International Journal of Industrial Organization 31 (2013) 331–341
Contents lists available at SciVerse ScienceDirect
International Journal of Industrial Organization
journal homepage: www.elsevier.com/locate/ijio
Advertising expenditure and consumer prices☆
Ferdinand Rauch ⁎
University of Oxford and CEP (LSE), Brasenose College, Oxford OX1 4AJ, United Kingdom
a r t i c l e
i n f o
Article history:
Received 23 April 2012
Received in revised form 27 April 2013
Accepted 1 May 2013
Available online 12 May 2013
JEL classification:
H25
M37
R10
a b s t r a c t
This paper studies the effect of a change in the marginal costs of advertising on advertising expenditures of firms
and on consumer prices. I make use of a policy change in Austria, that involved an increase of the taxation of advertising in parts of the country, and a simultaneous decrease in other parts. I show that advertising expenditures
of firms move quickly in the opposite direction of the marginal costs of advertising. Consumer prices increase
with advertising in some industries, and decrease in others. This effect correlates with informational differences
in advertisements across industries.
© 2013 Elsevier B.V. All rights reserved.
Keywords:
Advertising
Advertising tax
Advertising and prices
1. Introduction
This paper studies the effect of a change of the marginal costs of
advertising on both advertising expenditures and consumer prices.
It makes use of a policy change in Austria that directly affected the
marginal costs of advertising, and consequently advertising expenditure. While previous works have estimated the impact of advertising
on consumer prices for certain goods, this is the first study to investigate the effect of advertising costs on consumer prices for all major
industries and representative data for equilibrium effects of an entire
economy. As I show below, advertising increases consumer prices in
some industries, and decreases them in others. This heterogeneous
effect correlates with the information content of advertisements
across industries.
There are at least three important reasons why advertising has
been of interest to economists: first, advertising has been debated at
length in the theoretical economic literature as it is closely tied to
the central topic information and search as well as entry barriers
and product quality. Throughout advertising has remained a controversial topic, with contradicting policy recommendations, see below.
☆ Thanks to Peter Egger, Klaus Gugler, Maarten Janssen, Alan Manning, Guy Michaels,
Martin Pesendorfer, Steve Pischke, Stephen Redding, Philipp Schmidt-Dengler, Daniel
Sturm, John Sutton, Chad Syverson, John Van Reenen, and two very helpful anonymous
referees and participants at the NOeG, EEA, SAEe conference as well as the participants
at seminars at Bocconi, CEMFI, the HEC business school, the LSE, Oxford, Royal
Holloway College, Warwick, the Universities of Glasgow, Munich and Vienna for their
comments, and thanks to ESRC for financial support.
⁎ Tel.: +44 7721 488037; fax: +44 1865 271089.
E-mail address: [email protected].
0167-7187/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.ijindorg.2013.05.001
Empirical evidence on the reaction of equilibrium market prices to
changes in advertising might be helpful to guide the debate on how
equilibrium market prices react to changes in advertising costs.
Second, advertising itself is an increasingly important business
activity. In the United States, media advertising accounts for almost
2 percentage points of GDP and in Europe for around one percent.1
In Austria, on which this paper focuses, advertising accounted for a
share of 0.009 of GDP in 2000. This was a substantial increase from
the year 1990 when the share of advertising in GDP was only
0.0061 (see Grohall et al., 2007). Advertising is one of the main
sources of revenue for the media industry and the internet as well
as for cultural and sporting events and a better understanding of
advertising is relevant for businesses in all these industries.
Third, the taxation of advertising is a recurring policy idea. Here I
mention only a few examples. While there are many cities and
towns worldwide that tax local advertising, there have also been
frequent attempts to introduce advertising taxes at state or national
levels. In 1987 the Florida legislature enacted a sales tax on a range
of services that included advertising. In a heavy storm of protests the
advertising tax was attacked as “unfair, unwise and unconstitutional”
(Hellerstein, 1988), and was repealed only six months after its enactment. More recently, in 2006 the Pennsylvanian senate discussed a bill
(Senate Bill 854) that attempted to introduce a six percent sales tax on
advertising in that state, but was not enacted (see Philadelphia
Business Journal 2006). In Europe, the Slovak Republic charged a tax
on all advertising expenditure which was eliminated when Slovakia
1
See Arkolakis (2008) and Kosmelj and Zabkar (2008).
332
F. Rauch / International Journal of Industrial Organization 31 (2013) 331–341
entered the European Union in 2004. Hence despite few actual observations of taxation of advertising at state or national levels, it remains
a recurring and important political subject, and an idea that is periodically discussed. Almost all countries have laws that ban advertising of
certain products, like cigarettes or health related products. These bans
are likely to have effects similar to a substantial increase of an advertising tax.
For this investigation I make use of a policy change in Austria in
2000 that harmonized regional taxes on advertising expenditure,
thus simultaneously increasing the tax in some parts of the country
while reducing it in others, and leaving them roughly similar in a
third group. Using this natural experiment, I provide two empirical
contributions in this article. First, I show empirically that the taxation
of advertising has a strong effect on the advertising expenditures of
firms. Second, I show that consumer prices move in different directions across industries. This variation allows me to classify industries
by the different ways in which their advertising works. This second
finding may be understood by the presence of conflicting informative
and persuasive forces. The economic literature has long distinguished
these two forms of advertising. Typically, persuasive advertising is
advertising which shifts demand outwards and/or decreases elasticities of substitution, both of which serve to increase market prices.
Informative advertising increases competition through improved
information and thus reduces consumer prices. 2
Throughout the debate advertising has remained a controversial
topic: some economists have argued that there are excessive amounts
of advertising, which therefore may be a good target for taxation,
while others have suggested that underprovision of advertising might
provide a case for a subsidy.3 The main cause of these conflicting policy
conclusions is that advertising can be seen as persuasive or as informative (see Bagwell, 2007). Butters (1977) defines these two views as
“advertising as a set of psychological ploys which induce consumers
to buy products or brands that they otherwise would not buy”, or as
“a provision of information which allows consumers to make more discriminating choices within the framework of a fixed set of preferences”.
This distinction has roots further back in the work of Alfred Marshall
(1919) who defined similar categories with the names of combative
and constructive advertising. The persuasive view of advertising typically sees changes in preferences in the form of an outward shift of
demand, a decrease of elasticities of substitution between products, or
increased monopoly power of firms, and thus increasing market prices,
while the informative view sees increased information for consumers,
thus stronger competition and lower market prices (see Nelson 1970,
1974, 1978 or Stahl 1989, 1994). A closely related distinction was
brought forward by Johnson and Myatt (2006) who call two related
types of advertising ‘hype’ and ‘real information’.
The persuasive view of advertising suggests that advertising shifts
demand outwards. Its proponents have often called for a tax on advertising. Kaldor (1950) asserts a harmful effect of advertising and
suggests the introduction of a tax on advertising. Further, Sutton
(1974) makes the distinction between generated sales and diverting
sales from advertising, where the described case would be encountered if there were only diverted and no generated sales. Finally,
Gasmi et al. (1992) suggest that the advertising game between
Pepsi and Coca Cola is a predatory competition that hardly serves to
generate sales for the industry, and should be taxed. By the other
view, advertising might serve as a transporter of information. This
2
Ackerberg (2001) argues that advertising that provides product information can be
identified, as it should only attract consumers that are inexperienced with the brand.
However, this definition of informational advertising is different from the one used
in this paper.
