- The document analyzes the tradeoff between exploration and exploitation in the video game publishing industry. It examines how the exploration index (ratio of new games to total games published) is affected by performance and external events like new console launches. - Regression analysis finds a negative correlation between performance and exploration index, indicating higher-performing publishers focus more on exploitation through sequels and licensed titles. There is also a weak negative correlation with external events like new console launches. - The results provide insights into how endogenous and exogenous factors influence publishers' strategies around balancing exploration of new ideas and exploitation of past successes.

1. 1
Balancing Bets and Losses:
An Empirical Study on Exploration and Exploitation in
the Video Game Industry
Federico Bertazzoni 1732569
Tudor Carstoiu 1408558
Simone Di Carlo 1529932
Andrea Muttoni 1408731
Bocconi University - Fundamentals of Innovation and Industrial Change - Group 20
The paper aims to provide an empirical insight into the tradeoff between exploration and
exploitation, in a decision making context, for publishers in the videogame industry. We consider
the top 10 publishers for 8 consecutive years (from 2003 to 2010) and ranked their success using
a proxy composed of the average review score of the published games every year. We consider
sequels and licensed titles to be exploitative, and new original games to be explorative. By
comparing the ratio of explorative titles over total titles published per year we obtain an
Exploration Index. We create a model and perform several regressions to test two hypotheses:
(1) how does performance affect the Exploration Index? and (2) how does an external event, a
new console launch, affect the Exploration Index? Our results suggest a negative correlation
between performance and Exploration Index indicating that as publishers perform better, they
tend to focus on exploitation and a further weak negative correlation between Exploration Index
and external events. Our data, although meticulously aggregated, proved to have several
limitations discussed in the paper. Nevertheless, our results offer valuable insights on the
delicate tradeoff of exploration and exploitation.
Acknowledgements
The authors would like to express their sincere gratitude to Hakan Ozalp, whose knowledge of the
video game industry and patient feedback were of invaluable help on our empirical journey.

2. 2
SECTION 1
Introduction
Often our decisions depend on a higher level choice: whether to exploit well
known but possibly suboptimal alternatives or to explore risky but potentially
more profitable ones.
(J.D. Cohen et al. - 2007)
Decision-making within firms, when launching a new product, has always been a matter of
trade-offs. Firms constantly face a choice between following a success path, risking to
become obsolete, or focusing on innovation and explore new unknown endroits. It is
therefore a matter of balancing bets and losses. Philosophically speaking, as Aristotle said,
the golden mean is the desirable middle between two extremes, one of excess and the other of
deficiency. It is therefore clear that a fine-tuning between the two is the appropriate path, but
little is known about how performance of the firm over time or exogenous events influence
this equilibrium.
Exploration relates to value creation, or increasing the knowledge base, pre-empting the
future, while exploitation is related to value capture, or reusing the existing knowledge,
mastering the present.
When firms excessively focus on exploration, there is the risk is that the costs of
experimentation come without benefits, ideas remain undeveloped and there is an overall lack
of distinctive competences, which come about through continuous learning by doing along a
consistent trajectory.
Conversely, when most of the focus is on exploitation, the risks are that the firm gets locked-
in to suboptimal equilibriums (local maxima) and cannot adapt to changing circumstances.
This can create path dependence and core capabilities can eventually become core rigidities
(J. March – 1991). When a business finds the right balance between exploration and
exploitation then this business can be defined as ambidextrous (Duncan 1976, March 1991).
The aim of this paper is to empirically model the effects of this tradeoff in the video game
industry. Year after year, video game publishers face the dilemma of exploring by creating
new original titles or exploiting through sequels and licensed titles based on past successes.
The video game industry has enormously grown over the years and this path seems to be
confirmed by current market forecasts. In 2012 the industry was worth almost $80 billion,
combining software, gaming revenue and devices, according to DFC Intelligence. A special
report by the Entertainment Software Association (ESA) stated that the U.S. interactive
entertainment software industry is, and has been, one of the most rapidly growing industries
in the United States. From 2005 through 2009, the computer and video game industry

