APPLICATION
OF MULTI-SIDED PLATFORMS: GOOGLE MARKET POWER IN INTERNET ADVERTISING INDUSTRY
Darko Milosevic1
Jane Paunkovic2
1Università LUM Jean Monnet, Casamassima, Italy
email:darkomi. de@gmail. com
2Faculty for Management, Zajecar, Serbia email: jane. paunkovic@fmz. edu. rs
ABSTRACT
The main objective of paper
is to analyze in an empirical way the
economic theory behind multi-sided platforms of Mobile advertising industry. In order to analyze the industry in a
comprehensive way, the study establishes an empirical model of monopoly that
proves that certain negative effects can lead to a reduction in the level of
innovation. The conclusion of the model show that
there is a positive relation between Google’s
shares and the firms shares that operate in the high-tech sector.
KEYWORDS
Multi-sided platforms, Internet Advertising industry, Market
Power, Mobile search engines.
APSTRAKT
Glavna
tema ovog rada je istraživanje industrije mobilnog oglašavanja. Cilj istraživanja
je da analizira ekonomsku teoriju korišćenja „multi-sided“ platformi mobilne oglasne
industrije. Da bi se na celovit način analizirala industrija, u studiji utvrđen
je empirijski model monopola koji dokazuje da određeni negativni efekti mogu
dovesti do smanjenja stepena inovacija. Zaključak modela pokazuje da postoji
pozitivna povezanost između akcija kompanije Google i udela kompanija koje
posluju u visoko-tehnološkom sektoru.
KLJUČNE REČI
Višestruke
platforme, Internet industrija reklamiranja, tržišna snaga, mobilni internet pretraživači.1. INTRODUCTION
Evans
(2013) define two-sided market as interaction between operators by reducing
transaction costs and creating a greater value. In a market with such a high
degree of change companies not just grab a large market share to get some
market power but they continually develop new inventions to remain at the
forefront without being overtaken by competitors. In some cases it may
experience a phenomenon known as economies of scale: doubling the
number of lines of code and increasing more than double costs.
It is
possible therefore to affirm that the marginal cost is very high and in some
cases it can exceed ex post marginal costs distribution of the software itself,
which in the field of software is next to zero or even negative on at least one
side. Evans and Schmalensee (2013) identified two broad categories of network
economies: externalities of use (usage externality) and externalities
of belonging (membership externality). Externalities of use provide an increase
in value for both managers and business owners that from one side fit their strategic
goal, and for another give benefit to the end user who wants some kind of
service or product. Result must reached critical mass, because this create
certain limit to the number of multi-sided platforms that may operate in the
market, viewed as barriers to entry for
possible new competitors.
Rochet
and Tirole (2003) developed model of multi-sided platforms. They identify two
types of calculation in a multi-sided platforms considering asymmetry
of consumers cost: customer-defined "marquee" and customer loyalty
independent on prices. Since the possibility that consumer will change platform
is very low, for the loyalty these customers applied high prices.
2. BACKGROUND
Hoffman and Novak (2000) propose standard measurement constructs and point out potential problems in Internet advertising pricing models. Langheinrich et al. (1999) develop a linear programming model for nonintrusive targeting to increase click rates. Tomlin (2000) highlights potential problems with the linear programming approach and proposes a solution using traffic theory. Kohda and Endo (1996) propose an advertising agent, which selects ads based on consumers’ indicated preferences. Baudisch and Leopold (1997) also propose user configurable ads where the user indicates her interests.
Landes and Posner
(1981) accept the Lerner Index (1946) as the authoritative measure
of market power, and write that “If we knew the elasticity of demand facing
firm . . . , could measure its market power directly . . ., without troubling
ourselves about what its market share was. Price-cost margins may reflect
“superior skill, foresight and industry” that is the very result of
competition.
Starting from models
published recently by Lianos and Motchenkova (2013) has been developed a new
model that considers the effects of the monopoly in the field of search
advertising. For the purpose of modelling a search engine market we adapt a
modification of Armstrong (2006) multi-sided market model, where two sides pays
(advertisers and customers).
