G Model STRECO-454;

No. of Pages 14

ARTICLE IN PRESS
Structural Change and Economic Dynamics xxx (2011) xxxxxx

Contents lists available at ScienceDirect

Structural Change and Economic Dynamics

journal homepage: www.elsevier.com/locate/sced

Innovation strategies, process and product innovations and growth: Firm-level evidence from Brazil
Micheline Goedhuys a , Reinhilde Veugelers b,c,d,e,
a

UNU-MERIT (United Nations University, Maastricht Economic and Social Research and Training Centre on Innovation and New Technologies) and Maastricht University, The Netherlands b University of Leuven, Belgium c ECOOM, Belgium d Bruegel, Bruxelles, Belgium e CEPR, London, United Kingdom

a r t i c l e

i n f o

a b s t r a c t
Using World Bank ICS 20002002 data from Brazilian manufacturing rms, this paper identies innovation strategies of rms in particular internal development (technology make) and external acquisition (technology buy) and their effect on successful process and product innovations. It subsequently explores the importance of process and product innovations for rm growth. Successful process and product innovations occur mostly through technology buy (mostly through the purchase of machinery and equipment), either alone or in combination with a technology make strategy. The option of only relying on internal development is less successful. The results on rm growth indicate that innovative performance is an important driver for rm growth. It is particularly the combination of product and process innovations that signicantly improves rm growth. Both innovation and growth performance are supported by access to nance. Skills of workforce and management matter, but not necessarily tertiary education levels. The impact of international linkages on innovative and growth performance is mixed. 2011 Elsevier B.V. All rights reserved.

Article history: Received January 2010 Received in revised form January 2011 Accepted January 2011 Available online xxx JEL classication: O12 O31 O54 Keywords: Make and buy innovation strategies Process and product innovations Firm growth Brazil

1. Introduction Technological progress is at the heart of development. But the innovation processes that underlie technological progress in developing countries have very different characteristics from that in developed countries. Innovation in these countries largely occurs through the absorption, adaptation and mastery of already elsewhere developed technologies, often sourced from abroad, rather than the invention of entirely new technologies. Clearly there is a huge heterogeneity across developing countries and rms in the importance of innovation for growth. Among the factors identied as potential barriers

Corresponding author at: KULeuven, MSI, Naamsestraat 69, B-3000 Leuven, Belgium. Tel.: +32 16326908. E-mail address: reinhilde.veugelers@econ.kuleuven.be (R. Veugelers). 0954-349X/$ see front matter 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.strueco.2011.01.004

for growth and the adoption of new technologies are heavy regulatory burden, the quality of institutions, severe nancial constraints and macro-economic uncertainty (Bastos and Nasir, 2004; Dollar et al., 2003; Eifert et al., 2005). In addition, technology diffusion depends on the extent to which rms are exposed to foreign and new vintage technologies through trade, FDI and migration of human capital and on the rms ability to absorb and adapt the technologies to which they are exposed. A number of studies on innovation activities of rms in particular industries in Latin America and elsewhere in the developing world (e.g. Bell and Pavitt, 1995) stress the importance of investment in domestic capabilities to generate and manage change in technology. These capabilities encompass specialised resources such as skilled labour. To shed more light on the link between adoption of new technologies, innovation and rm growth, this paper analyses micro-evidence from a sample of manufacturing rms

Please cite this article in press as: Goedhuys, M., Veugelers, R., Innovation strategies, process and product innovations and growth: Firm-level evidence from Brazil. Struct. Change Econ. Dyn. (2011), doi:10.1016/j.strueco.2011.01.004

G Model STRECO-454; 2 No. of Pages 14

ARTICLE IN PRESS
M. Goedhuys, R. Veugelers / Structural Change and Economic Dynamics xxx (2011) xxxxxx