3. 3
achieved real annual growth of 10.6% per year. By comparison, the entire U.S. economy
grew by only 1.4% per year during the same four-year period.
Since the early days the market composition has been divided among console producers,
game publishers and developers. As a consequence the market has been characterized by
vertical integrations in order to acquire greater market share.
In our research in the video game publishing industry, we defined a game as explorative if it
was new and original at the date of release meaning it had no prior sequels and was not based
on licensed content. On the contrary, a game is considered exploitative if it is a sequel or
licensed title, in other words building upon existing success. We define sequel as a new
version of an already existing game, and licensed a game takes the story or the name or both
of some famous external content.
After defining what exploration and exploitation means in the video game context, we
considered the development and marketing costs of an explorative versus an exploitative
game. Our intuition led us to believe that an explorative game should have both higher
development and marketing costs because it is something completely new and therefore
requires more resources. Consequently, when sequels are launched, we assumed lower
development and marketing costs because they were incremental improvements on existing
titles and notoriety. Our research proved exactly the opposite of our preliminary intuitions.
The next extract gives us a hint about how things work in the industry: "In an interview, in
May 2009, Sebastien Puel stated that the development team working on Assassin's Creed II
had increased to 450 members, and the development team's size had tripled since the first
game." (Kotaku - 2009)
The paper is structured as follows: section 2 states the hypotheses; section 3 explains our
methodology; section 4 illustrates our variables and our model; section 5 sums up our results
and conclusions.

4. 4
SECTION 2
Hypotheses
In this section we briefly outline the hypotheses that underlie our analysis. We are used to
considering videogames as a form of leisure, not a company’s product with revenues and
costs. However, if we try to look at the videogame industry from an economic point of view,
we can suppose that the type of game made by a company is linked to some endogenous and
exogenous factors. In testing for this interaction we examine the interrelationship between the
publishing strategy and its performance as a proxy of endogenous elements and between the
publishing strategy and the launch of a new console.
Our first null hypotheses is that high levels of performance has no effect on the Exploration
Index, that is, there is no relationship between the publishing strategy and the average ratings
received by critics.
The first alternative is that when performance is high, the Exploration Index is greater, as
firms feel safer and have greater economic security to risk publishing completely new games.
A second alternative is that the publisher, being risk averse, with high levels of performance,
exploits successful games because of path dependence and the value trap.
Our second null hypotheses postulate that new console launches have no effect on
Exploration Index, that is, the endogenous factors doesn’t influence the choice of firms
publishing strategy. Arguably for video-game industry most important product market
outcomes is the time it takes to bring a game on new consoles, in order to have first mover
advantage; the second null hypotheses would test that the new console launch does not effect
on the type of game published by companies. Again there is at least two alternative
hypotheses.
The first alternative is that there is a positive correlation between a new console launch and
exploration because a new generation console may stimulate novelty.
Alternatively there could be a negative correlation between external events and Exploration
Index due to conversion of same game to the new console format or due to the fact that firms
hedge their new console launch, an inherently explorative act, with “safe” exploitative titles.