3. RESEARCH METHODS
Our approach to
modeling advertisers side of the market is compared to Edelman et al (2007),
Varian (2007), Ellison and Ellison (2004), Chen and He (2006), Athey and
Ellison (2011), or White (2008). The
economic models relating to Mobile multi-sided platforms
requires the presence of due or more economic agents and interaction with
common platform that connects between all parties by creating a higher value.
Two types of multi-sided platforms: physical and software platforms. The
physical platforms where markets served as a meeting point between the
merchants, whose purpose was to sell their wares, and buyers who provide a
wider choice concentrated in a limited space. More modern forms of physical
platforms, in ways similar to the search engine markets, are the online booking
hotels that make available beds and rooms using Mobile applications and services.
A key element of economic models is
represented by network economies or indirect network externalities, where
platform increases in value when increases the number of agents. To understand
fully the workings behind the models of multi-sided platforms it is crucial to
focus on the price structure because it explains how to maintain balance in
various contractors. It is assumed only the externalities of use and costs for
transaction and not for membership in the platform. In this model, the profit
of the platform is determined α in the following way:
(3.1)
and 
;
(3.2)
The first of the two conditions recalled the condition
of Lerne Index (1946) summarizing
the operation of the equilibrium price in a monopoly in relation to the
elasticity Ei of the application. The
second condition implies that there is a direct proportionality between the
demand and its elasticity.
The second model was developed by Armstrong (2006) and, analyzes different types of externalities and consequently also the costs analyzed are of different nature. Armstrong suggests a platform, in which externalities belonging to thus costs that are incurred by contracting for membership in the platform.
The
profit function elaborated by Armstrong is the following: The second model was developed by Armstrong (2006) and, analyzes different types of externalities and consequently also the costs analyzed are of different nature. Armstrong suggests a platform, in which externalities belonging to thus costs that are incurred by contracting for membership in the platform.
p = (P1
C1)
D1 (P1, Q2)
+ (P2 C2)
D2 (P2, Q1) (3.3)
C1)
D1 (P1, Q2)
+ (P2 C2)
D2 (P2, Q1) (3.3)
The
profit function analysis relationship between the demand of the group, the
price applied to the group and number of operators in the second group j by the expression: Di(Pi,Qj) con i = 1,2
and i ≠j.
In
order to properly solve this model to find the optimal solutions and to add a
further hypothesis we need to develop linearity of both demand functions.
According to the model developed by Armstrong it is:
, con i,j = 1,2 and i ≠ j (3.4)
In
this expression we introduced two additional elements: Ɛi that is the elasticity of demand Di with respect to price Pi taking constant Qj; θij is a term that indicates the positive impact of a
growth of Qi in
the application of the group j.
3.1
The Monopoly Model
The methodology is structured according
to theory approach and worked with Google
case study to create a holistic
comprehension of Market power. The iterative process consisted
literature review and (empirical) data analysis. We present here the
analysis for a Market power platform in order to focus on the possible
threat of abuse of dominant position by Google in the internet search engine
market. Motchenkova et al. (2013) in the more technical paper extend the model
to an asymmetric oligopoly setting, which allows to also analyze the interplay
between market dominance, network effects, and incentives to innovate in the
search engine market.
In the case of search
engines, in theory, there are two different prices pA and pU,
first applied to advertisers and second to users. The initial assumptions
about the price, in a formal way turn out to be: pA >
0 and
pU = 0. The main difference with the basic model developed by
Lianos and Motchenkova, for considering the incremental costs incurred by the
platform in that they consider only the marginal cost to attract another
advertiser in the platform, strictly positive as to serve new advertisers is
necessary to support the additional costs cA > 0.
The marginal costs instead of attracting new users are considered equal to
zero.
