (pun intended) throughout this year. Your willingness to help me and peer students with their work has been remarkable. In addition, your support while writing this thesis was superb and without your coding skills we as students of marketing would know a lot less. Then another special thanks to prof. dr. Jaap Wieringa. The lectures you gave this year were of outstanding quality in my view and also during tutorials you were very helpful.

Lastly, I want to thank my friends, family and fellow students for their thoughts and support during this rather stressful period of thesis writing. Thanks a lot!

Redmer Nijboer

(4)

4

1. Introduction

(7)

7

2. Literature review

2.1 Similarities to traditional art

(8)

8 2.2 Similarities to the music industry

(9)

9 to labels and therefore artists, which have already taken negative stances on the rise of streaming platforms (Byun, 2016). The financials of other major streaming platforms other than Spotify and Pandora are difficult to assess as they fall under large corporations (i.e. Amazon, Apple, Google), but given the royalty payments for each (Information is beautiful, 2019), it seems reasonable to expect similar net incomes for these services. In sum, it seems that the music streaming industry with its current pricing is not likely to persist in its current form and, more importantly to our analysis, does not display the price dynamics as observed in video games. The now outdated and aforementioned CD pricing is considered to be more similar to game pricing, however, research regarding dynamics is lacking.

2.3 Similarities to the film industry

Next to music, films are also an entertainment product which can be considered a form of art and also shows similarities to video games. Furthermore, it has experienced similar developments to music with regards to its product offerings with the introduction of streaming services as a replacement for DVDs and Blu-Rays. In fact, two years ago, the market for streaming, led by firms as Netflix and Amazon, outsized the market for physical copies of films and, like for music, this trend is expected to continue in the upcoming years (Fruhlinger, 2019: Sweney, 2017). The DVD market, however, has persisted for quite a number of years and, like the music industry’s CD market, resembles the market for video games more closely than the streaming market. A formal competitive analysis of the six major film studios in Hollywood (i.e. Columbia, Disney, Fox, Paramount, Universal, Warner) has been conducted by Mukherjee & Kadiyali (2018). In their analysis the authors estimate demand for a DVD as a function of several variables, including film studios, film genres, ratings, state dependence of the consumer, seasonality effects and time from theatrical to DVD release. In addition, they separate their estimations by accounting for the degree of freshness of a certain consumer, thereby dividing their sample into two groups. It is found that indeed such a distinction between two types of consumers is justified suggesting that the willingness-to-pay of different groups of consumers differ, thus for these different groups the estimated effects for demand differ. The freshness-sensitive group are more freshness-sensitive to price, shorter release times (both theatre to DVD and DVD newness), animated and R-rated movies. Moreover, they show more positive responses to several genres of films, including action/adventure, comedy and drama. These results are informative when it comes to our price analysis of games. Like films, video games have a variety of genres as well (e.g. shooter, puzzle, action), for which, in line with these findings, consumers may have systematically different preferences, which may be propagated into distinctive willingness-to-pays as well. Hence, the price determination in genres of films might share commonalities with those of a video game. Therefore, we construct the following hypothesis.

H1: Prices of games of the same genre are affected equally.

(10)

10 between the two types of retailers over time. This finding suggests that initial prices are set by the retailer and that there is a market-wide force that decreases prices. Besides it shows similarities to the findings of Rabinovich et al. (2008) in the market for CDs. Additionally, in line with expectations, popular DVD titles are priced differently than random titles. Moreover, individual variability of prices varied substantially, signifying the heterogeneity among the DVDs. It is also found that for dotcom retailers the price dispersion between DVD types (i.e. popular vs. non-popular) increases over time. To summarize, in the market for DVDs the determinants of initial prices are popularity of a title, the type of retailer and for the development of prices there exists a market-wide trend as well as a price dispersion effect DVDs for dotcom retailers. We note here that the sample of DVDs the authors analysed was rather small, only consisting of 26 titles per DVD type, measured at five multichannel and five dotcom retailers. This could be considered a weak aspect of the analysis conducted by the researchers, yet the findings provide some more insight into the workings of DVD pricing. From these two findings we establish another hypothesis.

H2: Popular video game titles are priced more highly than less popular titles. 2.4 The video game industry

(11)

11 100,000 to 900,000 owners, discount averages were 153, 269 and 333 days for 50%, 66% and 75% respectively in 2014. Again, similar trends were observed for the year 2015, denoting 67, 110, 189 and 353 days for discount averages of 25%, 33%, 50% and 75% respectively. All these figures are summarized in table 1. To provide a little more insight some data are linearly inter- or extrapolated by using the values closest to the to be calculated value. However, as the analysis is informal and executed over a small time frame, there are no clear conclusions to be drawn. Moreover, the data are incomplete and inconsistent. The same author analyses the price dynamics of games that do not launch at the traditional price of 60 euros, but at a sub-30 euro price. That is, the smaller titles are analysed as well and the price dynamics of these games are summarized in table 2. Again, like in table 1, interpolation and extrapolation were applied. From these two meagre sets of data we can, with a considerable amount of creativity, identify some preliminary trends about the development of prices for games. Firstly, it seems that for both sets of games the more popular ones take longer to get steeply discounted. Intuitively this is in line with expectations as popularity proxies for demand. This exploratory inference also displays parallels to the findings by Li and Tang (2011) for the DVD market. Furthermore, from the data observed, there appears to be a trend of triple-A titles at the $60 launch price point take longer to get reduced in price than games that are released at prices below $30, if we allow normalization for the amount of owners for each set of games. When we do not consider this there is no pattern to be observed. Hence, we still know relatively little about price dynamics of games on Steam. The main inference from this analysis is that bigger titles’ prices remain higher over time. Such a force may be determined by the influence of the large publishers in the video game industry trying to maximize profits as they need to report to shareholders. Another explanation could be the relative production quality of these type games compared to smaller titles, as a result of which their perceived value and price are sustained over time.

MEDIAN AVERAGE DAYS NEEDED TO REACH DISCOUNT

Games having 900K+ owners Games having 100K-900K owners

DISCOUNT(IN %) 2014 2015 2014 2015 25 - 61 - 67 33 58 106 - 110 50 192 223 153 189 66 348* 333* 269 353 75 435 395* 333 445*

Table 1: Steam discounts for games with launch price of $60. * Marks inter- or extrapolated data points.