from Brazil. In recent years, Brazil has been able to play an increasingly important role in international trade, production and innovation. While this pattern of emerging success has given rise to a signicant amount of research documenting the performance of the economy at the aggregate, the micro-evidence on the factors that underlie Brazilian rms success remains less abundant, especially when it comes to studies on larger samples, including various industries, and controlling for the variety of factors that can inuence rm performance. This paper, using rm level data from the World Banks Investment Climate Survey (ICS) collected in Brazil in 2003, contributes to the literature on innovation and development in several ways. Taking a micro-econometric perspective, the paper investigates in more detail not only how the interplay between external technology sourcing (a technology buy strategy) and internal technology development (a technology make strategy) affects innovation outcomes, but also how the successful introduction of new products and processes in turn drives rm growth. Both the innovation and the growth analysis also consider the importance of other framework conditions like access to nance, skills, ICT, (international) openness and competition. Our focus on Brazil allows analysing the case of a middle-income country in fast development, where innovation strategies are becoming increasingly more important for sustaining growth, particularly the interaction between technology buy and technology make and the interaction between product and process innovations. Also on the role of foreign technologies, trade and FDI, Brazil is an interesting case. Although it has opened up to the global economy, it still carries the reminders of a closed economy set-up, relying on its own large internal market. Our study takes on board many of these inuences to study what drives or impedes Brazilian manufacturing rms to introduce new products and processes successfully and grow. Before we present our results in Section 5, we rst briey review several strands of the literature (Section 2), discuss the specics of Brazil (Section 3) and introduce our data (Section 4). 2. Literature review Several strands of the literature are relevant for understanding the link between innovation and rm growth in developing countries: the macro-economic literature on innovation and growth, the literature on FDI, technology spillovers and growth, and the micro-literature explaining rm growth. It is widely recognised in the macro-economic literature that R&D and innovation are a major driver of economic growth. Since the classic Solow residual paper (Solow, 1956), it has been recognised that rates of factor accumulation do not account for the major part of economic growth. The endogenous growth literature (see Romer, 1994; Grossman and Helpman, 1991; Aghion and Howitt, 1998) identies innovation efforts as a major engine of growth. In these models, the rate of growth of a country is determined by its initial level of development, the creation of new knowledge within the country and the

absorption and exploitation of knowledge, independently of where it is created. Spillovers from technology elsewhere developed, have been identied as particularly important drivers in endogenous growth models (a.o. Grossman and Helpman, 1991). However, knowledge does not spill over easily and costlessly, but rather depends on countries ability to effectively absorb new technologies (Lall, 1992). Accessed knowledge needs to be combined with a sufciently developed absorptive capacity (Cohen and Levinthal, 1989) or social capability (Abramovitz, 1986) in order to deliver growth. Absorptive capabilities depend on many factors, including the extent to which a country has a technologically literate workforce and a highly skilled elite; promotes an investment climate that encourages investment and permits the creation and expansion of rms using highertechnology processes and products; permits access to capital; and has adequate public sector institutions to promote the diffusion of critical technologies where private demand or market forces are inadequate (World Bank, 2008). A second set of factors explaining technological progress is own indigenous innovative capacity, which becomes increasingly important as a country progresses closer to the technology frontier (Hoekman et al., 2005). First, own R&D complements the adoption of existing technology because it is a component of absorptive capacity. Foreign technologies frequently need to be modied so that they are suitable for domestic circumstances. Countries tend to acquire technology more readily when domestic rms have R&D capabilities and when public research laboratories and universities have relatively close ties to industry. But, at higher levels of development, own R&D increasingly may also start to substitute adoption of existing technologies, allowing generation of new technologies, particularly in these areas where the country has developed comparative strengths. Empirical macro-analysis conrms the importance of innovation for catching-up (Nelson, 1993; Kim, 1997). Fagerberg et al. (2007) and Fagerberg and Srholec (2008) for a large cross-section of countries nd signicant effects of technological capacity (both creation and absorption) to be signicantly related to growth. But although a well functioning innovation system seems critical for development, they also conrm the importance of the quality of a countrys institutions. For catching-up countries, international transfer of technology, through trade (exports and imports) or FDI, is an important source for growth. Many endogenous growth models have emphasized technology spillovers from the North to the South as a vehicle for productivity growth of the South (e.g. Grossman and Helpman, 1991; Coe et al., 1997). However, the empirical evidence on the effects of international technology transfer is less clearcut (Hoekman and Smarzynska Javorcik, 2006). Although earlier studies based on industry level cross-sectional data found statistically signicant horizontal spillover effects in developing countries (Blmstrom and Persson, 1983; Blmstrom, 1986; Kokko, 1994), more recent studies using panel data sets, correcting for rm or sector specic xed effects, nd no positive within-industry spillover effects

Please cite this article in press as: Goedhuys, M., Veugelers, R., Innovation strategies, process and product innovations and growth: Firm-level evidence from Brazil. Struct. Change Econ. Dyn. (2011), doi:10.1016/j.strueco.2011.01.004