3
Some examples for these different viewpoints are: Pigou suggested a tax on advertising in 1929, in addition Dixit and Norman (1978) argued for the possible presence of
excessive amounts of advertising. Stivers and Tremblay (2005) present the case for a
subsidy. Meurer and Stahl (1994) and Stegeman (1991) discuss models that can have
both outcomes.
idea has been formalized in models closely linked to the large literature on consumer search, but instead of consumers searching for
products, firms search for consumers via advertising. Here
advertising provides useful information to consumers such as the existence, the quality, or the price of a good. This idea has been formalized in models of informative advertising as for example in Butters
(1977) or Stahl (1989). In these models advertising expenditure has
a marginal effect on a firm's demand that will correspond to the marginal advertising costs it faces. Therefore a change of the cost function
will likely change advertising expenditure, and thus demand. It follows that in these models the taxation of advertising has in general
a clear effect on firm variables: more advertisements increase competition in the final goods market and thus lower prices. As demonstrated by Stahl (1989), in these models a subsidy for advertising may be
desirable. 4 Thus the discussion of information versus persuasion in
advertising follows a large existing literature, and the marginal price
responses could be informative about whether informative or persuasive forces dominate in an industry.
This paper proceeds as follows: Section 2 describes the data and
empirical strategy used to estimate the effects of a change in marginal
advertising costs on advertising expenditures and consumer prices,
and Section 3 presents the main empirical results. Section 4 concludes.
2. Empirical strategy and data
The empirical investigation relies on a policy change of the tax on
advertising in Austria in 2000. Austria is one of the few countries in
the world, and the only OECD country, that collects a tax on advertising. The nationwide tax is officially called ‘Werbeabgabe’, and locally
referred to as ‘Werbesteuer’. It covers advertising for goods and services from all industries. A constant fraction of advertising expenditure that a firm pays to the media has to be paid by the advertising
firm to the authorities as advertising tax. There are only a few exempt
companies or publications, such as advertising expenditure for content in student run school magazines, or the advertising by churches
and benevolent non-profit organizations. The tax includes all television and radio spots, advertisements in newspapers and magazines,
and expenditure for all other publicly displayed advertisements. The
Austrian advertising tax does not include advertising on the internet.
However, the internet was not yet a relevant outlet for advertising in
2000 on which this study is based and thus not a major concern here.
For details concerning the tax see Grohall et al. (2007).
The advertising tax was introduced in Austria in 1927, and has
remained in place ever since without interruption. Up to the year
2000 it was collected at regional levels with different tax systems in
different regions, whereby the location of the publication in which
the advertisement appears and not the location of the firm determines the payable tax rate. The states Tirol and Burgenland 5 did not
collect any advertising tax. The amount payable in the other states
was typically ten percent of advertising expenditure. In the state of
Salzburg the tax was only collected in the city of Salzburg and not
in the rest of the state, and the state Vorarlberg had a tax of only five
percent (see Bundesgesetzblatt, 2000; VÖZ, 1995). At the overall national level, at which the large majority of firms operate and most advertisements are made, the tax was also ten percent up to the year 2000.
After a change of the law, which took effect on June 1st, 2000, the
tax has been collected at the national level with an overall tax rate of
5% (WKO, 2002) for national, regional and local advertising alike.
Hence the year 2000 brought about an increase of the advertising
4
Grossman and Shapiro (1984) argue however that in the case of differentiated
products advertising can lead to an inflation of the number of firms, which does not
suggest the case for subsidy.
5
Throughout I refer to the nine regional units of Austria as states. In German these
units are called Bundesländer. In other publications they may be referred to as provinces or regions.
F. Rauch / International Journal of Industrial Organization 31 (2013) 331–341
tax from zero to five percent for local advertisers in Tirol and Burgenland, while in most of the other states and for nation-wide advertising
the tax rate dropped from ten to five percent. For a more detailed
legal description of the tax before and after the policy change see
the legal description (Bundesgesetzblatt, 2000). Fig. 1 shows a map
of Austria and highlights the tax changes. In most of the following
empirical analysis I omit the states Salzburg and Vorarlberg, where
the initial taxes differed at sub-state level, and focus on the difference
between tax increase states and tax decrease states.
Two common sets of critiques of differences-in-differences or
similar estimators relate to possible endogeneity of selection of the
treated group and the timing of the shock. As for endogeneity, since
in the present case the policy was of a unifying nature, the difference
between the two treatment groups emerges from prior decisions
whether or whether not to install a tax on advertising. This difference,
and thus the selection into different treatment groups, dates back to
the year 1927 when some states decided the matter differently from
the rest. Between the years 1927 and 2000, the Austrian First Republic
under which the law was established was replaced by a homemade
dictatorship, then under German rule, after which the country was
governed by the four allies, finally becoming the current republic in
1955. In addition the country experienced the Great Depression, the
Second World War, several changes of currency, and membership of
the European Union. Thus the economic and political conditions, the
legal environment and the structure of the media in the considered
states changed dramatically over this period, and there is reason to
think that many political differences in the initial period were moderated over this period.
Another concern is that it may be argued that the two treatment
groups in the presented experiment are not comparable. Since they
lived under different tax regimes prior to the policy change, they
might have selected themselves into the more suitable one. There
are several answers to this. First, in the regressions concerning advertising expenditure the sample is restricted to businesses that advertise in one state only. Since Austrian states are small (the median
populated state Tirol had less than 700,000 inhabitants in 2000), a
regression that only involves local advertisers concerns mainly
small local businesses whose location decisions often go back to the
birthplace of their founders. Second, I provide a robustness check of
within firm reallocation of advertising expenditures. In this specification the treatment groups consist of the same firms, and only studies
within firm reallocation of advertising, and thus do not allow for a selection bias. Third, it should be noted that Austrian states are small in
international comparison, and they are less important policy makers
than states in other countries. For example, states do not collect
income taxes and receive most of their budget and policies from the
Austrian national government.
The timing of the policy would invalidate the estimation if it were
chosen with considerations that favor one of the two groups, or if it
was anticipated. The report of the finance committee of the Austrian
parliament, which drafted the law to harmonize the taxation of
advertising across states, lists the following reasons for its decision
Fig. 1. This map shows the heterogeneous changes of advertising taxes in Austria in
2000. In five states, most parts of the country, the tax was reduced from ten to five percent. In Tirol and Burgenland it went in the other direction, from zero to five percent.
Vorarlberg and Salzburg had mixed tax regimes in state subdivision units.
333
Table 1
Frequency of articles on Tirol in newspapers.
Year
Economist
NYT
WB
Zeit
1996
1997
1998
1999
2000
2001
2002
2003
2004
n.a.
0
1
0
0
1
0
0
0
7
6
13
11
5
8
5
5
7
n.a.
n.a.
702
748
741
776
863
795
735
21
20
11
26
24
27
15
22
28
Note: This table shows how often the keyword “Tirol” or “Tyrol” was mentioned in the
following newspapers: The Economist, The New York Times, Wirtschaftsblatt and Die Zeit
in different years.
to bring this proposal forward (see Bundesgesetzblatt, 2000): it refers
to the lengthy debate about the usefulness of such a tax in the country, it lists administrative complications for trans-state businesses due
to the different tax regimes, and calls for a general harmonization of
taxes to avoid tax competition, although it does not see any signs of
such a tax competition taking place, in fact the minutes note that
there has been remarkably little adjustment, without pointing to specific references. The minutes from the discussion in the corresponding parliamentary subgroup show that there was one particular
cause for the timing of the initiative: A decision of the constitutional
high court of Austria (VwGH) from the year 1998 ruled that each
local authority may only tax the advertising value generated strictly
on its territory. As apparent from the minutes, this decision made
the collection of the tax practically impossible in the case of local
radio stations. In turn parliament felt that the law had to be adjusted.