MEDIAN AVERAGE DAYS NEEDED TO REACH DISCOUNT Games having 250K+ owners Games having 30K-250K owners

DISCOUNT (IN %) 2014 2015 2014 2015 20 98 25 99* 69 33 100 115* 40 168 140* 50 166 279 182 158 66 311 - 218* 209* 75 348 - 239 237 80 369* 462 90 446

(12)

12 Considering the findings of Li & Tang (2011) regarding market-wide price effects as well as those of the aforementioned observations regarding the price developments of video games we hypothesize the following.

H3: Prices are affected equally due to a market trend.

Next to the variation in prices of games due to discounts, prices can also fluctuate significantly per country on the Steam platform. Even when accounting for differences in exchange rates the prices of games can differ dramatically, with discrepancies as large as 60% (PCGamer, 2014). The exact origins of these differences are unknown, however, it seems reasonable that regional trends, exchange rates, demand and economic conditions dictate them. However, despite all of these conditions prices may differ across countries. In the empirical part of this paper we aim to shed some light on this by analysing the following hypothesis.

H4: Prices of different genres are set independently of the country they are sold in. 2.5 Recent developments in video game pricing

At this point it may be sensible to mention a couple of recent trends in the video gaming industry with regards to the games’ pricing. One of these developments is the rise of downloadable content which, as of 2010, represent 24% percent of sales in console gaming (Lizardi, 2012). Downloadable content (DLC) can be viewed as an extension to a game that gets released after an x amount of time after the original game is released. Although not new to the gaming world, the increase in their prevalence is noteworthy. These packages are meant to increase the monetization of a video game and to increase the total price of the video past its industry standard €60 price point. Furthermore, the DLCs can be viewed as a manner to increase a game’s life span. The rise of DLC vary depending on the size and content of package. For example, DLC can be minimal (e.g. cosmetics, weapon) often referred to as ‘micro-transactions’ or sizeable (e.g. another world-expanding storyline), each with corresponding prices. As a result of the increase of consumers’ habituation to DLC, it is suspected that some video games are not released in full but some content is removed such that it can be released as DLC in the future (Taormina, 2015: Cox, 2012). As a consequence of the increase of the sheer amount of DLC offerings per title, publishers have introduced so called season passes for video game titles, allowing players that have purchased such a pass to download (almost) all DLC that becomes available for the game. The rise of DLC has yielded the popularization of a new business model in video games; the free-to-play model with in-game transactions. One of the most notable exponent of this model is the game Fortnite which, as of August 2018 has 78.3 million active monthly users across all platforms and has gathered as of June 2018 $1.2 billion in revenue. Looking at individual revenues instead of aggregates, a study by LendEDU finds that 70% of Fortnite players spends $85 on average (Iqbal, 2019). Given the traditional $60 price for a title, the revenue captured by the publisher are almost the same with a free-to-play model (calculated to be $59.5). However, with the free-to-play model, the barriers to entry are significantly less. Fortnite’s free-to-play model is not alone in being highly successful. In fact, the majority of the most played video games worldwide utilize this model (Wikipedia, 2019). 2.6 Factors affecting the number of players

(13)

13 undertaken and the forthcoming systematic patterns of the consumption of these products have been lacklustre. Like for the pricing in the previous chapter, quite a bit more research has been conducted for other entertainment products. For example, the film industry has been investigated quite heavily. By estimating a multiple regression model, Litman (1983) identified a number of significant variables affecting the success of a movie at the box office, a a measure of success similar to our variable of number of players of games. Among the variables were the production costs of the movie, most genres, a dummy for a Christmas release, award nominations as well as award winnings. Most of these are proxies for quality (i.e. production costs, award variables). Further research in this strand of the literature expanded the set of variables and type of models used to analyse the same problem. Factors like having a movie star in the cast, informational signals, whether the movie was a sequel or not and the reception by critics have been investigated (De Vany & Walls (1999): Ravid (1999): Pangarker & Smit (2013). It is found that all play their part in determining the performance of a movie at the box office. For video games, however, only a few of these variables may apply (e.g. sequel, award variables, critics reception) whereas others (e.g. informational actor, having a movie star) may not. Whether these variables are transferred to the world of video gaming and if they exert forces in shaping its components has yet to be examined by the academic world.

For the CD industry we would ideally have access to the data of streaming services such as Spotify as these would provide us with insights into how genre- and country factors affect the listening behaviour. Still, this data would be devoid of the price changes and their effects; the thing we are interested in. As these data would by no means ever be shared with another party other than the firm itself, the academic literature is what we have to resort to. Like mentioned before, the entertainment industry has, despite its size not received extensive attention over the years. The sparse amount of research that has been conducted delved into the determinants of CD sales. This is a measure most similar to the amount of players of a game as there has been no academic investigation into the determinants of music listens. In a paper written almost a decade ago, Elliot and Simmons (2011) study this matter. What they find is that there are little systematic determinants of the amount of CDs purchased, other than the Amazon sales rankings. In the same paper, they also investigate for the effects of ranking on the amount of CD sales, yet, contrary to the movie industry, the authors are unable to find a significant relationship. Therefore, as of yet, not many determinants have been found other than the lagged effects of the dependent variable. Such an effect may also be of relevance for the gaming industry, in which network effects, either via multiplayer games or via social interaction, influence the amount of players of a video game. As a consequence of the network effects, the number of players of a video game may display nonlinearities which can be initiated by price decreases. This type of nonlinear relationship appears not to be restricted to the box offices revenues, but transmission into the DVD sales of the movies as well. McKenzie (2009) studied the Australian DVD retail sales following a theatrical release. It was discovered that the movies with higher theatrical revenues, the star movies so to say, exhibited increasing returns to their box office revenue when it came to their DVD sales. In the video game world there is, however, no pre-release similar to the movie industry. One could draw similarities to a game release and the subsequent DLCs releases, yet, given the interdependencies between these two the similarity runs eery.