5. 5
SECTION 3
Methodology
Our data was collected from a user-contributed website called Mobygames1
. Each game has
detailed information including title, release date (split geographically), average ratings,
description, genre, and platforms. Considering Mobygames could be seen as the Wikipedia
equivalent for video games, although we cannot rule out human error, the vast majority of the
information is correct and verified by multiple users (including the site admins). We cross-
referenced each data with other internet sources such as Wikipedia and official websites of
publishers.
We considered explorative any title that was new to the market, with no prior prequel/sequel
and no licensed content attached to it. Consequently, anything else we marked as exploitative:
successive titles such as prequels and sequels and licensed titles.
Our measure of performance was an average of the individual game ratings. While it would
be ideal to have precise financial data for each game, publishers do not subdivide balance
sheets and other financial data according to each title, making it very difficult to extract
accurate financial data for each game if not for the most popular titles. We think that an
average of the individual game ratings for a given year acts as a worthy proxy for financial
success.
Our game database contains over 3200 titles over 8 years and 10 publishers. Out of these, due
to uncertainties, redundant titles (e.g. “Deluxe Edition” released on the same day), and our
personal platform restrictions, our filtered total is 1564 titles. The most time consuming part
of the data collection was manually browsing the website extracting individual fields for each
game. We started with this method for around half the titles, but having spent an excessive
amount of hours, we decided as a group to find a more efficient method; both to manage time
constraints and preserve data accuracy. We therefore automated this first phase of the data
collection (extracting data from the website) via a custom programmed web crawler
programmed ad hoc. The crawler, based on specific filters (company, year) extracted all titles
from the website. This not only allowed us to save incredible amounts of time (what used to
take us 10 hours, now took 10 minutes) but most importantly eliminated any human error in
the extraction of the data.
We then categorized each title as either explorative or exploitative. As it required semantic
categorization, it was done by hand to avoid errors. Although it required manual verification,
it proved to be not as time consuming and could be done in parallel more efficiently.
1
http.//www.mobygames.com

6. 6
SECTION 4
THE VARIABLES
In what follows, we describe the main variables and the way they are defined and collated.
Table 1 shows the descriptive statistics.
AVERAGE RATING is a crucial variable because it is our proxy that represents the
publisher performance. We faced the possibility to consider other variables such as game
revenues but this one is the most significant and was the most consistent to find. We extract
the average rating for each game taken from Mobygames and Metacritic. For each firm for
the eight years taken in consideration we calculate the mean of game reviews. Eighty
observations compose our panel data, 8 years per 10 publishers, the rank is a number between
0 and 100.
EXPLORATIVE INDEX is a dummy variable that takes the value 1 if the game is in the
license/sequel category or 0 if the game is completely original. Also here we have eighty
observations. This is an important strategic variable as we are interested in measuring
exploration and exploration in firms publishing strategy. There is no easy way to obtain this
data, hence this variables reserve some discussion due to some uncertainty in classification.
EXPLORATIVE INDEX EXCLUSIVE is another variable that we use to catalogue each
games in explorative or exploitative category. However the main difference is that games
considered here are exclusive to a specific console. It is a dummy variable that takes the
value 1 if the game is in the license/sequel category or 0 if the game is completely original as
well. We use this new variable to investigate the relationship between new console launch
and firms publishing strategy. Some firms sign an exclusive license agreement to publish
game only for one console and this behavior could amplify the difference in balancing
exploration and exploration due to exogenous factors.
Going in depth, we consider both the Exploration Index for all games and for exclusive
games. Both of these are expressed by a dummy variable. We can notice the maximum value
is 0.258. This indicates that on average one out of four games published by a Top 10
Publisher is either licensed or a sequel.
If we compare the average of Exploration Index and exclusive Exploration Index, we find
exclusive Exploration Index is almost 3% larger: games published exclusively for a console
are on average 3% more explorative. It is a small difference, but was an interesting
springboard to investigate exploration and exploitation in the video game industry.
Table 1
VARIABLE OBSERVATION MEAN STD DEV MIN MAX
Avg rating 80 72.48588 4.896883 60.46 86
E-index 80 0.2376875 0.1794876 0 1
E-index
Exclusive
76 0.2584053 0.2225663 0 1

7. 7
Variable Correlations
Correlating average rating and Exploration Index (80 observations), we can observe a slightly
negative correlation between average rating and Exploration Index.
Avg rating Expr index
Avg rating 1.0000
Expr index - 0.1092 1.0000
Correlating Average Rating Lag 1 and Exploration Index (79 observations), we find a
positive correlation between average rating and Exploration Index.
Avg rating lag 1 Expr index
Avg rating lag 1 1.0000
Expr index 0.1635 1.0000
Correlating Exploration Index of exclusive games with console launches (76 observations),
we find a positive correlation between average rating and Exploration Index.
Avg rating Expr index
Avg rating 1.00000
Expr index 0.1979 1.00000
Model
We constructed the following model to run our regression:
where:
Y= Exploration Index
i = Publisher
t = time from 2003 to 2010
α = constant
x changes based on our regression. In our first regression x is average rating in t-1, in our
second regression it is average rating at time t, and in our third regression it represents a new
console launch.