With no objections, the change was waved through parliament and
implemented as quickly as is possible for legal changes on national
level. None of these reasons suggests that the timing of the harmonization was chosen in a way that would benefit a particular state.
Before the harmonization there was a recurring demand from the
chamber of commerce (WKO) and some journalists to abolish the
tax altogether, but no political party or representative took the matter
into their program. Thus there was no reason to anticipate the harmonization for the year stated prior to the court ruling at the end of
1998. This court ruling was not reported in the press, hence an
informed firm could have reasonably anticipated the tax change
only after the parliamentary announcements in mid 1999, no more
than six months before its introduction.
Another circumstance to invalidate the natural experiment would
be if Tirol and Burgenland were simultaneously affected by an important other shock that drove or biased the results. I checked the
archives of several national and international newspapers to see if
the name Tirol was mentioned significantly more often in 2000
than in the other years of the sample. This was not the case in the
archives of the Economist, the New York Times (which has been
used in economic studies before to indicate importance of events,
the Wirtschaftsblatt, the only Austrian daily that is primarily interested in economic matters, and Die Zeit, a German newspaper that
reports frequently on Austrian events, see Table 1. Similar tables
for the control group states also do not show an unusual quotation
frequency of either state in the year 2000. Similarly, incomes were
comparable between the two treatment groups: with a GDP per
capita of 26,300 Euro, Tirol was the fourth richest state in Austria in
2000, a position it held in all prior years of the sample (i.e. back to
1995), and much smaller Burgenland was the poorest state.6 Estimations
along the lines of Eq. (1) did not suggest that any state experienced a
break in its time trend of income in 2000.
6
Source of regional GDP data: Statistik Austria online database.
334
F. Rauch / International Journal of Industrial Organization 31 (2013) 331–341
The data on advertising expenditures come from “FOCUS Research
& Consulting Austria” (see Focus, 2008). This company measures
square centimeters of advertisements in newspapers and magazines
as well as seconds of advertisements on TV and radio. Using the
advertising price lists of these publications, they estimate the advertising expenditures of firms. The company supplied me with their
complete dataset from 1995 to 2005, which records advertising
expenditure per firm, year, publication and industry. Further they
provide the area where each publication is available in each year,
and an industry classification for each of the advertising companies.
This dataset does not include all publications available in Austria,
but with over 400 news providers it covers all major ones and a
wide range of small local magazines. Table 2 reports the number of
regional advertisements recorded, the number of regional media
(typically local newspapers, here only shown if they are exclusive to
one state), the log average cost of an advertisement and the number
of firms advertising in each of these publications, by state and also
separately for the post and pre treatment period. In total, the dataset
contains about 700,000 advertisements for the period considered.
Table 2 shows that in terms of firms per publication and average
advertising expenditures the values for the state Tirol (that carries
most of the different treatments) do not differ widely from values
of other states. The number of advertisements and regional media
in the treatment state Tirol is at the top of the distribution, in the
post period together with Steiermark. The relatively high number of
regional media in Tirol in this sample could be due to geographical
reasons. Tirol shares only a small fraction of its border with other
Austrian states, hence other states have more trans-state advertisers.
Burgenland exhibits the smallest number of firms per publication,
reflecting its small size and its comparatively low GDP per capita.
The analysis of consumer prices relies on two different datasets
provided by the Austrian federal statistical office (Statistik Austria).
They provide price indexes for twelve different industries and their
subcategories, classified by the COICOP (Classification of Individual
Consumption According to Purpose) system of classification from
the UN, on a state level and over the period from 1997 to 2003. The
disadvantage of this classification is that it is not fine enough to
distinguish individual products, and only informs about four digit
product groups. Its advantage is that it spans the complete universe
of consumer products in Austria, and is designed to be comparable
across states and years. It is the most detailed aggregation of products
for which the Statistik Austria was willing to share its complete data
set. In addition, Statistik Austria provided me with a selection of 40
randomly selected individual products at the most detailed level of
aggregation in different states in the same panel.
Fig. 2 displays prices of goods for the main COICOP groups for Tirol
and Burgenland and the mean price for that good for the remaining
six states in Austria, with Salzburg and Vorarlberg omitted. I also
omit communication (COICOP two-digit item 8) which does not
vary at state level, as the national mail and the phone companies all
operate nationwide with the same prices, from the estimations.
Typically I estimate an equation of the following type:
yit ¼ Post t β1 þ TaxInci β2 þ Post t TaxInci β3 þ it :
ð1Þ
This regression is similar to a difference-in-difference estimation,
with the exception that it uses two treatment groups with different
treatment, and no unaffected control group. Post is a variable that
takes a value of one for the year of 2003 and a value of zero in the
year 1997. TaxInci is a dummy variable that indicates the firms advertising in the states of Tirol and Burgenland, the states with the tax increase. The interaction Post × TaxInci gives the differential coefficient
of interest. In the regressions robust standard errors are clustered by
state. Although the used data are a panel with many time dimensions,
I typically focus on the third years before and the third years after the
treatment (1997 and 2003) in order to avoid correlation of errors in
the regression (see Bertrand et al., 2004).
3. Results
3.1. Effect on advertising expenditures
As a first approach to the data on advertising expenditures I display
total advertising expenditures for tax increase, tax decrease and other
states separately. In Fig. 3 I display these three series as indexes
normalized to one for the year 1996. As the figure shows, tax decrease
states exhibit a higher growth rate of advertising expenditures. There
is a small difference between tax increase and tax decrease states in
1999 and 2000, which might reflect some firms that anticipated
the tax change. This difference widens greatly from 2000 onwards,
in the expected direction. Also, as expected, the line for the “Other”
states, that experienced a tax change in between the change of the
two other sets of states, lies between the other two states. The central
of these three lines is closest to a neutral control group. The tax decrease states are closer to that curve in the years after 2000 than the
tax increase states, which could suggest that the differential effect
mainly comes from reactions of firms in the tax increase states.
Given that the treatment differs at state level, I only use local advertisers, defined as firms that advertise in one state only to see how
growth rates of advertising reacted to the tax change. Since most advertising in Austria is at the national level or across more than one state,
this restriction reduces the number of observations in the dataset to
less than ten percent. In addition I only use firms that have positive advertising expenditures in all eleven years of the panel from 1995 to
2005 to obtain a balanced panel. The resulting dataset consists of 877
observations in the states of Tirol and Burgenland and 1120 in the
other states, and includes a complete panel of total advertising expenditure for each of these firms and each year. The data report the expenditures as they arrive at the publications, and thus report the expenditure
net of taxes. Table 3 provides p-values from a two sided test of the
equality of growth rates in both sets of states. The test only suggests a
difference in the year 2000 with a p-value of 0.03 in the two sided
test, while in all other years the p-values of a simple test of differences
of means are above 0.05 and also above 0.1. As a placebo test I replace
the states with different tax changes with each of the other five states
of Austria (again excluding Salzburg and Vorarlberg and in addition
Tirol and Burgenland), and rerun the exercise for all these six states
and nine years. These p-values from a test for the difference of means
are reported in the right panel in Table 3. In the placebo table there is
no difference significance at one percent level, and two that are significant are at five percent level.
Table 3 suggests that the difference of growth rates of advertising
expenditures as received by the media between states is 25%. If firm
expenditures are considered, this difference is less, since those firms
that increased their expenditures also had to pay less tax. If the 9.5%7
tax difference is accounted for, a 9.5% cost difference resulted in a
15.5% difference of advertising expenditures. Thus on average the
estimates suggest that a one percent increase of advertising costs results
in a 1.6% reduction of advertising expenditure, conditional on firms not
exiting from the advertising markets. This is the estimate of the elasticity of advertising expenditures with respect to marginal advertising
costs.
