1. INTRODUCTION
Traditionally, empirical R&D-market structure analyses are at
the industry level, and disaggregation varying from 2 to 5 SIC digits,
using patents of invention as appropriability mechanisms (Lee (2005),
Arora (2008)). Cohen (2010) reviews recent studies and remarks how
modern data bases and panel analyses fill in the gaps in the previous
literature. Cohen and Levin (1989) and Cohen (2010) build a sound
literature review about this subject. On the other hand, Cohen, Nelson
and Walsh (2000) and Hall and Ziedonis (2001) show us that firms use a
mix of appropriability mechanisms (MAM).
In this paper we display empirical evidence about the relationship
among R&D, market structure and mix of appropriablility at the firm
level in Brazilian manufacturing. This allows us take account of more
fundamental sources of variation in the innovative behavior and
performance of firms and industries using firm level appropriability
mechanisms. It is possible thanks to two detailed surveys conducted by
the Brazilian Census Office (IBGE): the Industry Annual Survey (PIA) and
the Technological Innovation Survey (PINTEC).
Our empirical results suggest that market share and MAM increases
firm's R&D likelihood. As far we know, this is the first
empirical study in this field to examine Brazil and one of the few to
look at firm level data.
This paper has 3 remaining sections: section 2 reviews patens
limits and mix of appropriability mechanisms importance; section 3 shows
data base and variables; section 4 shows our econometric results; and
section 5 conclusions.
2. FROM PATENTS TO MIX OF APPROPRIABILITY MECHANISMS
There are limits to using patents as an appropriability indicator.
At least since Scherer (1965) we have known that a straight count of
patents has two limitations: (1) the propensity to patent an invention
of given quality may vary from firm to firm and from industry to
industry; and (2) the quality of the underlying inventions varies widely
from patent to patent.
Nowadays it is clear that not only patents but also others
intellectual protection tools have positive effects on the economy.
Copyright laws, for example, incentivize technological innovation and
allow better price discrimination in the US VHS and DVD market
(Mortimer, 2007).
In fact, intellectual protection through patents is not always the
best option for many firms. Cohen, Nelson and Walsh (2000) try to
explain why some American companies register patents and others not.
Analysing data from 1478 R&D labs in the American manufacturing
industry in 1994, they found that firms protect their innovation profits
not only through patents but also using a mix of intellectual property
mechanisms, which include industrial secret and leading time. Among
those mechanisms, patents are the fewest used while industrial secret
and leading time are the most common.
Hall and Ziedonis (2001) agree with Cohen, Nelson and Walsh (2000):
patents are not always the best option. In some cases there is a patent
paradox, as illustrated by their empirical study of 95 firms'
pattern standards in the US semiconductor industry between 1979 and
1995--an industry whose main characteristic is fast technological change
and cumulative innovation. The results show that those firms don't
always use patents to protect their R&D profits--which is a paradox
in a high and fast technological change sector.
While patents sometimes are not the best option in developed
countries, data limitations make developing countries a less than ideal
source of information about innovation. In general, data on patents have
three important restrictions: i) they measure inventions not innovation,
ii) patent standards change according to country, industry and process
and iii) companies frequently use alternative protection tools such as
industrial secret and leading time (Gorodnichenko, Svejnar and Terrell,
2008).
At the least, we should note that besides many formal and informal
apropriability mechanisms, such as patents and designing, advertisement1
is a protection option that is sometimes far more efficient than formal
ones.
In fact, if choosing between registering an innovation in a patent
office or showing it to as many potential buyers as possible, the second
option could be financially better than the first one. And once an
innovation is associated with a company, competitors will have extra
difficulty because when they imitate or create, they will need to
persuade potential buyers that their products are as good as or better
than those from companies which first innovated and advertised it.
Advertising may serve as a signal of product quality or R&D
effort; or it may be that both R&D and advertising are strategic
investments and thus affect each other. The relationship between
advertising, R&D, and market structure advertising increase
perceived quality (Shaked and Sutton, 1987).
3. DATA AND VARIABLES
Our data is taken from the Industrial Annual Survey (Pesquisa
Industrial Anual, PIA) and the Technological Innovation Survey (Pesquisa
de Inovacao Tecnologica, PINTEC), both produced by Brazilian Census
Office (Instituto Brasileiro de Geografia e Estatistica, IBGE). PIA is a
firm level industrial annual survey to Brazilian manufacturing and
PINTEC2 is a firm level technological innovation survey. PIA and PINTEC
are connectable through a common firm identification number. We match
PIA and PINTEC for years 2003 and 2005 editions and get a short
unbalanced panel with 14,000 firms.