(14)

(15)

15

4. Methodology

4.1 A visual representation of the DHFM

The model that will be used to assess the dynamic pricing effects in the market for video games will be similar to the one developed and used by Moench et al. (2009). The authors applied the so called dynamic hierarchical factor model to measure economic activity by factoring 447 time series, thereby reducing dimensionality. Additionally, upon usage of this model, they were able to exploit the time inconsistencies within their data. Moreover, the model of Moench et al. is hierarchical of nature, allowing for a ranking of effects. Specifically, the structure of the model can be viewed as an accumulation of different blocks which each have their own distinct effect on the prices. The imposed design causes a difference in the amount of variance of each block, which is dependent on the exposure of a certain block b to the common block higher in the hierarchy. Hence, the structure of the hierarchy implies that there is a degree of coefficient restriction in the model. Thus, the hierarchy allows for greater flexibility in estimating price effects.

(16)

16 From the literature, several price effects have been identified. Although holding for different products, these may be applicable to the video game market given the similarities between them. From the analysis of Li & Tang (2011) it is found that there is some type of a market-wide pricing effect that affects all prices in the market of CDs. Given the similarities between videos games and CDs with regards to nature of the products, systemic pricing force affecting all games seems plausible. Furthermore, a genre type pricing effect as well as a country-level pricing effect is included. The genre effect is resultant from the findings of Mukherjee & Kadiyali (2018) and the forthcoming hypothesis that different pricing strategies and preferences exist between genres. Here, it may be true that some genres are more competitive or popular than others, which subsequently leads to either a downward or upward pricing trends. The country effect is a level in the hierarchy that accounts for the fluctuations in prices between countries whilst controlling for exchange rates. In the data, prices for the same game may differ as much as 75%, as in the case of Rocket League (Steamdb, 2019). It seems that exchange rates by themselves cannot fully account for such a deviation in the prices and therefore it seems reasonable to allow for such an effect in the model. In constructing the hierarchy in our dynamic hierarchical model we place the market-wide effect at the top of the hierarchy, as this effect is said to influence all games analysed. Furthermore, we consider the fact that DVDs are being demanded and priced differently due to their genre and popularity. Therefore, it is hypothesized that these genre pricing effects are dominant over country effects as games of a certain genre are considered to be in direct competition with one another in multiple countries. Consequently, the country effect is placed the lowest in the hierarchy. Another reason for this is that the variation in prices may differ for no apparent reason from game to game dependent on the country and therefore assuming dependencies of genre effects on country effects is deemed incorrect. Furthermore, it is plausible that certain games may be more popular in certain countries due to for example network effects or local preferences. Whilst sharing common block effects, there is room for individual deviation, which are modelled for by including error terms for each common block. All of the above considered, the hierarchy of effects that influences the prices of video games can be visualized as above in figure 1.

4.2 A formal representation of the DHFM

Correspondent to this hierarchy we can specify the model in a more formal manner. In its formulation we account for the effects coming from the common block movements as well as block-specific movements. This means that all but one of the levels of the hierarchy experiences shared effects. As the prices develop over time, the effects for a block every common block b are serially correlated over time. Considering all of the above we specify the following formulas to represent the model.

𝑃_𝑏𝑖𝑡 = ρ_{𝑐.𝑏𝑖}(𝐿)𝐶_𝑏𝑡 + ε_{𝑃𝑏𝑖𝑡} 𝐶_𝑏𝑗𝑡 = ρ_{𝐺.𝑏𝑗}(𝐿)𝐺_𝑡+ ε_{𝐶𝑏𝑗𝑡} 𝐺_𝑏𝑘𝑡 = ρ_𝑀𝑘𝑡(𝐿)𝑀_𝑡+ ε_{𝐺𝑏𝑘𝑡}

𝑀𝑘𝑡 = ε𝑀𝑘𝑡

(17)

17 algorithm. In the model, (1) refers to the actual observed time series in the data, which are in our case the weekly prices of the video games. We find the data in the lowest part of the hierarchy in figure 1. In its definition we find that the prices of games are directly influenced by country effect as well as an idiosyncratic component. The amount of country factors in the model is set at five, where the countries are selected based upon their relative share of total revenue in the PC gaming industry. This will be elaborated upon in the upcoming chapter. In turn, in the second formula, the country-level block is affected by genre-level next to an individually varying component. The number of gaming genre blocks is determined by a hierarchical cluster analysis, which again will be further explained in the next chapter. Resultant from this analysis, four factors are identified. Lastly, in formula (4) the relationship between the genre block and the market-wide effect is formalized. As the market-wide effect affects all games, there is only one block for this level.

4.3 The usage of factors for regression

When the DHFM analysis is conducted a number of factors and idiosyncratic components are generated. Analysing these would already provide an interesting insight into competitive workings of the video games, especially since little research has been done in this sector. It is, however, not the aim of this thesis. Rather, we utilize these variables for a regression. Here, our dependent variable is the amount of players of a certain game genre in a week, the data of which can also be found on the Steamdb website. By using the factors as independent variable we can get a sense of how different pricing effects affect the number of players of a video game. For example, it seems plausible that by a market-wide pricing decrease (i.e. in a sale season) the amount of players for a certain game genre would increase. Alternatively, for indie games, which normally have a lower starting price, prices may be rather stable while experiencing a sharp increase in the amount of players. This may be due to the game catching tailwinds by the gaming community and/or the forthcoming networking effects. Hence, the relationship between indie game prices and the amount of players of these games may be quite different than other games.

The Twitch (streaming website for games) viewer count may serve as another independent variable as it proxies for potential players. On the other hand, it may result in endogeneity issues as it may be the result of players of the game looking for an alternative to consume the game, a phenomena referred to as ‘backseat gaming’. The influence of Twitch on the video gaming industry has been rather remarkable. From 2012 to 2018 the amount of average Twitch viewers has increased 1000% (TwitchTracker, 2019). Therefore, despite concerns of causality, it seems reasonable to consider this influence in our regression model. While the majority of our data comes from one source, the country-specific data needs to come from another one. The discrepancies in prices between countries needs to be accounted for and a major contributor of which, the exchange rates, will be considered. These will be retrieved from Investing.com, which provides a weekly dataset on exchange rates.