To see how the effect differs across industries I provide estimates
from a differences-in-differences regression as described in Eq. (1) in
the upper part in Table 4. I focus on the third year before and the third
year after the treatment (the years 1997 and 2003), which is the most
conservative strategy to address the problem of auto-correlation of errors (see Bertrand et al., 2004). I use only firms that are present with
7
Tirol and Burgenland experienced a five percent increase of costs from 1 to 1.05,
and the other states a decrease of 4.5% from 1.1 to 1.05.
F. Rauch / International Journal of Industrial Organization 31 (2013) 331–341
335
Table 2
Descriptive statistics.
Population
Nominal GDP/cap.
Number of
advertisements
Number of
publications
Log mean
expenditure
Number of firms
per publication
1997 (1000)
1997 (Mio)
1995–1999 (1000)
1995–1999
1995–1999
1995–1999
3962
9538
19,186
11,980
7156
16,401
12,640
2601
415
277
561
1526
1364
511
1184
659
344
1542
14.6
19.2
18.6
21.9
26.3
19.4
23.2
23.7
32.8
3.5
11.7
11.8
17.2
14.0
61.6
22.5
10.1
10.8
7
10
10
8
8
13
15
7
12
11.9
12.3
12.6
13
12.6
13.1
12.3
12.6
13.4
126
258
246
347
293
257
276
281
217
Post
2003 (1000)
2003 (Mio)
2001–2005 (1000)
2001–2005
2001–2005
2001–2005
Burgenland
Kärnten
Niederösterreich
Oberösterreich
Salzburg
Steiermark
Tirol
Vorarlberg
Wien
Source:
SA
276
558
1553
1385
518
1190
681
356
1601
SA
18.6
22.8
22.1
26.4
30.6
23.3
28.4
28.5
38.4
SA
8.5
15.9
20.1
26.5
21.2
22.3
31.1
12.7
13.5
Focus
11
12
18
14
15
21
21
10
16
Focus
12.13
12.44
12.49
12.91
12.55
13.1
12.4
12.6
13.5
Focus
153
283
253
439
309
248
318
262
166
Focus
Tax change
in 2000
Area
Pre
Burgenland
Kärnten
Niederösterreich
Oberösterreich
Salzburg
Steiermark
Tirol
Vorarlberg
Wien
Up
Down
Down
Down
Other
Down
Up
Other
Down
Note: Summary statistics for the nine states of Austria. The last four columns consider only regional advertisements, defined as those that are published in publications that appear
in only one state. The source abbreviation SA stands for Statistik Austria (2010), Focus refers to Focus (2008).
positive advertising expenditure in both these years. Further I again use
only advertisers that advertise in a single state. The industry classification is taken from the advertising expenditure dataset and similar to
the one constructed by its supplier. I only provide results for industries
with at least 30 overall observations. I include state fixed effects in all
regressions and cluster standard errors by state. There is evidence for
Fig. 2. Prices. Note: State price index separately for states with increasing and decreasing marginal costs for advertising and 95% confidence intervals around these means. All
COICOP 2 consumer price categories are included, except for the prices of Mail and Telecommunications, which do not vary at state level. The states Salzburg and Vorarlberg
which have no clear tax change are omitted from the figure.
336
F. Rauch / International Journal of Industrial Organization 31 (2013) 331–341
Levels, expenditures (index, 1996=1)
2
Table 3
Differences of advertising expenditure growth rates.
1
1.5
1997
1998
1999
2000
2001
2002
2003
2004
1996
1998
2000
2002
2004
2006
year
Tax decrease
Other
Tax increase
Fig. 3. Total advertising expenditures in separate graphs for states with tax increase
(Burgenland and Tirol), mixed tax changes (Salzburg and Vorarlberg) and tax decrease
(the five remaining states). The time series are converted to an index that is one in
1996 for all sets of states. The displayed 95 percent confidence intervals based on
robust standard errors are computed from year fixed effects.
a significant reduction of advertising expenditures in Tirol and Burgenland for some industries. All coefficients that differ significantly from
zero at a ten percent level of significance show a negative coefficient
in the differential coefficient, and also the overall mean effect is negative
and significantly different from zero.8
In addition I analyze the within firm reallocation. For this exercise
I again observe only firms that have positive advertising expenditures
in both the years 1997 and 2003 in Tirol or Burgenland and at least
one other state. For each firm I aggregate all advertising in states outside Tirol and Burgenland. Thus for the regression there remains a
sample with four advertising expenditure observations per firm,
within and outside of Tirol in the years 1997 and 2003. This estimation strategy can provide robustness with respect to a possible selection effect, since in this specification the two treatment groups consist
of different advertising expenditures of the same firms. In addition,
this estimation provides typically more observations since most
of the firms in Austria advertise in more than one state. Using
this sample I estimate in OLS a differences-in-differences regression
using again robust standard errors, and in this exercise firm fixed
effects. The results are reported in the second part in Table 4. There
is stronger evidence for a reduction of advertising expenditures in
Tirol within firms overall, and for all industries except tourism,
coefficients do not vary strongly across industries.
In addition I verify if these expenditures correlate with advertising
intensity of industries, but find no evidence for such a correlation. The
number of newspapers in which firms decide to advertise did not
decline significantly with the advertising tax and remained very
stable in states with either direction of the tax change alike. In states
with an increase of the tax the mean number of newspapers per firm
changed from 2.9 to 2.6 between 1995 and 2005, while in states with
a tax decrease the mean number changed from 2.3 to 2.2. I did not
find a differential effect for any of the years in between.
An additional concern is that the prices that newspapers charge
advertisers could have changed, and that most of the effect is
absorbed this way. While I do not observe the prices that newspapers
charge directly, I can find evidence from two different channels that
this effect is not too pronounced. First, there are types of advertising
for which prices have not changed, apart from the tax change, for
8
I repeat the estimation of column of the first panel using the states classified as
‘Other’ in Fig. 1 as control group. Coefficient ‘Post × Treat’ is −0.24 with a robust standard error of 0.0105, significant at the 5% level of significance. Comparing the ‘Other’
states with the tax decrease states, the differential coefficient is −0.027 with a robust
standard error of 0.1.
Tirol + Burgenland
Kärnten
Niederö.
Oberö.
Steiermark Wien
td − tu
p
p
p
p
p
p
0.13
0.10
−0.004
0.25
−0.19
0.01
−0.001
0.14
0.349
0.422
0.978
0.033
0.123
0.941
0.993
0.228
0.683
0.167
0.574
0.469
0.066
0.219
0.543
0.07
0.798
0.083
0.033
0.294
0.525
0.527
0.484
0.211
0.231
0.882
0.723
0.74
0.036
0.53
0.739
0.643
0.871
0.358
0.099
0.761
0.2
0.315
0.315
0.545
0.157
0.912
0.847
0.744
0.25
0.634
0.937
0.786
Note: The left part in this table provides results of tests for differences of mean growth
rates of advertising expenditures between states with decreasing and increasing
marginal tax rates for advertising (tu − td) and the p-value from a two-sided test of
the difference of the coefficients. The right side in the table provides the p-values of
corresponding placebo exercises with each state of the sample with tax decreases tested against the others. The states Salzburg and Vorarlberg are omitted throughout.
example handing out samples in the high street or supermarkets, or
sponsoring or hosting events. The biggest such type of advertising
that is included in the Focus data is unsolicited mail advertising. I
find no substitution into such advertising in either the treatment of
control state for this type of advertising. Note also that 55% of local
advertisements are in national newspapers, that sell the opportunity
to advertise only in selected parts of the country. These newspapers
do not charge different per-reader rates in different states. Thus for
this majority of publications, even for local advertisements, advertising rates did not change.