8. 8
Does Performance Affect the Exploration Index?
First Regression: Exploration Index vs. Average Rating at time t-1 (Lag 1)
VARIABLES Constant Average Rating at t-1
Exploration Index
-0.0657323
(0.845)
0.0041921
(0.365)
In this first regression we want to test if the Exploration Index of year X is correlated with the
average score of the games published the previous year (x-1). The aim of this regression is to
see whether ratings affect their publishing strategy (explorative vs. exploitative).
We find this regression to be insignificant as shown by our p-value, thus not rejecting our
first null hypothesis. We are confident with our premise and analysis but there is a high
chance of heteroskedasticity or endogeneity in the regression due to the limitations of our
dataset.
Heteroskedasticity happens when the assumption of constant variance for the error term is not
fulfilled. This drawback does not lead to biased beta coefficient, but to biased variances.
Therefore the t-tests are not reliable. This kind of problem is very common in cross-sectional
dataset like ours. By using the Breusch-Pagan test we find that the variable Average Rating is
heteroskedastic. We then used an ordinary least square regression with a White correction.
The beta remained statistically equal to zero.
Second Regression: Exploration Index vs. Average Rating at time t
VARIABLES Constant Average Rating at time t
Exploration Index
1.0064
(0.004)
-0.0106
(0.028)
We decided to complement the first regression with a second one regressed at time t instead
of time t-1. We found a p-value lower than 5% meaning that the null hypothesis can be
rejected: average rating has an affect on the Exploration Index. As firms perform better, they
tend to decrease their levels of exploration and focus on exploitation. In practice this means
publishing licensed or sequel types of games.
To support our finding, we analyze the position of the market leader in terms of average
performance: Electronics Arts. During the period we see how a higher performance with
respect to the industry average and competitors, corresponds with relatively low Exploration
Index.

9. 9
On the other hand we can deduce that when a firm shows bad performance it is easier that
firm rise up the risk publishing new original game. This view is confirms Henrich Greve’s
theory that as firms perform worse than their historical average or compared to competitors, it
takes on more risk.
We clearly see this in the case of THQ, who coincidentally went out of business in 2010.

10. 10
Does A New Console Launch Influence Publishing Strategy ?
Third regression: Exploration Index of Exclusive Games vs. New Consoles
VARIABLES Constant Console
Exploration Index of
exclusive titles
0.2124547
(0.000)
-0.0919011
(0.060)
In this regression we want to explore the firm’s behavior related to the new console launch.
We took into consideration the previous 3 years relative to the console launch. This new
index however takes in consideration only the titles published exclusively for a console such
as PlayStation or Xbox. We consider exclusive games so as to strengthen the effective
relation between new console launches and the explorative strategy of companies. We note
how some of the publishers like Sony and Microsoft are both publishers and console
producers.
In this regression the p-value is slightly above the typical 5% threshold of statistical
significance. If we raise the tolerance threshold slightly, we find the correlation is negative.
In order to compensate for the relatively weak p-value, we performed a Breusch-Pagan test
and found there is homoskedasticity between the variables.
This could be due to the fact that publishers prefer not to expose themselves excessively, in
order to better exploit the network effects of previous successful games in a new generation
console launch. Exploitative titles are cash cows, and a safer bet. Since new consoles have an
inherent explorative component, we could reason that publishers prefer to hedge this form of
exploration with a greater amount of exploitative titles.
Another contrasting view could be that, as we approach the launch a of a new console,
publishers divest their explorative resources of original titles for new-generation consoles,
releasing relatively more cash cows in the previous generation of consoles.
Robustness of our data
We next discuss some potential limitations of the data and a number of robustness checks we
performed to deal with these concerns. A common issue with the use of survey data is the use
of retrospective data and the bias such data can impose. Heteroskedasticity is the first
problem we faced, it happens when the assumption of constant variance for the error term is
not fulfilled. This drawback does not lead to biased beta coefficient, but to biased variances.
We use the Breusch-Pagan test to check if heteroskedasticity happens and adjust the mistake.