The
hypothesis is based on the idea that customers marquee are very valuable to the
platform to take advantage of the strong network economies that can develop due
to the increase of users on the platform. To make sure that there is a growing
number of users who use the platform we need to consider the marginal costs of users strictly
positive cU > 0. As we can see Google costs for R&D increasing over time, so model contains
company investments to innovate in the field. In this model F(k) is the function of innovation and it is an
increasing function k, the costs
incurred in order to innovate, the greater the improvement in quality, then F 0(k)
> 0. The utility advertisers function can
be formalized as:
uA(k,pA,nU)
= αAnU + k -pA (3.5)
where the value is
determined as a function of growing respect to innovation and to the number of
users and depends negatively on the price imposed on them. In the function αA is the benefit that
advertisers are derived from the interaction with users, or network
externalities, and in this case is strictly positive, because more are
users who will see advertisements the greater the utility derived for
advertisers.
The utility for users
can be determined in a similar way, component represented by negative price is
canceled because the initial hypothesis imposed that:
pU = 0: uU(k,nA)
= αUnA + k (3.6)
Consequently
it is possible to define the abundance of the two groups as increasing
functions of the large number of groups: nU = ϕU(uU); nA
=ϕA(uA); where ϕi,
with i = A, U, is an increasing
function in terms of utility, and it is possible to assume that the ϕi> 0.
Having formalized all
the essential elements of the model can determine the function of profit for
the monopoly company:
∏(k,pA)
= nA(pA cA) cUnU F(k)
cA) cUnU F(k)
The
profit function can be rewritten in terms of utility, operating certain changes
to the formulas defined previously. Starting from the equations 5.1 and 5.2, we
can rewrite the terms pA as functions ui, thus obtaining the following formulations:
k = uU
αUϕ(uA) (3.7)
αUϕ(uA) (3.7)
and pA = αAϕ(uu) + uU αUϕ(uA)
uA (3.8)
αUϕ(uA)
uA (3.8)
The payoff can be
formalized as:
∏(uU, uA)
= ϕ(uA)[αAϕ(uU) + uU αUϕ(uA) uA cA]
cUϕ (uU) F(uU
αUϕ(uA)) (3.9)
αUϕ(uA) uA cA] cUϕ (uU) F(uU
αUϕ(uA)) (3.9)
Assumptions of this model function of welfare considers the
surplus user group U(uU)
and
advertisers A(uA), essential for
comparing the price and efforts to innovations in the two situations uii = ϕi(ui) with i = A,U and ω(uU,
uA) = ∏(uU, uA)
+ uU(uU) + uA(uA) (3.10)
3.2 Maximizing
monopoly profit
Maximizing the
profit function of the company expressed as monopolistic utility function in
equation 3.11, we get the price charged to advertisers market equilibrium:
(3.11)
The price charged is a function of the marginal cost to attract
additional advertisers modified to some parameters that can be analyzed
separately:
·
αUαAϕ`(uU)nA represents the disutility that an extra group of
advertisers leads to the group of users, so the company needs to be calibrated
for the price especially considering this component. As defined above it it is
assumed that αU is negative and therefore this component is positive and
is added to the marginal cost of increasing its value, then pA > cA.
·
αUcUϕ`(uU)nA this part is the contribution of the marginal cost of the
users in the price of the final balance. If CU necessarily increase the price charged to advertisers
will increase, as the company should consider in its function even more money
to get back the costs, although incurred to attract members of the other group.
·
is the
elasticity of participation of the members of group A, which leads to increase in equilibrium price.
is the
elasticity of participation of the members of group A, which leads to increase in equilibrium price.
The
function innovation of profit in a market where there is a monopoly is instead
determined as follows:
(kM)
= αAϕi(uU)nA + nA 
cUϕ0(uU) (3.12)
3.3 Maximization of
total welfare
When
it maximizes the function of welfare 5.6, instead of solely that of profit, the
equilibrium solutions are slightly different, in favor of the members of the
two groups and not only of the platform.