3.2. Effect on exit from advertising
The estimations so far use a non-balanced panel, as only firms that
do not stop advertising are considered. I investigate exit from advertising separately, by estimating Eq. (1) using a probit estimator, for the
same industries as in Table 4. The table demonstrates that there is
indeed increased exit from advertising in states where the advertising
tax increased, with the most statistically significant coefficient in the
clothing industry. This finding suggests that the coefficients presented
in Table 4 underestimate the true effect, as the underlying data exclude
many firms that reacted to the advertising tax change by exit.
Following this finding on exit, I estimate the aggregate effect as in
Table 4 only for those industries, where I do not find significant
evidence for exit from advertising on the five percent level of significance in Table 5. The aggregate effect is statistically significantly
different from zero at five percent level of significance, and not significantly different from the effect in Table 4.
4. Prices
In the literature there is some partial, empirical evidence of how
advertising affects prices. In particular, several studies make use of
bans of advertising in certain areas for certain goods. They found
that advertising seems to decrease prices for eyeglasses (Kwoka,
1984), children's breakfast cereal (Clark, 2007), and drugs (Cady,
1976). Milyo and Waldfogel (1999) present evidence that advertising
decreases the price of advertised goods in liquor stores, while it increases the price of non-advertised goods in the same stores. On the
other hand, Gallet and Euzent (2002) suggest that advertising to
sales ratios have a positive effect on the supply price in the brewing
industry, although Gisser (1999) found no significant relationship in
the same industry. The present study is the first that provides such
elasticities for all industries.
Again the estimation follows the strategy similar to the common
differences-in-differences approach described in Eq. (1). To avoid autocorrelation in the errors I use one observation per time period and
unit, the mean price for each state and industry for the years 1997
F. Rauch / International Journal of Industrial Organization 31 (2013) 331–341
337
Table 4
Advertising expenditures across industries.
(1)
All
(2)
Audio
Differences between local advertisers
Post
0.0778
(0.0971)
Post × TaxInc
−0.307⁎⁎
(0.0966)
State fixed effects
Yes
Observations
1064
With tax increase
428
Within firm reallocation
Post
0.117⁎⁎⁎
(0.106)
Post × TaxInc
−0.191⁎⁎⁎
(0.011)
Yes
1048
Firm fixed effects
Observations
(3)
Construction
(4)
Services
(5)
Leisure
(6)
Retail trade
(7)
House garden
(8)
Automotive
(9)
Clothing
(10)
Tourism
−0.105
(0.167)
−0.283
(0.167)
Yes
56
23
0.296
(0.327)
−0.860⁎
(0.327)
Yes
101
48
−0.0160
(0.200)
−0.170
(0.200)
Yes
123
44
0.0536
(0.654)
0.0245
(0.654)
Yes
30
22
−0.114
(0.557)
−0.494
(0.557)
Yes
43
21
0.0958
(0.0974)
−0.270⁎⁎
(0.0974)
Yes
114
52
0.343⁎⁎
(0.132)
−0.287⁎
(0.132)
Yes
326
100
−0.352
(0.223)
−0.372
(0.223)
Yes
113
51
−0.110
(0.228)
0.184
(0.228)
Yes
85
40
0.178⁎⁎⁎
(0.412)
−0.186⁎⁎⁎
(0.045)
Yes
64
0.109⁎⁎⁎
(0.346)
−0.184⁎⁎⁎
(0.034)
Yes
124
0.157⁎⁎⁎
(0.246)
−0.224⁎⁎⁎
(0.027)
Yes
174
0.072
(0.525)
−0.206⁎⁎⁎
(0.050)
Yes
56
0.061
(0.398)
−0.199⁎⁎⁎
(0.037)
Yes
88
0.127⁎⁎⁎
(0.270)
−0.209⁎⁎⁎
(0.024)
Yes
108
0.129⁎⁎
(0.531)
−0.207⁎⁎⁎
0.022
(0.350)
−0.136⁎⁎⁎
(0.035)
Yes
68
0.040
(0.496)
−0.830
(0.626)
Yes
40
(0.044)
Yes
28
Note: Both parts in the table report diff-in-diff estimates of advertising expenditures using the years 1997 and 2003 in different panels for all firms and by sector. The first panel
considers local advertisers; the second panel considers within firm reallocations for cross-state advertisers. Robust standard errors are in parentheses, in the upper panel they
are clustered by state.
⁎⁎⁎ Denotes significance at one percent level.
⁎⁎ Denotes significance at five percent level.
⁎ Denotes significance at ten percent level.
Table 5
Exit from advertising.
Post
Post × TaxInc
State fixed
effects
Observations
(1)
All
(2)
Audio
−0.233⁎⁎⁎
(0.033)
0.174⁎⁎
(0.071)
Yes
−0.000656
(0.170)
0.689⁎
(0.404)
Yes
16,470
707
(3)
Construction
(4)
Services
(5)
Leisure
(6)
Retail trade
(7)
House garden
(8)
Automotive
(9)
Clothing
(10)
Tourism
−0.289⁎⁎⁎
(0.102)
−0.0411
(0.214)
Yes
−0.191⁎⁎
(0.083)
0.385⁎⁎
(0.193)
Yes
−0.219
(0.226)
0.839⁎⁎
(0.400)
Yes
−0.416⁎⁎⁎
(0.154)
0.101
(0.297)
Yes
−0.226⁎⁎
(0.110)
0.371⁎
(0.217)
Yes
−0.129
(0.085)
−0.0833
(0.212)
Yes
−0.191
(0.139)
0.777⁎⁎⁎
(0.267)
Yes
−0.426⁎⁎⁎
(0.089)
−0.313⁎
(0.177)
Yes
1492
3647
421
862
1378
1884
1120
2083
Note: The table shows random effects probit estimation results of firm exit, which is defined as a dummy variable that takes a value of one in the year in which a firm stops to
advertise. Estimates are provided for all firms, and by sector. Post is a variable equal to one in the years 2001–2004 and equal to zero in the years 1995–1999, the data are
collapsed into these two periods.
⁎⁎⁎ Denotes significance at one percent level.
⁎⁎ Denotes significance at five percent level.
⁎ Denotes significance at ten percent level.
Table 6
Prices, main industry groups.
(1)
Food
Post
TaxInc
Post × TaxInc
State f.e.