TABLES 1 and 2 (that should be read through lines) show us that
among 14,379 firms average Market Share (MS) is 0.9%, with standard
deviation 3.9% and 75th percentile 1.4%. Average price cost-margin (PCM)
is 64%, with median 71%, standard deviation 23.7%, 5th percentiles 23%,
and 75th percentile 82%.
Finally, the average advertisement/net revenue ratio (ADV) is 0.3%
with median 0.02%, standard deviation 1.2% and 75th percentile 0.3%
To sum up, these descriptive statistics show us that market share
is lower than 1.5% for at least 75% of the firms in this sample, and at
least 50% of them have price cost-margin larger than 71%. They also show
us a significant dispersion for all variables described.
TABLE 3 (that should be read through columns) show us that among
five writing protection mechanisms, patents of invention (PI) is used by
6.22% of the firms in our sample, utility model patent (UMP) by 5.47%,
industry design register (IDR) by 4.98%, trade marks (TM) by 22.97% and
copyright (C) by 2.44%. Among the three strategic protection mechanisms,
design complexity (DC) is used by 2.59% of the firms in our sample,
industrial secret (IS) by 10% and leading time to competitors (LTC) by
5.74%. Other appropriability mechanisms, which each firm specifies
differently, are used by 2.8%.
A firm can at the same time register a patent, have a design
complex and spend on advertisement. Hence, it makes sense consider a mix
of appropriability methods (MAM). We create MAM qualitative variable
that is one to firms that used more than one appropiability mechanism,
include advertisement, and 0 on the contrary. 49.74% of the firms in
this sample used a mix of appropriability methods.
4. ECONOMETRIC MODEL AND RESULTS
4.1. Econometric model
Because our R&D dependent variable is a binary one, we use a
binary response model. Probit is the binary response models most
commonly used in applications. It is typically estimated by maximum
likelihood which has good properties in large samples. In particular, it
is asymptotically efficient (Horowitz and Savin, 2001).
Generally, a probit model can specified as
(1) P(Y=1|X) = G(X[[beta].sub.k]) = G([[beta].sub.0] +
[[beta].sub.1][X.sub.1]+ (...) + [[beta].sub.k][X.sub.k]), where G(.) is
a normal cdf.
The [[beta].sub.k] gives the signs of the partial effect of each
[x.sub.k] on the response probability; and the statistical significance
of [x.sub.k] is determined by whether we can reject [H.sub.0]:
[[beta].sub.k] =0. In sum, the signing of [[beta].sub.k] determines
whether the independent variables have a positive or negative effect on
the binary dependent variable (Wooldridge, 2002).
However, there is neglected heterogeneity problem. According to
Blundell, Griffith and Van Reenen (1999) the empirical relationship
between innovation and market share could be artificial because of
unobserved heterogeneity deriving from the different technological
opportunities and appropriability conditions facing firms and therefore
the association could be spurious because of poor data quality.
It is possible that market share and the particular types of
appropriability mechanisms may be related to unobserved forces that also
drive R&D as a specific kind of unobserved heterogeneity. For
example, the dynamics of market shares will depend crucially on the
speed and effectiveness of rivals' responses. In this respect,
different industries display widely differing characteristics. Many
patterns of interaction may emerge between a leader and its rivals, and
these patterns will reflect various factors, some of which (such as the
beliefs of rival firms, for example), are very difficult to measure,
proxy, or control for in empirical studies (Sutton, 2007). On the other
hand, the pattern of appropriability varies according to firm's
strategy. In some cases it is useful focus on writing mechanisms as in
pharmaceutical industry. In other cases it is better strategic
mechanisms as industrial secret in high technological sectors. But it is
always rational a mix of appropriability mechanism, advertisement
inclusive. The mix composition, however, is another source of unobserved
heterogeneity.
In fact, omitting variables which are not independent of the
included explanatory variables results in biased and inconsistent
estimated coefficients. The omitted variables set could have variables
such as management ability, technological opportunity and preference for
innovation.