In
this second scenario, the equilibrium price that advertisers are required to
pay is determined as:
(3.13)
The function of innovation instead is increased by a factor equal to nU. The equilibrium
result turns out to be:
F `(k*)
= αA 
(uU)nA + nA cU (uU) + nU (3.14)
(uU)nA + nA cU (uU) + nU (3.14) 4. DATA COLLECTION
For
the construction of model was chosen as the dependent variable the annual series on the shares issued by the
industry leader, Google Inc. The stock prices were
taken from series of the financial data NASDAQ Yahoo Finance, Statistica and Google for period
2005 – 2015. The minimum point corresponds to the first month of listing the
company with a value of 129.60 per share and then have an overall increasing
trend with a peak reached in February 2014 totaled 608.43. Although the overall
trend is positive. Apple Inc. mobile
search engine is the main competitor of Google,
and significantly contributed to the growth of high-tech industry in recent
time. Performance of the shares of Apple
is always growing. The minimum number amounted to 7.68 was reached in June
2006, while the peak of 126.59 was achieved in June 2015. The performance of
the shares of the Microsoft, during
the observed period, had same growing trend in the series. Minimum point had
been reached in 2009 with a value of 18.94 per share, while reached peak in
June 2015 is 46.09. Yahoo is the main
competitor of Google basic search
engine, and vertical online search advertising. Yahoo had two different periods of growth. First period of the
course of actions is decidedly negative and second period from last quarter of
2012 begins with slow recovery. Today Yahoo
value per share is 40.51.
The Nasdaq Industrial has been selected to
benchmark manufacturing sector, listed on stock exchange where Google and other companies was
previously analyzed. The performance of the actions of this title, are strictly
increasing. After the crisis period where value reaching its minimum of
1,021.35 in February 2009, from mid-2010 began the second half of the field
touching a peak of 2,595.40 in December 2012. Latest data available for analysis
is June 2015 with value of 4,141.20.
The last independent
variable represents Google's annual revenue related to online advertising.
The performance of Google's revenue
is growing, with the low point recorded in the first observation and the peak
for the last observation.
The first dummy variables, or binary variables
(take the value 0 or 1) is constructed
to be a negative. This variable
was not significant in any of the models built. The second variable concerns
the innovations
and the acquisitions made over the years. This could be caused by the
many similar operations carried out by Google
that impact stock prices.
5. RESULTS
To build an econometric model it is necessary that the
variables used stationary or the series analyzed must have a constant mean and variance. The first test
is used to test the unit root Augmented
Dickey-Fuller whose H0
implies non-stationarity of the series analyzed. All series are analyzed for
this test, where p-value is smaller than the critical α value 0.05.
In Appendix
Model 5.1 shows the output of the linear regression model built using the statistical software Gretl. Table summarizes the main
elements needed to consider the initial suitability of an econometric model.
With this model we try to explain the performance of the shares of Google, the dependent variable choice,
through different independent variables such as: the stock market price of Apple; the actions of Microsoft in mobile search industry; the
performance of securities of Yahoo; actions
delayed by three periods of Nasdaq
Industrial; Google's revenue, on
advertising type search site, delayed by 5 times. The first point note to
concerns is the significance of the independent variables included in the
model. For this we can examine coefficient
with t-statistics and p-value of each variable to be less
than the critical value α. A key element to understand the goodness of the
model is the R2 value, or
more appropriately adjusted R2.
This factor represents the ratio of the explained total variation of value
ranges between 0 and 1, with 1 being the construction of a model that can
explain perfectly the mechanism. In the model built in this chapter is the
value of adjusted R-squared of 0,97. The value is very high and adjusted
R-squared can be considered sufficient to
accept. Model perfectly correspond to the actual values of the dependent
series, in fact in some places there is overlap between the estimated and
actual values. From the output of the time
series plot we can check the value taken by the Durbin-Watson statistic that measures the autocorrelation of the
first order among the debris. The value of this statistic varies from 0 to 4,
in the analyzed model DW assumes a value close to 2 and this indicates that
there is in correlation between the error terms.
Verification
of the model. The basic assumptions
of the model can be summarized in the following list: linearity of the
parameters; regression must be deterministic; the error terms must be
uncorrelated; presence of homoscedasticity between failures; errors should be
normally distributed.