Obs
0.0203⁎⁎
(0.00554)
−0.00698
(0.00437)
0.0367⁎⁎⁎
(0.00310)
Yes
(0.00433)
42
(2)
Alc Tob
(3)
Garment
(4)
Housing
0.0656⁎⁎⁎
(0.00424)
0.0101⁎
0.0101
(0.0115)
0.0119
(0.0137)
0.0100
(0.00721)
Yes
(0.0123)
42
0.0493⁎⁎⁎
(0.00707)
0.00983
(0.00511)
−0.00345
(0.00566)
Yes
(0.00221)
56
(0.00436)
−0.0122⁎⁎⁎
(0.00269)
Yes
(0.00337)
42
(5)
Furnishing
0.0398⁎⁎⁎
(0.00662)
0.0107⁎
(0.00480)
−0.000850
(0.00638)
Yes
(0.00280)
98
(6)
Health
(7)
Transport
(8)
Recreation
(9)
Education
(10)
Hotels
(11)
Other
0.0885⁎⁎⁎
(0.00393)
0.00282
(0.00530)
0.00819
(0.00503)
Yes
(0.00353)
56
0.0709⁎⁎⁎
(0.00336)
0.0195⁎
−0.0239⁎⁎⁎
(0.00440)
−0.00400
(0.0139)
0.0223
(0.0126)
Yes
(0.00618)
56
0.252⁎⁎⁎
(0.0357)
−0.00465
(0.0346)
0.104⁎⁎
(0.0284)
Yes
(0.0396)
14
0.0738⁎⁎⁎
(0.00376)
0.0110⁎
0.0601⁎⁎⁎
(0.00465)
0.0166⁎⁎
(0.00536)
−0.0128⁎
(0.00622)
Yes
(0.00218)
42
(0.00524)
0.0155⁎⁎⁎
(0.00402)
Yes
(0.00374)
70
(0.00901)
−0.0226⁎⁎
(0.00808)
Yes
(0.00370)
56
Note: Diff-in-diff estimates of prices, for all pooled prices and for industries (all two digit COICOP industries are included except mail and telecommunication which does not vary at
state level). I use the most disaggregated prices available for each COICOP-2. The treatment group consists of the states where advertising taxes increased (Tirol and Burgenland).
Salzburg and Vorarlberg are omitted from the estimation. The pre period uses collapsed prices from 1997 to 1999, and the post period from 2001 to 2003. The coefficient on
Post × TaxInc reports the differential effect. Robust standard errors are clustered by state.
⁎⁎⁎ Denotes significance at one percent level.
⁎⁎ Denotes significance at five percent level.
⁎ Denotes significance at ten percent level.
338
F. Rauch / International Journal of Industrial Organization 31 (2013) 331–341
Table 7
Prices, aggregate magnitude and placebo test.
(1)
Pool 3 years
Post
TaxInc
Post × TaxInc
State f.e.
State trend
Observations
0.052⁎⁎⁎
(0.002)
0.008⁎⁎
(0.002)
0.006⁎⁎⁎
(0.001)
Yes
602
(2)
Trend
0.610⁎⁎⁎
(0.0088)
0.013
(0.0092)
0.0276⁎⁎
(0.0076)
Yes
Yes
602
(3)
Pool, weights
(4)
1999–2001
(5)
1999–2002
(6)
1999–2003
(8)
1999–2004
(9)
1995–1997
(10)
2003–2005
0.044⁎⁎⁎
(0.002)
0.008⁎⁎⁎
(0.002)
0.0034⁎⁎
(0.001)
Yes
0.0218⁎⁎⁎
(0.00202)
0.00866⁎⁎
(0.00322)
0.00446
(0.00258)
Yes
0.0446⁎⁎⁎
(0.00174)
0.00961⁎⁎
(0.00291)
0.00439⁎
(0.00208)
Yes
0.0614⁎⁎⁎
(0.00158)
0.0109⁎⁎⁎
(0.00277)
0.00472⁎⁎
(0.00152)
Yes
0.0724⁎⁎⁎
(0.00203)
0.0111⁎⁎⁎
(0.00297)
0.00439⁎⁎
(0.00165)
Yes
−0.235⁎⁎⁎
(0.00473)
−0.00394
(0.00429)
0.00489
(0.00334)
Yes
0.0334⁎⁎⁎
(0.00145)
0.00998⁎
(0.00465)
0.00290
(0.00252)
Yes
602
602
602
602
602
602
602
Note: Diff-in-diff estimates of pooled prices. For each industry the most disaggregated prices available are used. The treatment group consists of the states where advertising taxes
increased (Tirol and Burgenland). Salzburg and Vorarlberg are omitted from the estimation. The coefficient on Post × TaxInc reports the differential effect. Robust standard errors
are clustered by state. In the second column I include a linear trend. In the third column prices of industries are weighted by their weights in the national statistics, and by the share
of regional advertising. All other columns display unweighted results.
⁎⁎⁎ Denotes significance at one percent level.
⁎⁎ Denotes significance at five percent level.
⁎ Denotes significance at ten percent level.
to 1999 for the pre-treatment effect, and the mean for 2001 to 2003
for the post-treatment effect. Table 6 reports the estimates for all
COICOP class 1 industries, except communication that does not vary
at state level. To increase the number of observations I use the most
disaggregated level for which I can access price information, which
goes down to COICOP class 3 level, so that sub categories for industries are used. Some examples to illustrate the most disaggregated
level of industry available to me would be ‘Beer’, ‘Gas for heating’ or
‘Bicycles’. In these regressions I apply state fixed effects and cluster
robust standard errors by state. This table confirms what is observable in Fig. 2: there is evidence that prices increased in the industries
food and education, while they decreased for alcohol and tobacco,
transport, hotels and restaurants. Food sold in grocery stores is an
industry that frequently advertises on the streets of Austria with
concrete price information, often with direct reference to the competition. Hence it is not surprising to find an informative effect in this
industry. On the other hand, tobacco and alcohol are known for
their persuasive advertisements that appeal to emotion.
The overall effect is estimated in Table 7. The first column presents
the baseline specification, the second column includes a linear state
trend, and in the third column I weight prices by their importance
in the price index. Again these estimates suggest an increase of prices
in Tirol and Burgenland, however with a possible delay. The differential coefficient is positive, but not significant when the years
1999–2001 are compared. The effect remains at a similar magnitude
for the years 1999–2002, 1999–2003 and 1999–2004, but it increases
in significance. This could point to slow adaption to the new equilibrium. Columns (9) and (10) provide two placebo estimates for the
periods before and after the treatment (1995–1997 and 2003–2005)
and show no differential treatment effect. To find the magnitude of
the overall price change, I use the weights for each of the COICOP
one digit groups provided by Statistik Austria (2010). These weights
indicate the relevance of a given price in the overall price basket
for the inflation computation. I also adjust the treatment effect to
correspond to the difference of the strengths of the shock in different
industries by multiplying it with the share of advertising in each industry that is on the state level. 9
9
The differential coefficient of column (1) when replacing Tirol and Burgenland with
Salzburg and Vorarlberg (the ‘Other’ states in Fig. 1) is 0.0015 with a robust standard error
of 0.0005, significant at the five percent level of significance. The coefficient shows the
expected sign, and is smaller in magnitude. Comparing the ‘Other’ states with the tax
decrease states, the coefficient is −0.0014 with a robust standard error of 0.0005, again
significant at the five percent level, again showing the expected sign, and again smaller
in magnitudes.
The differential treatment effect appears strictly at the regional
level. Advertising at the national level experienced a reduction of
the tax on advertising from 10 to 5%. This reduction is similar for
national advertisers in the states with differential tax changes. The
treatment is thus stronger in industries in which the share of local
advertising is higher, and price effects can be expected to be larger
as well. I compute the share of local advertising for each industry
from the Focus data, and plot it against the estimated differential
price effects in Fig. 4. The figure shows COICOP category 2 prices for
which I found matching industries in the Focus data. See Appendix
A for details on the match. The figure shows a strong positive correlation between the share of local advertisers, and the price response.
Note that for those industries with a lower share of advertising, the
treatment effect is also weaker. If an industry exhibits 50% of its
advertising at the local level, its treatment effect is half. In this figure
it can be inferred that absolute marginal price responses are roughly
similar for different industries.