The fixed effects probit analysis treats fixed effects as
parameters to be estimated along with p. But, treating fixed effects as
parameters to estimate lead to potentially serious biases. This
doesn't happen if we consider random effects (Wooldridge, 2002;
Artes, 2009). There is any specific solution available to unobserved
heterogeneity problem and probit random effects sounds the best option
in our case.
In the general form our specification is
(2) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
This can be re-written as
(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
[R&D.sub.it] is a dummy variable (3) to firm i at time t, which
is 1 if firms had expended on R&D and 0 otherwise.
[MS.sub.it] is market share (4), [MS.sup.2.sub.it] market share
squared, [LNPCM.sub.it-1] and [LNPCM.sub.it-2] are the first and second
lags of the log price-cost margin, [MAM.sub.it] is the mix of
appropriability mechanisms, [MS.sub.it]*[MAM.sub.it] is the market
share-mix of appropriability indicator interaction, and
[MS.sup.2.sub.it]*[MAM.sub.it] is a market share squared-mix of
appropriability interaction term.
[MAM.sub.it] allows us control for systematic firm differences in
appropriability on firm R&D decisions. [MS.sub.it]*[MAM.sub.it]
tells us if market structure and appropriability are (or not) mutually
independent in their relationship with industry R&D performance,
i.e., if it controls the market share effects on R&D firm decision
for systematic firm differences in appropriability.
The tension between the often-cited inverted-U hypothesis and the
diverse empirical results indicates that the available empirical
evidence is inconclusive, and thus much remains to be learned regarding
the relationship between market structure and industry R&D
performance (Cohen and Levin (1989), Lee (2005), Aghion et al (2005)).
[MS.sup.2.sub.it] provides support for the inverted-U hypothesis. And
[MS.sup.2.sub.it]*[MAM.sub.it] it allows us check if appropriability
strategies influence the inverted-U relationship. Last, Tit is the firm
random effect to avoid the neglected heterogeneity problem.
We expect that [[alpha].sub.1], [[alpha].sub.3], [[alpha].sub.4],
[[alpha].sub.5], [[alpha].sub.7] will have positive sign as market
share, lagged profit and appropriability mechanisms should all have a
positive effect on a firm's R&D decision. And [[alpha].sub.2]
and [[alpha].sub.6] could be positive or negative as either U or and an
inverted-U market share-R&D relationship are possible.
As Artes (2009) remarks, it has been argued that companies that
perform R&D are more likely to obtain higher profits in the future,
and market structure will shift accordingly. If innovation and market
structure are simultaneously determined, the estimation suffer from an
endogeneity problem. Previous literature that tested endogeneity between
market structure and innovation is contradictory. The problem is
difficult to address because there are lags between successful
innovations, and the accumulation of profits is able to affect the
monopoly power of the company. Furthermore, it seems plausible that the
lags vary from firm to firm.
In our sample, firms are observed during a short period of time
over which market conditions are unlikely to change in response to
firms' innovative activities. In that sense the panel being short
could alleviate the possible endogeneity.
A traditional interpretation of the innovation-market power
correlation is that failures in financial markets force firms to rely on
their own supra-normal profits to finance the search for innovation. The
availability of internal sources of funding are useful for all forms of
investments, but may be particularly important for R&D. External
sources of finance may be more expensive (due, for example, to
asymmetric information) and there is the risk that rival firms could
acquire valuable information if a firm seeks external funding for its
innovation projects (Blundell, Griffith and Van Reenen, 1999).
Our lagged profits capture the fact that past profits can help
finance innovation and also helps to avoids endogenity problem,
especially because of the simultaneity between R&D expenditure and
profitability.
4.2 Regressions results
TABLE 4 show us that market share, lagged profits and mix of
appropriability mechanisms (MAM) increase R&D likelihood. There is
also a non-linearity R&D-market share relation as [MS.sup.2 ]has
negative impact on R&D decision. Market share and MAM is not
significant. And market share square and ad interaction keeps sign.
Those sings are according to expected.