(18)

18 𝑌𝑡 = Α + ∑ 𝛽𝑖𝑗 𝑚 𝑗=1,..,𝑚 𝑌𝑡−𝑗+ 𝜀𝑡 𝑡 = 1, 2, … , 𝑇,

(19)

19

5. Data collection and manipulation

From the literature review and formed hypotheses, we have some conditions on the type of data that needs to be collected. Firstly, as a technical requirement, the data needs to be balanced. That is, the video games included need to have prices for the entire sample period. As games are not all released at the same time, we need to consider games that have been released for some period. Moreover, for the dynamic hierarchical factor model to work, the data utilized needs to be stationary. Furthermore, there needs to be a diverse selection when it comes to the genres of games in order to capture genre-level effects. Hence, each genre needs to be represented sufficiently. In addition, a distinction needs to be made between popular and less popular games. This quality of the data is required to assess the price effect of popularity, as found in DVDs, in video games. Lastly, to gauge the importance of country-level effects, country-specific data has to be appended to the dataset.

5.1 Data sources and sample selection

For our data we resort to multiple sources. Firstly, the source of our pricing data will be a third-party Steam database (steamdb.info), which is a tool developed to track the applications and packages that Steam has in its database. As the name suggests, the data tracked are only for Steam, a PC gaming platform. By choosing this source, we naturally exclude the price data for console games. Hence, a significant portion of total gaming market (25% if counted by revenue (Wijman, 2018)) will not be considered. Furthermore, the data for games offered on other PC gaming platforms (e.g. EA Origin, Blizzard, Amazon) are left out as well. Ideally, we would have included these data, however, due to the lack of accessible data sources these are impossible to fit with the Steamdb data. The database includes also the amount of players that every game has every day. Furthermore, the amount of viewers watching streamers play on Twitch is incorporated. Both serve as an indicator of engagement. The former takes on a more active version, whereas the latter takes on a more passive role. The amount of players will be used as a dependent variable on which the factors obtained from the DHFM will be regressed upon. Furthermore, to account for exchange rates between countries all prices will be converted to express them in USD prices using data from Investing.com.

There is, however, an additional complication for analysing the prices in the dataset. In the literature, we referred to the rise of free-to-play titles with in-game purchases. Within the list of the 98 most played games we find that thirteen of the titles utilize such a business model. As we are studying price dynamics over time and these games do not display any price fluctuations in the time dimension, they are not properly matched with the other data. One could argue that the DLC that can be purchased in game can be considered a substitute for the price, however, this method is met with several complications. Most importantly is the fact that we cannot assess the average player’s expenditure on DLC as this data is in a private company’s hands. The weekly dynamics of this statistic is therefore also impossible to assess. Consequently, the ‘prices’ for these games cannot be truthfully accounted for. As a result, the free-to-play games are left out of the dataset and we are left with 85 purchase-to-play games.

(20)

20 to analyse. The latter issue will be dealt with by opting for a ‘last observation carried forward’ procedure (LOCF), which fills the NAs in the data set with the value of the observation that is last observed. For example, if in week four the price of game x changes from 60 dollars to 40 dollars this would produce a data entry in week t. No other data entry transpires until a new price change occurs in the subsequent weeks t+n where n is unknown. The LOCF lets the prices in the subsequent weeks (i.e. NAs) take on the value of the price in week t. Lastly, some of the games did not have enough variation in its prices and therefore too little data entries for them to consider in the data analysis (e.g. Factorio only had three price changes in the entire sample period). The two games that possessed this little variation were omitted from the data set. After considering all of the above, a sample of 59 video games is left for the period from the 1st of January of 2017 until 31st of December 2018. For this period, all games in the sample possess complete data. It is recognized that there exists an asymmetry in terms of the release date of the different games, which could potentially have an effect on the pricing dynamics.

Then there is the issue of aggregation which we need to account for. The data extracted from the Steamdb website is data at the daily level. As daily level data is subject to heavy fluctuations this level of aggregation is deemed to be undesirable and will thus be aggregated to the week level. The characteristics of the pricing on the Steam platform are, however, on a discount basis, meaning that the prices are fairly constant over time alternated by periods of sharp discounts. It is sensible to assume that the effect of the discount is what drives purchasing behaviour of the customers. Hence, for this reason minimum prices of a game for each week are chosen to be the variable of interest. With regards to the amount of players and Twitch viewers in each week the mean value of each week is deemed to suffice. Furthermore, it was found that the price data did not possess enough variation necessary to run the DHFM algorithm. Therefore, random noise was added to the data such that the conditions of the algorithm were satisfied. This jitter has only been added for the DHFM analysis and the subsequent regression analyses and not for any of the summary statistics or visualisations in this paper.

5.2 Data selection 5.2.1 Country level

Next to the genres we need to select of number of countries and corresponding currencies in which the prices will be expressed. For our sample, we include the largest PC gaming markets as the price developments in these markets are the most relevant for all stakeholders. Hence, the following countries and their corresponding currencies are considered.

Country Market size in millions USD Currency

China 13,100 Chinese Renminbi (RMB)

United States 5,300 US Dollar (USD)

South Korea 2,340 South Korean Won (KRW)

Japan 1,800 Japanese Yen (JPY)

Germany* 1,400 Euro (EUR)

Table 3: Selection of countries and their market size. Source: VGSales (2019). *Given that the currencies of the countries will be used in the analysis, the market that represents the euro will larger than just Germany.

(21)

21 the price of video games, which are more interesting for our analysis. Below is a visual overview of how the different currencies have performed on a weekly basis vis-à-vis the US dollar over the sample period.

Figure 2: Weekly development of value of currencies of interest vis-a-vis the US dollar. Source: Investing.com

Over the sample period, different trends for the different currencies can be observed. For the euro, the year 2017 was a solid period in terms of the currency’s value as we see, despite some minor depreciations, an overall appreciation against the dollar. The peak value occurred in early 2018, after which the price of the euro started to steadily decline, albeit at a lesser rate than its appreciation in the preceding year. For the Japanese yen the pattern is rather different as it is subject to more volatility over the sample period, displayed by steeper slopes and more frequent fluctuations. Like for the euro, the USD strengthened for almost the entirety of 2018 against the Japanese currency, only showing a relative weakening in the last month. This might be due to the relative poor performance of the US stock markets, as a result of which capital was transferred to an investor ‘safe haven’ like Japan (Yahoo Finance, 2019). In line with this hypothesis, the pattern of the Korean won is again more in line with the one of the euro. It is marked by steady increases until the beginning of 2018, after which it started to decrease in value rather marginally. Lastly, the Chinese renminbi performed rather well until the third month of 2018, similar to all other currencies. The year 2018, however, China experienced the fastest depreciation of its currency relative to the others considered. Despite the pace at which the currency lost its value, the relative value decrease against the dollar does not differ substantially when compared to the other currencies.