Normality
of residual. Fundamental
characteristic for is that the residues are normally distributed. To test hypothesis it is possible to use linear regression models and normality test. The hypothesis implies that
the distribution of errors according to normal.
Test
for hypothesis of normal distribution:
Chi-square(2) = 0,667 with p-value 0,71655. Test for normality of
d_l_Az_Google: Doornik-Hansen test = 8,9181, with p-value 0,0115734; Shapiro-Wilk
W = 0,77981, with p-value 0,00823477; Lilliefors test = 0,307589, with p-value
~= 0,01.
In this case, the p-value is higher than the critical
value α and so this leads to accept the null
hypothesis of normality of the residuals.
Figure 5.1: Test of normality
Frequency distribution for uhat2, obs 1-10 number of bins = 5, mean = -1,7053e-014, sd = 33,9625.
Test for null hypothesis of normal distribution: Chi-square(2) = 0,667 with p-value 0,71655.
Homoscedasticity. The second essential characteristic concerns the homoscedasticity,
or the presence of constant variance. To test this hypothesis, we can use the White's test, whose null hypothesis H0 provides for the absence
of heteroskedasticity. Test statistic: TR^2 = 6,358807, with
p-value = P(Chi-square(6) > 6,358807) = 0,384219
Autocorrelation. For the autocorrelation of the residues using the test
of Ljung-Box. Only to lag 1 there is
a first output -0,26 from
the down limit of the bars. This leads to affirm that the residues
are not autocorrelated.
Unadjusted R-squared = 0,448380; Test statistic: LMF =
0,812842, with p-value = P(F(3,3) > 0,812842) = 0,566; Ljung-Box Q' =
1,36757, with p-value = P(Chi-square(3) > 1,36757) = 0,713;
Figure 5.2: Residual autocorrelation function
The significance of the model. Power Model
shows how the performance of Google
shares is heavily influenced, in a positive way, by the performance of other competitors. This shows
that the sector in which the company operates is a very active, the more
businesses grow the more influence each other. In this way, industry
innovations affect not only a company but most of the firms involved, and this
creates a greater incentive for companies to innovate thanks to the strong
presence of positive externalities. The negative sign of the trend coefficient
is when the industry is in decline the high-tech industry appreciates. There is
a positive relationship between independent variable and total revenue of Google, indicating that an increase in
revenue as well as stock prices is improved. This feature is natural as
continuous increase in revenue positively affects stock prices.
Structural
Break. For verification optimum of an
econometric model we need to
determine if there is the presence of some structural breaks. Usually structural breaks occur in times of
deep financial crisis, which
adversely affects the stock prices. The
first test used to verify the presence of break points is the QLR test. Quandt likelihood ratio test for structural break at an
unknown point, with 15 percent trimming: The
maximum F(3, 2) = 87,1498 occurs at observation 2011; Asymptotic p-value =
1,332e-058 for chi-square(3) = 261,449.
According to the test the point at which is more reliable
signal of the structural a break is 2011. For the calculation of
structural breaks we used Chow test,
whose null hypothesis H0
regards the non-presence of break points. It is necessary that the p-value obtained from the test has a
value α higher than the critical point 10% previously indicated in the output
of the QLR test. Chow test for structural break at observation 2011; F(4, 2) = 0,85886
with p-value 0,6005; Chow test p-value 0,6005 < QLR test Asymptotic p-value
= 1,332e-058.
Figure 5.3: QLR and
corrolation-matrix
We can
see that the highest values of p-values
are found in 2011. In regard to this model, it has been verified the presence of break points.