Those regression results suggests that 1) mix of appropriability
mechanisms efficient as R&D protection efforts among Brazilian
industrial firms. This makes sense since firms use a intellectual
protection portfolio, especially useful in countries with low legal
enforcement (Hall and Ziedonis (2001), Cohen, Nelson and Walsh (2000),
Gorodnichenko, Svejnar and Terrell (2008)), 2) market share affects
positively firm's R&D decision, i.e., market concentration and
R&D expenditure are positively correlated in low appropriability
sectors, which suggests that market concentration supplements low
R&D appropriability or that technological competence (or
opportunities) may not be fully exerted (or exploited) when market
structure remains atomistic (Lee, 2005), 3) there are non-linearities,
with is according to Cohen and Levin (1989), Lee (2005), Aghion et al
(2005), 4) lagged profits increase R&D likelihood, probably because
of failures in financial markets as Blundell, Griffith and Van Reenen
(1999) suggests.
5. CONCLUSION
This study presents empirical evidence on the relationship between
R&D, market structure and a mix of appropriability mechanisms in
Brazilian manufacturing firms. Our results suggest that market share,
lagged profits and mix of appropriability mechanisms (MAM) increase
R&D likelihood. There is also a non-linearity R&D-market share
relation. Those results are according to a couple of recent studies, as
Hall and Ziedonis (2001), Cohen, Nelson and Walsh (2000), Gorodnichenko,
Svejnar and Terrell (2008), Cohen and Levin (1989), Lee (2005), Aghion
et al (2005) and Blundell, Griffith and Van Reenen (1999).
In sum, our results about R&D and market structure not only
contribute to the recent debate about this subject, but also highlight
the effect of mix of appropriability mechanisms besides patents,
especially for an important emerging economy such as Brazil.
REFERENCES:
Aghion, P et alli Competition and innovation: an inverted
U-relationship. Quarterly Journal of Economics, vol 120, 2005, p.
701-728
Arora, A; Ceccagnoli, M; Cohen, WM. R&D and patent premium.
International Journal of Industrial Organization, vol 26, 2008, p.
1153-1179
Arte, J Long-run versus short-run decisions: R&D and market
structure in Spanish firms. Research Policy, vol38, 2009, p.120
Bagwell, K. The economic analysis of advertising. In: Armstrong,
M.; Porter, R. Handbook of industrial organization. Amsterdan: Elsevier,
2007. p. 1701-1844.
Blundell, R; Griffith, R; Reenen; J. Market Share, Market Value and
Innovation in a Panel of British Manufacturing. The Review of Economic
Studies, Vol. 66, No. 3 (Jul., 1999), pp. 529-554
Caves, R.E. e Porter, M.E. Market Structure, oligopoly, and
stability of market shares. The Journal of Industrial Economics, v. 26,
n. 4, p. 289-331, June 1978
Cohen, MW Fifty years of empirical studies of innovative activity
and performance. In Hall, BH and Rosemberg, N. Handbook of the Economics
of Innovation. Amsterdan: North-Holland, 2010. p. 129-213
Cohen, M.W; Nelson, R.R and Walsh, J.P Protecting their
intellectual assets: appropriability conditions and why U.S.
manufacturing firms patent (or not), NBER Working Paper 7552, February,
2000
Cohen, W.M. and Levin, R.C. Empirical Studies of innovation and
market structure. In: Schmalensee, R. and Willig, R.D. Handbook of
Industrial Organization. Amsterdan: Elsevier, 1989. p. 1059-1107
Gorodnichenko, Y.; Svejnar, J. and Terrell, K. Globalization and
innovation in emerging markets, NBER Working Paper 14481, 2008
Hall,B.H. and Ziedonis, H.R. The patent paradox revisited: an
empirical study of patenting in the U.S. semiconductor industry,
1979-1995, The RAND Journal of Economics, Vol. 32, No. 1, Spring 2001,
pp. 101-128
Horowitz, J.L; Savin, N.E. Binary Response Models: Logits, Probits
and Semiparametrics. Journal of Economic Perspectives,Volume 15, Number
4,Fall 2001,Pages 43-56
IBGE. Pesquisa Industrial Anual, PIA. 2001, 2002, 2003, 2004 and
2005 editions.
IBGE. Pesquisa de Inovacao Tecnologica, PINTEC. 2003 and 2005
editions.
Lee, C. A new perspective on industry R&D and market Structure,
The Journal of Industrial Economics, Volume LIII, March 2005
Mortimer, J.H. Price discrimination, copyright law, and
technological innovation: evidence from the introduction of DVDs, The
Quarterly Journal of Economics,Vol.122(3) August 2007, p. 1307-1350
Scherer, F. M. Firm Size, Market Structure, Opportunity, and the
Output of Patented Inventions. The American Economic Review, Vol. 55,
No. 5, Part 1 (Dec., 1965), pp. 1097-11
Schmalensee, R., Inter-industry studies of structure and
performance, In: Schmalensee, R.; Willig, R. D. Handbook of industrial
organization. Amsterdan: Elsevier, 1989. p. 951-1009.