(22)

22 video game prices. Therefore, to avoid this problem, the currencies are corrected for by using weekly exchange rate data. All local currency prices of a video game will be expressed as if they were in US dollars.

5.2.2 Genre level

As mentioned earlier in this chapter, our data is subject to a couple of conditions which need to be satisfied. Most of these can be satisfied by selecting the right sample of video games. We choose our sample to be the best selling games of 2016, as measured by gross revenue (Steam, 2019). All of these games have been released some time ago and thus the availability of price data will not be an issue. In this set of games, a fourway distinction has been made between the video games. Specifically, these categories are platina, gold, silver and bronze, consisting of eight, seven, eleven and 35 titles respectively. Thereby, we have found a qualitative proxy for the popularity. It should be noted that the games in our sample are quite heterogeneous when it comes to their characteristics. By the use of user tags the games are categorized according to a large variety of properties. Each game on the Steamdb possesses twenty of these tags. In order to shed some more light on the types of genres we have in our sample, we take the most important five tags from each game. These are manually put in an Excel file and dummy coded. Thus, if a characteristic for a game is in the top five tags, the title receives a one for that category. In the case a characteristic does not belong to the top five characteristics of a game, the value for this variable becomes zero. Applying this procedure to all game titles we end up with a data file consisting of 36 different user tags. The distribution of the prevalence of the user tags are illustrated in figure 1 below. From the figure it becomes evident that the user tags of video games are distributed in Pareto-like fashion. From the 59 games in the sample, 48 of them have the ‘Action’ tag as one of the top five user tags. Furthermore, the tag ‘Adventure’ is highly popular amongst the most revenue generating games, appearing in 41 of them. Both of these categories can be viewed as rather general and applicable to a wide variety of games. On the other side of the spectrum, however, we find more specific game characteristics such as the ‘Comedy’ and the ‘Trains’ user tags. To account for systemic genre price dynamics in the DHFM it is a necessary exercise to reduce the amount of dimensions of our current set of game characteristics. If not reduced, the observations for each genre could be insufficient to provide reliable coefficients. Therefore, a hierarchical cluster analysis is conducted in order to decrease the dimensionality of the different types of genres.

(23)

23 Firstly, a dissimilarity matrix is calculated from our dummy coded data set. Then Ward’s method (1963) is applied, which generates clusters by minimizing the variance within a group. From the cluster analysis we obtain the four distinctive groups of games. The full dendogram of the cluster analysis can be found in appendix A. These four groups have subsequently been named according to a manual count of the most prominent tags as well as a subjective assessment of their characteristics. Below the clusters, their size, their given names as well as a selection of the games which comprise the cluster can be found.

Cluster (size) Name Examples of games

1 (18) Action/AAA Watch Dogs 2, FarCry 4, Fallout 4, Dishonored 2, GTA 5, The Witcher 3

2 (17) RPG/Action/Adventure Starbound, Endless Legend, Undertale, Terraria, Grim Dawn, Stardew Valley

3 (13) Indie No Man’s Sky, Youtubers Life, Garry’s Mod, DayZ, Subnautica, Rust

4 (11) Simulation/Strategy Civilization VI, Civilization V, Payday 2, ARMA 3, Stellaris

Table 4: Clusters resulting from Ward's method.

It is found that the different groups of games are distinct to a satisfactory extent. There is still some room for improvement as some of the clusters overlap somewhat, however, given the inputs of five user tags per video game as well as the chosen set of clusters this problem is unsurmountable. Therefore, the identified clusters are assessed to be suitable and will serve as the genres in our genre-level of the hierarchy. The cluster names are assigned to the corresponding games as a means of providing them with an overarching genre, derived from a manual assessment of genres of the games contained in the cluster. The different clusters have distinguished between several sets of games from which we can expand our set of hypotheses. Firstly, the cluster containing the indie (= independent developer) games are bound to display different price dynamics than other non-indie games. Given that the game has no brand value and recognition going its way it seems reasonable to expect lower price monopsony power. Furthermore, a lower price could attract buyers. Lastly, given the developmental costs of the game, the traditional 60 dollar price tag is not suitable for these types of games. Given all of the aforementioned we form the following hypothesis.

H5: Indie video games will have a lower mean and higher variance.

Given our set of indie video games, however, we may expect different price dynamics. As we are assessing the most grossing video games of 2016, the demand for these games is relatively high. Therefore, in the case where an indie game is highly successful, its price may increase due to demand forces. Yet, we deem this force to be less significant. Moreover, prices of differentiated products rarely increase.

(24)

24

6. Data descriptives

As with any data set, it is a useful exercise to first examine the data set’s characteristics. Therefore, in this section, the data set will be described with visualisations and statistics. We investigate the development of the prices on the market wide level, the country level and the genre level as per our hierarchy defined in 4.1 and 4.2. From this, we get a better feeling for the data and are perhaps already able to identify some trends in the data.

6.1 Market wide level

The data in the final data set has been subject to a fair bit of manipulation as mentioned in the previous chapter. In the data set analysed, the prices of 59 games are assessed over a period of two years, corresponding to 106 weeks. For each game, the prices are recorded in euro (EUR), Japanese yen (JPY), Korean won (KRW), Chinese renminbi (RMB) and the United States dollar (USD), all of which are exchange rate corrected. When considering price movement of all of the games’ prices expressed in one currency together we observe, unsurprisingly, a high degree of correlation, as visualized in figure 4 . The data show that the RMB prices correlate the least with the other prices, obtaining the lowest correlation when compared to the JPY (=0.82). The highest correlation is between the EUR and USD prices reaching a correlation coefficient of almost 0.97. A potential explanation for this are the similarities between economic development and consumer preferences of the European and North-American continent. These findings indicate that the market wide trend is rather strong and seems to be the dictating factor for the games’ prices. Hence, it is expected that this factor will have a great impact on the final prices. The DHFM analysis will assess this more formally.