Prices at industry level, which are aggregations, may be too broad
to cover the true effects. Statistik Austria generally does not give
access to its data on prices of individual goods, but they agreed to
provide me with a small random sample to perform a robustness
check. I asked for prices that are comparable across states, from products that may be sold by small local businesses. The data consist of 40
such prices, at the most disaggregated level available to Statistik
Austria. Examples of these units that they supplied are beefsteak in a
restaurant, an hour of a car mechanic or a car wash. Some of these
panel series are incomplete, with missing states or missing years. I
drop these products, and in Table 8 I reproduce the same regressions
as the ones on the industry level using these detailed prices. As is
apparent, the tables give the same signs as the ones based on prices
at industry level on the differential effect. However, the number of
observations is lower, which sometimes may be the reason for
lower statistical significance.
5. Persuasive and informative advertising industries
The observed differences in consumer price changes across industries point to possible differences of the parameters across industries.
Consumers may be more easily persuaded to buy certain products
such as alcohol and tobacco than to buy certain food. In an industry
with parameters such that persuasion is costly, advertising content
focuses on the informative aspect of advertising, while industries
with persuasive potential will make efforts to put additional elements
into their advertisements.
F. Rauch / International Journal of Industrial Organization 31 (2013) 331–341
339
Fig. 4. Absolute price response by industry. Note: This figure shows the differential absolute price response at level of industry classification COICOP 2 as estimated above against the
share of advertising on state level in total advertising from Focus data and a linear trend computed using OLS.
Table 8
Diff-in-diff estimation for detailed product prices for selected industry groups. The treatment group consists of the states where advertising taxes increased (Tirol and Burgenland).
Salzburg and Vorarlberg are omitted from the estimation. The coefficient on Post × TaxInc reports the differential effect. Robust standard errors are clustered by state.
Post
TaxInc
Post × TaxInc
Observations
(1)
Beer
(2)
Wine
(3)
Food
−0.00211
(0.00933)
0.114⁎⁎⁎
0.154
(0.149)
−0.12
(0.0971)
0.211
(0.149)
14
−0.00332
(0.00344)
−0.0491⁎⁎⁎
0.0237⁎⁎⁎
(0.0055)
−0.0313⁎⁎
(0.00852)
0.0553⁎⁎⁎
(0.00344)
132
(0.0085)
0.0772⁎⁎⁎
(0.0055)
84
(0.0247)
−0.0312⁎⁎
(0.00933)
14
(4)
Non alcoholic drinks
(5)
Food in restaurants
0.303⁎⁎
(0.0937)
1.285⁎⁎
(0.382)
−0.175
(0.104)
126
(6)
Services
5.846⁎⁎⁎
(0.419)
1.268
(23.91)
−0.839
(4.054)
28
⁎⁎⁎Denotes significance at one percent level.
⁎⁎Denotes significance at five percent level.
⁎Denotes significance at ten percent level.
As discussed in the Introduction, there are theoretical arguments
to presume a positive correlation between the degree to which advertising leads to a price increase, and the persuasive content of advertising across industries. To test this prediction I rely on a study from the
marketing literature that provides a meta-analysis of the information
intensity in advertisements from different industries (Abernethy and
Franke, 1996, from here on referred to as AF, Table 2). In their article
they summarize the findings from 60 studies that measure information content of advertising. Using the methodology developed by
Resnik and Stern (1977), which relies on the count of well defined
information cues such as if an advertisement contains explicit price
information or comparisons with competitors, they give estimates
of the information share across industries. I merge their industry
with the COICOP industries from Statistik Austria as far as this was
possible with confidence (see Appendix A). There is no correlation
between the information content of advertising and the share of advertising at the state level. 10 Fig. 5 shows a clear positive correlation
between the price response to advertising suggested by the estimates
in Table 6 and the information content reported by this meta-study.
The correlation between the two series is 74%, a regression with
10
In a regression of the information content on the state share with robust standard
errors the p-value is larger than 0.76.
robust standard errors yields a slope that is significantly different
from zero at five percent level of confidence. The slope coefficient of
the displayed line is 0.02 with a robust standard error of 0.0077.
This suggests that the informative effect of advertising is observed
in industries with high information content in advertising. The
correlation with other industry characteristics that I observe is poor,
for example a Herfindahl index of advertising expenditures does not
correlate with marginal price responses, with a p-value of 0.75.
6. Summary and magnitudes
The results show a significant relative reduction of advertising
expenditures from advertising markets in the states of Tirol and
Burgenland after they experienced a relative increase of the overall
advertising tax and for the large majority of industries. Further, the
results highlight a within-firm reallocation of advertising expenditure
out of Tirol. While the advertising expenditures react consistently
across industries, consumer prices increase for some and decrease
for other industries. In particular, there is evidence that prices
increased in the industries food, transportation, education and tourism while they decrease for alcohol and tobacco, health, leisure and
house and garden. In the context of this discussion, food (which
largely represents food sold by grocery stores) is a large sector in
340
F. Rauch / International Journal of Industrial Organization 31 (2013) 331–341
Fig. 5. Price response and information. Note: This figure shows the marginal average price response by industry as estimated above in Table 6 against information content of
advertising in industries as reported in the meta study (AF) and a linear trend computed using OLS.
which the informative effect of advertising dominates, while alcohol
or tobacco are sectors in which persuasion dominates.
The weighted differential growth rates in Table 6 suggest that
mean prices in Tirol were about 1% higher than in the other states
after the introduction of the tax. This suggests that the 9.5% difference
in marginal advertising costs, or the 25.5% difference of advertising
expenditures as seen by publications, or the 16% difference of firm
expenditures on advertising resulted in a one percent difference of
consumer prices. Note that this mean effect is an average of positive
and negative growth rates. The effects for certain industries or products can be much higher or lower as apparent in Table 6.
7. Conclusion
In this paper I use a policy change that affected the taxation of
advertising to estimate how advertising costs affect advertising expenditure and consumer prices across industries. The results suggest
that taxation of advertising is an effective policy tool to reduce advertising. I estimate that a one percent increase of the marginal tax of
advertising decreases advertising expenditure by 1.6%. This reduction
is similar across industries.
In a second stage, I consider how this change of advertising intensity affects consumer prices. I find that advertising increases consumer prices in some industries such as food in grocery stores, while
decreasing them in others, such as alcohol and tobacco. Suggestions
from the theoretical literature are ambiguous. Persuasive advertising
would increase consumer prices, while informative advertising would
decrease them. Persuasive advertising is seen as welfare decreasing,
hence there would be a case to tax or otherwise restrict advertising
in industries where advertising increases market prices. In fact
many countries already restrict advertisements for alcohol and tobacco. On the other hand, informative advertising that decreases market
prices is seen as welfare enhancing, thus policy makers should be
careful not to tax advertising in those industries in which advertising
decreases prices.
For the mean price across all industries, the informative effect of
advertising seems to dominate. I estimate that a ten percent increase
of advertising tax leads to a 0.5% increase across consumer prices.
By this estimation, a complete abolition of the five percent advertising tax in Austria would lower consumer prices by about 0.25%
age points. This effect however would differ across industries. The
estimates suggest that prices would rise for alcohol, tobacco and
other persuasive industries, while food in grocery stores would
become cheaper.
Appendix A
In this appendix I provide details on the match of industries in the
COICOP classification from the Statistik Austria with the industries
from AF, Table 2 on page 10. Since the industries used by AF do not
follow a standard classification, I have to link them based on the
name. Given that all industries in question are large, well defined
groups this is fairly straight forward in most cases. I match COICOP
subgroups with the most fitting industry based on names. Then I
compute for each industry a weighted mean of information values
for each industry, the weights were taken from the price index baskets provided by Statistik Austria (2010). For the table, the estimates
from the 1997–2003 differences-in-differences estimates were compared with the percent informative measures provided by AF in
Table 2 in page 10.