Shaked, A; Sutton, J. Product Differentiation and Industrial
Structure. The Journal of Industrial Economics, Vol. 36, No. 2. (Dec.,
1987), pp. 131-146
Sutton, J. Market Share Dynamics and the "Persistence of
Leadership" Debate. The American Economic Review, Vol. 97, No. 1,
pp. 222-241, 2007
Wooldridge, J.M. Econometric Analysis of Cross Section and Panel
Data, MIT press, 2002
Gilson Geraldino Silva-Jr, Catholic University of Brasilia, Brazil
(1) For an in depth economic analysis of advertisement, see Bagwell
(2007).
(2) PINTEC is similar to Spanish Survey of Business'
Strategies, but more detailed. For example, PINTEC has information about
appropriability mechanisms beyond patents.
(3) Artes (2009) also uses an R&D dummy as dependent variable
in a recent R&D-market structure study to Spain.
(4) Market share measures market concentration and firm size at the
same time. See Caves and Porter (1978) and Schmalensee (1989).
TABLE 1--continuous variables descriptive statistics
Variable Observations Average Standard Deviation
MS 14379 0.009 0.041
[MS.sup.2] 14379 0.0018 0.023
PCM 14379 0.64 0.237
ADV 14379 0.003 0.012
Source: Our tabulation from 2003 and 2005 PIA and PINTEC surveys
TABELA 2--continuous variables percetiles
Variable P5 P25 P50 P75 P95
MS 0.0001 0.0007 0.0034 0.014 0.11
[MS.sup.2] 0 0 0 0.0002 0.012
PCM 0.23 0.57 0.71 0.82 0.93
ADV 0 0 0.0002 0.003 0.028
Source: Our tabulation from 2003 and 2005 PIA and PINTEC surveys
TABLE 3--firms that used appropriability protection mechanisms
(%) PI UMP IDR TM C DC
Yes 6.22 5.47 4.98 22.97 2.44 2.59
No 93.78 94.53 95.02 77.03 97.56 97.41
Total 100.00 100.00 100.00 100.00 100.00 100.00
(%) IS LTC others MAM
Yes 9.99 5.74 2.80 49.74
No 90.01 94.26 97.20 50.26
Total 100.00 100.00 100.00 100.00
Source: Our tabulation from 2003 and 2005 PIA and PINTEC surveys
Table 4 : R&D, mix of appropriability and market structure--probit
panel regressions
DR&D
CONSTANT -2.26(0.072) ***
MS 10.42(1.80) ***
[MS.sup.2] -6.85(2.72) ***
[LNPCM.sub.t-1] 0.41(0.095) ***
[LNPCM.sub.t-2] 0.304(0.09) ***
MS * MAM 2.91(1.86)
[MS.sup.2] * MAM -7.967(2.868) ***
MAM 1.22(0.062) ***
LogLikelihood -6063.90
Test [X.sup.2]
All variables [X.sup.2] (7)=1164.20 ***
MS, M[S.sup.2] [X.sup.2] (2)=49.21 ***
MS * MAM, [MS.sup.2] * MAM [X.sup.2] (2)=9.40 ***
MS * MAM, [MS.sup.2] * MAM, MAM [X.sup.2] (3)=471.96 ***
Observations 14379
Source: Own tabulation from 2003 and 2005 PIA and PINTEC surveys
***, **, * means 1%,5% and 10% significance level, respectively
Note: DR&D is R&D dummy variable (1 if firm spent on R&D, 0 on
the contrary), MS is market share, [MS.sup.2] is market share
LNPCM.sub.t-1], [LNPCM.sub.t-2] is price cost margin log lag 1 and 2.
MAM is a mix of appropriability mechanisms dummy variable, which
includes advertisement (it is one of a firm used one or more
appropriability mechanism, inclusive advertisement, and zero on the
contrary). MS * MAM is market share- mix of appropriability indicator
interaction, [MS.sup.2] * MAM is market share squared- mix of
appropriability indicator interaction.