This preliminary finding provides some initial evidence of for our H3. The DHFM analysis will quantify this suspicion more formally and thus we will return to this point.

Figure 4: Correlation matrix of the country level prices.

6.2 Country level

(25)

25 looking at the development of prices and price differences over time for each of the currencies. What is noticeable from these plots (figure 5) is the extent to which there exists a cyclical sales season, as also found by Senior (2019). These sales seasons are uniform with regards to their timing for each of the currencies, indicating a market-wide trend as hypothesized by the hierarchy defined in 4.1. Thus, it can be inferred from the graphs below that a downwards pricing trend exists in the time dimension. This is not surprising given that the video games in the sample mature over time, thereby decreasing the novelty and the forthcoming value of the product. Additionally, the video games have to compete with newly released games. This force of competition also puts a downwards pressure on the prices of all video games.

Figure 5: Development of prices per currency.

(26)

26

Figure 6: Boxplots of the average prices and discount per country.

Again, these boxplots show the highly volatile pricing patterns on the Steam platform. The majority of the data points for prices and discounts are contained within a fairly small range, while on the other hand having significant, high ranging outliers. Furthermore, from the second boxplot it can be inferred that the discounts and premium are rather correlated over time. 6.3 Genre level

(27)

27 currencies (F= 0.26 <p <0.32). The timing of variance of these two genres, however, is somewhat dissimilar as displayed below in figure 7.

From the figures above, it can be observed that the genre level price deviations exhibit, like the country level, high levels of comovement. However, if one observes closely, it can be seen that there are also some asymmetric movements of prices between the genres. Specifically, for the different genres there are several periods in which the genres have opposite price movements spanning across multiple currencies. Most noteworthy are the differences between the shifts of

the third and fourth genre (i.e. ‘Indie’ and ‘Simulation/strategy’) and the first two genres (i.e. ‘Action/AAA’ and ‘RPG/Action/Adventure’). This hints at the fact that there is indeed some common dynamic for each different genre which influences the prices of the games contained within that particular genre. On the contrary, given the relatively small size of these two groups, the differences could be entirely driven by alterations in the prices of individual titles. In order to shed some more light on how the price differences between genres manifest themselves, the plots in figure 8 are presented.

(28)

28 Like before, the existence of comovement is ever evident, however, upon closer inspection the plots show also their fair share of differing movements in prices for each of the genres. The findings of highest discounts and premiums from week to week for the ‘Action/AAA’ is bolstered. The ‘Simulation/strategy’ genre, on the other shows the most divergent of trends. A reason for this could be that the games contained within this genre are infrequently released games and possess a great deal of depth in the gameplay, thereby demanding a heavy time investment of its players. Consequently, the life span of this type of game will be more extensive relative to titles in the alternative genres. As a result, these games compete to a lesser degree with other games in the market and are therefore able to weigh their individual pricing decisions more.

(29)

29 line of reasoning is mere speculative as no literature exists for the gaming industry. This finding provides support for the fifth hypothesis, which conjectures a lower mean and higher variance for ‘Indie’ genre games.

(30)

30

7. Dynamic hierarchical factor analysis

In this section the hierarchy as defined in sections 4.1 and 4.2 will be estimated. The method used is based on Moench et al. (2009). These authors applied the model to macroeconomic time series where they model for multiple common blocks and subblocks. Among the blocks modelled were the factors production, demand and housing. For each of these they imposed subblocks to make the hierarchy even deeper. For example, for the factor demand they considered retail sales, merchant wholesalers and domestic car retail sales. In turn, the model was used to reduce the dimensionality of the different time series by making them subject to the set hierarchy. Like Moench and his co-authors, we impose a subjective hierarchy that is based on literature and theoretical foundations, which have been discussed in the previous chapters. Our data is only comprised of different pricing data. The aim of the DHFM is to find the magnitude imposed commonalities for each of the blocks and see how they affect the pattern of variance of the weekly prices of the video games. Ultimately, we wish to extract the factors for each of the common blocks and use them for regression analysis explaining the amount of players for each game.

We start by briefly explaining the procedure used to estimate the DHFM, followed by the variance decomposition of the different factors as provided by the model. Lastly, we will use the model’s outcome to test for impulse responses.

7.1 Markov Chain Monte Carlo (MCMC)

The procedure used to estimate the dynamic hierarchical factor model is based on the Bayesian maximum likelihood method, specifically the Gibbs sampling using a Markov Chain Monte Carlo algorithm. By firstly imposing principal component analysis structure the coefficients are estimated via this mechanism. The coefficients estimated are used to predict the prices present in the data set and optimized by subsequently backwards smoothing a large number of times which improves the estimation of our coefficients. As a result, the coefficients are obtained that are most likely to fit the posterior distribution. These coefficients will then be used for further analyses such as the one in the following section.

7.2 Variance decomposition

7.2.1 Action/AAA title games (genre 1)

(31)

31

Figure 9: Variance decomposition of the Action/AAA genre

In the graph, the F refers to the market wide component, the G to the genre level component, the H to the country level component and the Z to the idiosyncratic component. Numbers one to five on the horizontal axis indicate for which country (i.e. currency) the decomposition holds. Firstly, it should be noted that each variance decomposition looks rather similar. Compared to the other groups (appendix C5), the market component has a relatively influence on the prices of this genre. On the contrary, the country level factors are a relatively important determinant of the prices. For each of the currencies, the majority of the variance of the prices is determined by the individual games’ idiosyncratic component, taking a share of approximately 40% for each currency (complete shares are found in appendix C). This indicates that the genre one games compete for a large part in monopolistic competition fashion with one another. Furthermore, it can be inferred that there are quite some deviations in the share of the country level. Particularly, the difference between the EUR and JPY stands out, the latter country level factor being twice as large as the one of the former. The interpretation of this finding is that the country level price movements of JPY weigh twice as heavy on Action/AAA games as they do in for EUR countries. This is suggestive of a different demand structure for each of the markets, making the JPY games more susceptible to general JPY market woes and relatively less to the global market. All the other factors do not differ that substantially from one another, implying that the Action/AAA genre price mechanisms are rather universal, not subject to either strong country level, genre level, or market forces.