11. 11
Then we face endogeneity in the regression due to measurement error, omitted variables and
possible simultaneity.
A classical hypothesis of the linear regression is that the error term should be orthogonal to
the regressor matrix:
E(X'
ε) = 0
This is a fundamental hypothesis for the validity of betas. If it was violated, the only way to
resolve it is to use instrumental variables. This can be tested when you already have an
instrumental variable regression.
The first cause of endogeneity could be omitted variable distortion. In order to avoid this
problem we should add the omitted variable in the regression, or use an instrumental variable.
In our case the Exploration Index is not sufficiently explained just by the rating, but other
factors, both endogenous and exogenous. Endogenous factors like: change in key employees,
game-by-game revenues, R&D costs. Exogenous factors could be: market trends in
preference types of games, geographical differences, external events. All of these factors are
very difficult to capture.
The second cause of endogeneity is measurement error. This means that the variables are
measured in a wrong way, thus the dataset is distorted. In our case, for example, the
performance and the rating could be distorted by the rating system applied. We opted for an
average measure of various rating sources using Metacritic and Mobygames. While this result
gives an accurate overall objective average of each game’s score, but in turn could dull the
differences in rating.
The last source of endogeneity is the simultaneity. This means that in our regression, not only
the average rating influences the Exploration Index, but also vice versa. To solve this bi-
directional effect we would need some instrumental variables that control for factors that only
one of the factors could explain.

12. 12
SECTION 5
Discussion & Conclusion
In general, the industry is very concentrated. Our top 10 publishers account for the vast
majority of the total games produced every year. We notice that the video game industry is
generally predominantly exploitative and it is the exploitative titles that benefit from the
largest amounts of resources, developer teams and marketing expenditures, contrary to the
intuitive belief that it is explorative games that garner the greatest amounts of resources.
Titles based on licensed content are also very popular and make for a significant portion of
the exploitative titles.
In the context of the resource-based view of the firm, we can see how capabilities and their
inherent tacit and specialized component reinforce the path dependence on exploitative titles.
In other words, making a game requires specific capabilities, largely composed of skills
acquired through learning by doing. The accumulation of these capabilities is expensive and
specialized. In making an exploitative sequel, the spillover of capabilities is almost perfect.
Most exploitative games only marginally innovate on secondary features and complementary
assets rather than core gaming mechanics. In making a completely different explorative title
however, these capabilities have a potentially very low spillover, as the competences required
to make a first-person shooter are very different from the ones required in a turn-based puzzle
game. Both core assets (game engine, developers) and complementary assets (art, music,
marketing) tend to specialize over time.
We found, through our first and second regression, that there is a negative correlation
between the average performance of a publisher and its explorative index. We find support of
this result by phenomena like the value network trap and in general the concept of path
dependence. Positive feedback (in our case positive reviews of the games) from customers
make it less attractive for firms to embark in exploration and risk financial returns and
reputation on new titles. Publishers like EA, who coincidently is a consistent top performer,
has the lowest Exploration Index out of all the publishers, focusing almost exclusively on
licensed titles, most of which are licensed sequels. Further evidence for these results can be
found in the latest releases of triple-A blockbuster games: GTA5 and Call of Duty: Ghosts.
Both of these recently released titles are exploitative and a sequel to a distant original concept,
and both broke landmark sales records in the first days of release totaling over a billion
dollars in revenue. Given the huge success of these exploitative titles, it is easy to see why
publishers prefer to divert their largest portion of resources to safe, cash-generating
exploitative endeavors.
Our third regression suggests that new console launches are slightly, albeit with weak
significance, negatively correlated with the Exploration Index. We saw how this could be