6. CONCLUSION
The online
advertising industry turns out to be a very healthy market and growing. The
findings coincide fully with the market situation in which the company acts
only for their own profit without taking into account the situation of the
entire market, leading to a balance with higher prices and less innovation,
which is more static nature of the market. This serves to remind how the sector
is highly dynamic and that the speed of change depends very much on the introduced innovations. The model showed that in a monopoly market price and the innovations
of balance are different depending on whether you want to maximize your profit (the company is rational and only then look at
its profit) or you try to maximize the
welfare of society. The case of welfare maximization is
optimal for all parties as advertisers are required to pay a lower price and
thus can increase the number of advertisements that want to publish as 
. This is positive because even smaller
advertisers can approach the world of online advertising, because access prices
are lower. As for the innovations they appear to be higher in the case of
maximization of the welfare. This increase of innovations brings a greater
benefit for all parties involved as we introduce continuously new and better
services that may be useful to the entire sector. In conclusion from this model
it is possible to say that a monopoly situation is not optimal, at least it is
decided to maximize the function of welfare.
. This is positive because even smaller
advertisers can approach the world of online advertising, because access prices
are lower. As for the innovations they appear to be higher in the case of
maximization of the welfare. This increase of innovations brings a greater
benefit for all parties involved as we introduce continuously new and better
services that may be useful to the entire sector. In conclusion from this model
it is possible to say that a monopoly situation is not optimal, at least it is
decided to maximize the function of welfare.REFERENCES
Book
Evans, David S., and
Richard Schmalensee. The
antitrust analysis of multi-sided platform businesses. No. w18783. National
Bureau of Economic Research, 2013.
Gawer, Annabelle.
"Platforms, markets and innovation: An introduction." Chapters (2009).
Lerner, A.P., The Economics of Control. Principles of
Welfare Economics, The MacMillan Company, New York, 1946.
Journal
Armstrong, M., Competition
in two-sided markets, RAND Journal of Economics, Vol.37, No.3, Autumn 2006.
Landes, William M., and
Richard A. Posner. "Market power in antitrust cases."Harvard Law
Review (1981): 937-996.
Lianos, I. and
motchenkova, E., Market Dominance and
Quality of Search Results in the Search Engine Market, Journal of
Competition Law & Economics, April 2013.
Rochet,
Jean-Charles, and Jean Tirole. "Platform competition in two-sided
markets." Journal of the
European Economic Association (2003):
990-1029.
(p.24).
Links
1.
Google's annual advertising
revenue. http://www.statista.com/statistics/266249/advertising-revenue-of-google (01.05.2018)
2.
International Telecommunication Union. http://www.itu.int/en/ITU-D/Statistics/Documents/facts/ICTFactsFigures2015.pdf (01.05.2018)
APPENDIX
Data - Stock Market Trends: Google, Apple, Microsoft,
Yahoo, Ind. Nasdaq, and Google Revenue
Model 5.1: OLS, using observations 2006-2015 (T = 10)
Dependent
variable: d_l_Az_Google
Coefficient
|
Std.
Error
|
t-ratio
|
p-value
|
||
const
|
−173,615
|
40,2098
|
-4,3177
|
0,01247
|
**
|
d_l_Az_Apple
|
−0,828555
|
0,690303
|
-1,2003
|
0,29625
|
|
d_l_Az_Micros
|
16,5081
|
4,10124
|
4,0251
|
0,01580
|
**
|
d_l_Az_Yahoo
|
2,94385
|
2,17545
|
1,3532
|
0,24741
|
|
d_d_G_websi_5
|
2,20021
|
0,728377
|
3,0207
|
0,03914
|
**
|
d_l_Nq_Indu_3
|
−0,0208966
|
0,0389632
|
-0,5363
|
0,62018
|
Mean dependent var
|
329,5243
|
S.D. dependent var
|
143,5653
|
|
Sum squared resid
|
1960,666
|
S.E. of regression
|
22,13971
|
|
R-squared
|
0,989430
|
Adjusted R-squared
|
0,976218
|
|
F(5, 4)
|
74,88815
|
P-value(F)
|
0,000484
|
|
Log-likelihood
|
−40,58166
|
Akaike criterion
|
93,16332
|
|
Schwarz criterion
|
94,97883
|
Hannan-Quinn
|
91,17171
|
|
rho
|
−0,119683
|
Durbin-Watson
|
2,227468
|
No comments :
Post a Comment
Note: only a member of this blog may post a comment.