7.2.2 RPG/Action/Adventure games (genre 2)

(32)

32 by large shares of the common market component and the idiosyncratic component (0.4507 and 0.4078 respectively). What this would indicate is that when it comes to the RPG/Action/Adventure games, the Japanese market prices follow the general market trend as well as the individual titles’ pricing strategy. A potential explanation for this could be that the genre two titles in our list are not in high demand in the Japanese market thus not allowing for significant pricing alterations. In support of this argument we mention the existence of the subgenre Japanese Role-Playing Game (JRPG). These games are RPG games made in Japan and mostly catered to Japanese players’ preferences, having a turn-based battle system and reliance on storytelling as opposed to the action-based battle system and reliance on imagination of players of the RPGs developed in the West (IGN, 2009). As a result, the other RPG games, like those on our list, are played less. In turn, for publishers there is less incentive to specify particular prices for the JPY region and thus make them follow the overall market trend. All other markets do not vary substantially from one another when it comes to their pricing of RPG/Action/Adventure games.

Figure 10: Variance decomposition of the RPG/Action/Adventure genre.

7.2.3 Indie games (genre 3)

(33)

33

significant need to stay price competitive.

Figure 11: Variance decomposition of the Indie genre.

7.2.4 Simulation/Strategy games (genre 4)

Lastly, visualised below, we have the variance decomposition of the Simulation/Strategy games. For each of the currencies, we find relatively little influence of the common market wide factor, just as for the Action/AAA titles. In fact, the Simulation/Strategy games are on average the least influenced by common market forces. This is in line with the characteristics of these games which can be considered ‘outsider’ titles due to the amount of freedom and control it provides to players. These characteristics generate heavy time investments into the in-depth gameplay by the players of these games. As a result, these titles’ lifetime is longer-lasting than other genres of games, which would allow for more stability against market trends. Moreover, in line with this argument, it is found that the idiosyncratic component of this group is the largest on average, indicating strong individual pricing power by each game. The remaining components have varying influences on the variance of the prices. For the EUR, we observe a relatively small country effect, suggesting little influence of regional price developments. The respective size of this component is similar to the USD region. On the other hand, the size of this component for the JPY and RMB are significantly larger than the EUR and USD region, 61% and 68% respectively. This hints at a different pattern of interest of the JPY and RMB regions relative to the EUR and USD regions, which could mean there is an incentive for price setters in these regions to adjust pricing for the former two.

(34)

34 From the variance decompositions there is evidence observed for differently levelled effects: market wide effects, genre effects and country effects. This is contrasting to the fourth hypothesis which states that the prices of different genres are set independently of the country they are sold in. Truthfully, there exist some country level effects but those are only partially responsible for the explanation of prices. Hence, we regard the fourth hypothesis as a partially rejected one. Furthermore, it should be mentioned that from these variance decomposition analyses it becomes evident that there is a market force having an effect on the games’ prices, thereby providing positive evidence for the H3. Thus, we find this hypothesis to be supported. In addition, regarding hypothesis one, we find partial evidence as well.

7.3 Common factors over time

From the DHFM analysis several factors are obtained for each of the hierarchy levels. Particularly, for the market wide level, three types of coefficients are generated, namely level, slope and curvature. For the genre and the country level, two types of coefficients are obtained: the level and the slope. All of these estimations are plotted over time in the graphs that follow. 7.3.1 Market wide block

(35)

35

Figure 13: Common dynamics ofmarket wide factors over time. Top panel: level factor. Middle panel: slope factor. Bottom panel: curvature factor.

7.3.2 Genre level block

For the genre level, we only obtain the factors for the level and the slope and therefore these are less information dense than the factors of the market wide level. As we have four different genres, the total amount of factors we extract are eight. Given the sake of brevity, we will mention only the most important findings for each of the forthcoming plots. Furthermore, it should be mentioned that the for the first two genre blocks the 95% confidence intervals are generated, whilst for the others genre block this is not the case.

7.3.2.1 Action/AAA games

(36)

36

Figure 14: Common dynamics of the Action/AAA genre factors. Top panel: level factor. Bottom panel: slope factor. 7.3.2.2 RPG/Action/Adventure games

For the second genre, a similar pattern as genre one is found when it comes to the level of prices. Each peak is correspondent to the sales periods on the Steam platform. When it comes to the slope of the genre we observe relatively little deviance from the mean value, with irregular peaks in between. This pattern holds until the end of the second year of observation where a sharp increase in the slope is observed. This indicates that from week to week the prices experienced a rapid surge in price.

Figure 15: Common dynamics of the RPG/Action/Adventure genre factors. Top panel: level factor. Bottom panel: slope factor.

7.3.2.3 Indie games

(37)

37 of this third genre moves up to one standard deviation of its mean slope. The slope of the Indie genre is quite different than the RPG/Action/Adventure games. Given both curves, the Indie games experience more often rapid changes in prices. This is inferred from the amount of peaks for the Indie genre.

Figure 16:Common dynamics of the Indie genre factors. Left panel: level factor. Right panel: slope factor. 7.3.2.4 Simulation/Strategy games

Upon inspecting the level factor of the fourth genre of games it seems that this genre is the most volatile of all of the genres. However, we would like to refer to the scale of the y-axis which inflates the degree of volatility of the price level. When this fact is considered the Simulation/Strategy are rather stable in prices with here and there some sharp decreases in times of sales seasons. The slope of this genre of games also seems bound to a high level of substantial volatility, however, this is again due to the y-axis. We note here that the slope of the Simulation/Strategy games does alter often, suggesting varying levels of steepness of price increases and decreases. Given the pattern displayed in the second graph, these changes succeed each other rapidly.

(38)

38 7.3.3 Country level block

Given the sheer number of blocks that have factors for both the level and the slope (total of 40), we restrict our analysis in this section to only two genre blocks, meaning that ten country levels will be analysed (five for each genre).

7.3.3.1 Genre 1, Action/AAA: subblocks of each country

In the figure below, we observe the level factors (left column) and slope factors (right column) for the different countries within the block of Action/AAA games. From top to bottom we find the factors EUR, JPY, KRW, RMB and USD country. Firstly, what stands out is the difference between both the level and the slope factors of the different countries in the first genre, signalling a substantial amount of asymmetry.