13. 13
dictated by two contrasting views. The first view is in favor of the negative correlation
whereby, given that a significant portions of our top 10 publishers are also console producers,
they hedge their explorative risk of a new generation console launch with more exploitative
titles. The contrasting view, which goes against the found correlation is that near the end of a
console generation, the amount of explorative games are less, as publishers focus their
explorative resources in developing games for the new console launch. A reason our
regression did not measure this effect could be that explorative games require longer
development time and would have made an appearance some time after the launch of a new
console. A reason for this could be that publishers prefer to wait until new consoles witness a
wider diffusion and not only appeal to niche users or early adopters.
In conclusion, we found several consistencies with the existing body of literature, for
example Henrich Greve’s theory of increased exploration as a reaction to negative
performance, but we hope our findings trigger further research. Our hypotheses should be
tested with different proxies, not just average game rating, and optimally take into
consideration other endogenous and exogenous factors mentioned in Section 4. A possibly
more accurate proxy could be revenues, or even better a monthly game-by-game expenditure
in development and marketing. Considering the powerful dynamics of this high growth
industry, and players’ constant dilemma of exploring uncharted territories or settling in
proven lands, we believe that deeper and broader inquiries will expose surprising new
insights and a lot of new lessons could be learned.

14. 14
References
Ashcraft, B., Assassin's Creed 2's Team Triples. Kotaku. 2009-05-19
Esa, Entertainment software association, Video Games in the 21st century: the 2010 report
Esa, Entertainment software association, Essential Facts about the video game industry, 2013
Duncan, R.B. ( 1976) ‘The Ambidextrous Organization: Designing Dual Structures for
Innovation’, in R. H. Kilmann, L. R. Pondy and D. P. Slevin (eds) The Management of
Organization Design. Volume I:
Strategies and Implementation , pp. 167-88. New York: Elsevier North-Holland.
Eggers, J. P. (2011). “All Experience is Not Created Equal: Learning, Adapting and Focusing
in Product Portofolio Management”, Strategic Management Journal.
Greve, Henrich R. 1998. "Performance, aspirations, and risky organizational change."
Administrative Science Quarterly 44(March):58-86.
Greve, Henrich R. 2002. "Sticky aspirations: Organizational time perspective and
competitiveness." Organization Science 13(1):1-17.
Greve, Henrich R. 2003a. "A behavioral theory of R&D expenditures and innovation:
Evidence from shipbuilding." Academy of Management Journal 46(6):685-702.
Greve, Henrich R. 2003b. "Investment and the behavioral theory of the firm: Evidence from
shipbuilding." Industrial and Corporate Change 12(5):1051-76.
Greve, Henrich R. 2007. "Exploration and exploitation in product innovation." Industrial and
Corporate Change 16(5):945-75.
Greve, Henrich R. 2008. "A behavioral theory of firm growth: Sequential attention to size
and performance goals." Academy of Management Journal 51(3):476-94.
March, J. Exploration and exploitation in organizational learning. Organizational science.
Volume 2, Issue 1, May 1991
Grohsjean, T. and Kretschmer, T. (2008). “Product Line Extension in Hypercompetitive
Environments: Evidence from the US Video Game Industry.” Discussion Papers in
Business Administration 2008, Munich
Grohsjean, T. et. all (2011). “Performance Feedback, Frm Resources, and Strategic Change”,
Working Paper No. 11-02.
Shilling, A., Technology Success and Failure in Winner-Take-All Markets: The Impact of
Learning Orientation, Timing, and Network Externalities, ACAD MANAGE J April 1, 2002
45:2 387-398;
Tschang, Ted F., Balancing the Tensions Between Rationalization and Creativity in the
Video Games Industry, Organization Science, v.18 n.6, p.989-1005, November 2007
Tushman, O’Reilly. “The Ambidextrous Organization”, HBS, http://hbr.org/2004/04/the-
ambidextrous-organization/ar/pr
Verona, G. A resource-based view of product development, The Academy of Management
Review; january 1999