The COICOP subgroups were merged as follows (the following
paragraph shows in bold the main COICOP industry, in italics the
COICOP two digit sub-industry, then in unformatted text the matched
AF industry): Health: Medical products and equipment: Medicine,
medical products; Out-patient services: Services; Hospital services:
Personal care; Clothing: Clothing: Clothing; Footwear: Clothing;
Hotels, Restaurants: Catering services: Financial, transportation,
travel; Accommodation services: Financial, transportation, travel;
Transport: Purchase of vehicles: Cars, Operation of personal transport
equipment: Financial, transportation, travel, Transport services: Financial, transportation, travel; House, water, electr., gas: Actual rentals
for housing: Laundry and household; Maintenance and repair of the
dwelling: Laundry and household; Water supply and miscellaneous
services: Services; Electricity, gas, and other fuels: Laundry and household; Durable goods: Furniture and furnishing, carpets, floor coverings:
Furniture, home furnishes, appliances; Household textiles: Household,
lawn, garden; Household appliances: Electronics; Glassware, tableware,
and household utensils: Household, lawn, garden; Tools and equipment
for house and garden: Furniture, home furnishes, appliances; Goods
and services for routine household maintenance: Services Recreation,
culture: Audio visual processing equipment: Hobbies, toy, transportation; Other major durables: Hobbies, toy, transportation; Other
F. Rauch / International Journal of Industrial Organization 31 (2013) 331–341
recreational items and equipment: Hobbies, toy, transportation;
Recreation and cultural services: Services; Newspapers, books and
stationary: Toys, leisure, entertainment; Package holidays: Services;
Miscellaneous: Personal care; Personal care; Personal effects; Hobbies,
toy, transportation; Social protection; Institutional; Insurance; Financial, transportation, travel; Financial services; Financial, transportation, travel; Other services; Services; Food: Food.
The match in Fig. 5 was done as follows (COICOP 2 for price data in
italics followed by normal text for the advertising data): Gas and fuel,
fuel; medical products, pharmaceutical and cosmetic products; shoes
(including repair), shoes; food, food; hotel services, hotels; leisure and
culture, touristic; household appliances, electronic appliances; water
supply, water supply; transportation, transportation; alcohol, alcohol;
furniture, furniture (trade); tools for house and garden, garden tools;
photography, optic trade; transportation services, services.
References
Abernethy, Avery M., Franke, George R., 1996. The information content of advertising: a
meta-analysis. Journal of Advertising 25, 1–17.
Ackerberg, Daniel A., 2001. Empirically distinguishing informative and prestige effects
of advertising. The RAND Journal of Economics 32, 316–333.
Arkolakis, Konstantinos, 2008. Market Penetration Costs and the New Consumers
Margin in International Trade. NBER Working Papers 14214.
Bagwell, Kyle, 2007. Handbook of Industrial Organization. In: Armstrong, Mark, Porter,
Robert (Eds.), North-Holland.
Bertrand, Marianna, Duflo, Esther, Mullainathan, Sendhil, 2004. How much should
we trust differences-in-differences estimates? Quarterly Journal of Economics
(February).
Bundesgesetzblatt für die Republik Österreich, Teil 1, Jahrgang 2000, Ausgegeben am
31. Mai 2000.
Butters, Gerard, 1977. Equilibrium distributions of sales and advertising prices. The
Review of Economic Studies 44, 465–491.
Cady, John F., 1976. An estimate of the price effects of restrictions on drug price advertising. Economic Inquiry 14, 493–510.
Clark, Robert C., 2007. Advertising restrictions and competition in the children's breakfast cereal industry. Journal of Law and Economics 50.
Dixit, Avinash, Norman, Victor, 1978. Advertising and welfare. Bell Journal of Economics
9, 1–17.
FOCUS Research & Consulting Austria, Advertising expenditure dataset, contact: Klaus
Fessel, database from 1995 to 2005, obtained on November 12th, 2008.
Gallet, C.A., Euzent, P.J., 2002. The business cycle and competition in the U.S. brewing
industry. Journal of Applied Business Research 18, 89–96.
341
Gasmi, F., Laffont, J.J., Vuong, Q., 1992. Econometric analysis of collusive behaviour in a
soft-drink market. Journal of Economics and Management Strategy 1, 277–311.
Gisser, M., 1999. Dynamic gains and static losses in oligopoly: evidence from the beer
industry. Economic Inquiry 37, 554–575.
Grohall, Günther, Gstrein, Michaela, Mateeva, Liliana, Schuh, Ulrich, Strohner, Ludwig,
Yegorov, Yuri, 2007. Eine ökonomische Analyse der Werbeabgabe, Research Report.
Institute for Advanced Studies, Vienna.
Grossman, Gene M., Shapiro, Carl, 1984. Informative advertising with differentiated
products. Review of Economic Studies 51.
Hellerstein, Walter, 1988. Florida's sales tax on services. National Tax Journal 41.
Johnson, Justin P., Myatt, David P., 2006. On the simple economics of advertising,
marketing, and product design. American Economic Review 96.
Kaldor, Nicholas, 1950. The economic aspects of advertising. The Review of Economic
Studies 18.
Kosmelj, Katarina, Zabkar, Vesna, 2008. A methodology for identifying time-trend
patterns: an application to the advertising expenditure of 28 European countries
in the 1994–2004 period. Metodoloski Zvezki 5, 161–171.
Kwoka, John, 1984. Advertising and the price and quality of optometric service. The
American Economic Review 74.
Marshall, Alfred, 1919. Industry and Trade. Macmillan Publishing.
Meurer, Michael, Stahl II, Dale O., 1994. Informative advertising and product match.
International Journal of Industrial Organization 12.
Milyo, Jeffrey, Waldfogel, Joel, 1999. The effect of price advertising on prices: evidence
in the wake of 44 Liquormart. The American Economic Review 89.
Nelson, Phillip, 1970. Information and consumer behavior. Journal of Political Economy
78.
Nelson, Phillip, 1974. Advertising as information. Journal of Political Economy 82 (4),
729–754.
Nelson, Phillip, 1978. Advertising as Information Once More in ‘Issues in Advertising:
the Economics of Persuasion. In: Tuerck, David G. (Ed.), American Enterprise Inst,
Washington.
Resnik, Alan, Stern, Bruce L., 1977. An Analysis of Information Content in Television
Advertising. Journal of Marketing 41, 50–53 (January).
Stahl, Dale O., 1989. Oligopolistic pricing with sequential consumer search. The
American Economic Review 79.
Stahl, Dale O., 1994. Oligopolistic pricing and advertising. Journal of Economic Theory
64.
Austria, Statistik, 2010. Basket of goods and services and weights. downloaded from
www.statistics.at on March 15th, 2010.
Stegeman, Mark, 1991. Advertising in competitive markets. American Economic
Review 81, 210–223.
Stivers, Andrew, Tremblay, Victor J., 2005. Advertising, search costs, and social welfare.
Information Economics and Policy 317–333.
Sutton, C.J., 1974. Advertising, concentration and competition. The Economic Journal
84, 56–69.
VÖZ, Verband Österreichischer Zeitungen, 1995. Luxussteuer auf Werbung - Dokumentation
zur Anzeigen- und Ankündigungsabgabe. Express Druck, Gutenbergstrasse, St. Pölten.
WKO, 2002. Österreichs Werbesteuer - Relikt und Unikat weltweit Press release of the
Wirtschaftskammer (Chamber of Commerce), Vienna 14th of November.