Figure 18: Common dynamics of the country level factors for Action/AAA games. From top to bottom: EUR, JPY, KRW, RMB, USD. Left column: level factor. Right column: slope factor.

(39)

39 preceding years (World Bank, 2019). Furthermore, we find that, like the other countries it is marked by discount seasons, however, the RMB country level factor seems to be the inverse of the other regions. Such a pattern is remarkable and may be possible due to the sheer market size of the China, allowing price setters to adjust prices for this region. If true, this finding would put further pressure on rejecting the H4.

Then, next to the level common country factors, we find the slope factor in the right column of the figure. Here, again, we find noticeable divergence between the different countries for the genre one games. For the EUR, we observe a rather stable pattern for the slopes with one major negative spike, indicative of a sharp negative change in prices for this region. In line with the EUR slope factors are those of the KRW, which are displaying a similar pattern. The KRW slope factors seem, however, less volatile, meaning that the changes in short-term prices in genre one are less extreme than in the EUR region. For the remaining countries we observe a bit more distinctive patterns of the slope factors. For JPY a volatile pattern of sizeable slope factors can be deduced, symbolic of quickly succeeding price changes. We also note here that in this region there is a tendency of the slope factors to move increasingly upwards over time, suggestive of the presence of more prices increases. For the RMB region, a volatile pattern is illustrated. The size of these peaks are, however, relatively little, especially for positive values, providing evidence for little very steep short term price increases. Near the end of the sample period, the size of the size of the slope becomes larger approximating a standard deviation from its mean level. In the final row we find the USD factor slopes for the first genre which display a stable pattern with little divergence from their mean level. Like for the RMB region, the USD’ slope factors increase in size over time. What this could signal, and this also holds for the RMB case, is that the games have matured as products. Hence a way in which sales could be boosted is by having sharp price decreases to instigate consumers to buy the video game.

7.3.3.2 Genre 2, RPG/Action/Adventure: subblocks of each country

Presented below are the level and slope factors of the second genre for the different countries in our data set. Like in the previous section, we will discuss the most striking patterns and their implications.

With regard to the country level factors of the RPG/Action/Adventure genre we find that, contrary to the previous genre, all countries share similar common dynamics. However, there are some minor differences between the different countries. Most notable is the distinctive steepness of the graphs. The EUR and the USD experience, like the games in genre one, the largest, negative factors for the second genre of games. Provided that the price setters of the games of both genres are rational this finding suggests that there is an incentive to offer steep discounts. Such an incentive could be the relative price sensitivity of consumers. Kübler et al. (2018) investigated this issue and showed that quite a number of European countries tend to be rather price sensitive. In the same paper, it is found that Japan and particularly China tend to be relatively price insensitive, which would provide an explanation for our the pricing pattern shown in figure 19. Unfortunately, the South Korean market was not included in their sample. Generally, the findings from these authors are in line with our data here.

(40)

40 sharp increases in prices of genre two in the EUR area. All other countries exhibit mostly similar country level common dynamics, although varying in intensity. For the negative slope factor values the USD region obtains the largest deviations, indicating rapid changes in prices in the short-term. On the contrary, the KRW region shows little deviations from the mean when it comes to the slope of the common country dynamic. JPY and RMB are somewhat in between the last two regions. At the end of the sample period, each region experiences its biggest slope factor, signalling a large change in the prices of the games of genre two for each country.

Figure 19: Common dynamics of the country level factors for RPG/Action/Adventure games. From top to bottom: EUR, JPY, KRW, RMB, USD. Left column: level factor. Right column: slope factor.

(41)

(42)

42

8. Vector autoregressive analysis

The ultimate aim of this paper is to explain the number of players of each game by exploiting the different price effects present in the video game market. The latter part has been achieved via the DHFM analysis and thus we continue with the former part. We start our regression analysis at the genre level. That is, we regress the generated factors on the number of players for the games in the first genre. We do this via a vector autoregressive model where we try to gauge the impact of shocks to a certain variable. The formal model is defined in 4.3.

Given the set of potential models and their interpretations, only the first two genres (i.e. Action/AAA and RPG/Action/Adventure) of games will be analysed for the amount of players. 8.1 Action/AAA games

In section 7.2 we quantified the composition of variance for each of the factors of the different countries. Here, we will relate those factors to the number of players of the games. This implies that for this genre all three market wide dynamics (i.e. level, slope and curvature) and the genre factors Gi1 and Gi2 (i.e. level and slope of the each genre) will be considered. Moreover, as we run multiple VAR analyses for each the countries (=5), the country level factors Hij1 and Hij2 will be used, where i stands for the genre, j for the country and the number indicate the level and slope factors respectively. In order to conduct this analysis, the data needs to satisfy the condition of stationarity, tested for by the Augmented Dickey-Fuller (ADF) test. Upon testing this condition, it was found that three variables did not satisfy this prerequisite for the genre of Action/AAA. Particularly, these are the mean average Twitch viewers and the country level factors H121 and H152. The outcomes of the ADF-tests can be found in Appendix D. The latter two non-stationary series were corrected for by taking first differences. The first one could not be resolved without making some serious concessions. Hence, it was decided to drop this variable for the VAR. Furthermore, we have to ensure that there exists no significant serial correlation in the VAR model in order to prevent inefficient estimation of the coefficients, under estimation of the error variance, under estimation of the regression coefficient variance as well as the forthcoming inaccurate confidence intervals (NCCS Statistical Software, 2019). The outcomes of these tests are also to be found in Appendix D.

To provide insights of how variables respond to one another, an impulse response analysis is conducted. In this type of analysis the response of a variable is measured as a result of a shock in another variable. We run the VAR analyses for each individual country and visualise the significant impulse responses. In particular we are interested in the number of players, so this will function as our response variable. The size of the shock is a positive estimated standard deviation of the disturbance term for the impulse. A full overview of the significant variables from these functions and relevant test statistics are outlined in Appendix E.

An investigation in the price determinants of video games and their effect on player bases

Faculty of Economics and Business

MSc Marketing Intelligence