G Model STRECO-454; No. of Pages 14

ARTICLE IN PRESS
M. Goedhuys, R. Veugelers / Structural Change and Economic Dynamics xxx (2011) xxxxxx 3

for developing countries (e.g. Grg and Greenaway, 2003). Using macro data, no or little support is found for openness to trade and foreign direct investment to matter for innovation and growth (Fagerberg and Srholec, 2008). One explanation for the difculty to nd evidence of positive spillovers from openness is the confounding impact of competitive effects from open markets (Markusen and Venables, 1999). In addition, the potential benets from FDI may not materialize, as multinational rms may protect their core know-how from dissipating to local rivals (Veugelers and Cassiman, 2004). An additional critical factor to exploit spillovers is the technological capability of indigeneous rms (Blomstrm and Kokko, 1998; Blalock and Gertler, 2009). Most of the empirical studies on developing countries have failed to nd robust evidence of positive knowledge spillovers from multinational investment, accounted for by the lack of absorptive capacity in these host countries (e.g. Aitken and Harrison, 1999; Narula and Dunning, 2000). The literature explaining rm growth has shown that growth is largely a stochastic process where many unidentied and unobservable (rm-specic) factors are responsible for the growth performance of rms. However, since the seminal work of Jovanovic (1982) rm growth is seen as a learning process: rms discover their true efciency levels and adjust their size accordingly, with more efcient rms growing into a larger size. This learning process is most apparent shortly after entry, which explains why small and young rms grow faster. Pakes and Ericson (1998), building further on Jovanovic (1982), developed an active learning model, in which the efciency level can be actively raised by rm-specic investments in innovation activities and R&D, thereby opening up the growth perspectives of rms. Several authors (e.g. Katz, 1987; Enos, 1992; Lall, 1992; Bell and Pavitt, 1995; Kim, 1997) have integrated these ideas with the literature on technology for catching up rms in developing countries, using an evolutionary perspective of economic change (Nelson and Winter, 1982). These authors underscore the importance of investment in domestic capabilities to generate and manage change in technology. In the absence of domestic technological capabilities specialised resources and skilled labour efciency gains and rm growth will not automatically follow from the acquisition of foreign technology and know-how embodied in machinery (Bell and Pavitt, 1995). Firm level technological change is understood as a continuous process, to absorb or create technical knowledge, determined partly through the acquisition of external inputs through FDI, licenses, know-how and technical service agreements, capital goods imports and partly through accumulating skills and knowledge by training human capital through formal education, on the job training, experience and in house R&D efforts to manage technological change. Recently a number of empirical studies emerged analysing innovation in developing country rms, its determinants as well as its impact on other rm performance indicators (see Raffo et al., 2008 for an overview). Using mainly innovation survey data, these studies follow Crpon et al. (1998) or variations thereof, explaining rm performance, mostly productivity, as a function of product and/or

process innovation, which in turn are explained by R&D and other innovation expenditures.1 For instance, Chudnovsky et al. (2006) nd in-house R&D and technology acquisition expenditures to enhance the probability of product and/or process innovations, which in turn attain higher productivity levels than non-innovators. The impact of innovation on rm growth is studied by Benavente and Lauterbach (2008) for Chile. They nd product innovations to stimulate employment growth, but nd no relationship between process innovation and growth. Equally interesting is the paper by Almeida and Fernandes (2008) who analyse innovation and international exposure, based on rm data from 43 developing countries, and nd importing and exporting rms, but not foreign owned rms, to be more innovative; they do not measure the further impact on rm performance. Raffo et al. (2008) equally do not nd a positive effect of foreign ownership on innovativeness, but a significant positive one on productivity. Overall, the literature paints a complex relationship between technology development (technology make) and the acquisition and absorption of externally developed (foreign) technologies (technology buy) and growth. At the macro-level, R&D and innovation seem important for development, but are no panacea for success. Depending on the initial country conditions, anking conditions such as education, nance, quality of institutions, governance and openness, need to be factored in. At the rm level, linking innovation strategies and growth requires taking into account the conditions for effective technology creation and absorption. 3. The case of Brazil As our analysis uses 20002002 rm level evidence from Brazil, this section characterizes the country specic setting of the data, particularly with respect to its innovation position, its development process and the framework conditions for innovation and growth. With the largest population in Latin America and the Caribbean, Brazil has reached in the past few years important economic, social and environmental advances (OECD, 2006). The country has expanded beyond agricultural commodities and low value-added manufacturing, to become a global competitor with a signicant group of rms being inserted in international markets via medium and high-tech goods (De Negri and Turchi, 2007). Through joint ventures with multinational corporations (MNCs), the country also has developed information and communications technology (ICT) and software industries that serve domestic and regional markets. However, in other hightech, high-growth sectors, Brazil is falling behind global competitors, particularly Asian. Brazil has made great strides in developing its own innovation system. Brazil is Latin Americas largest and most innovative economy, with about 1% of its GDP going to R&D in 20002003 (Lugones and Suarez, 2007, p. 156). However,

1 Examples from Latin America include the analysis of countries such as Argentina (Chudnovsky et al., 2006), Chile (Benavente, 2006), Mexico (Prez et al., 2005) and Argentina, Brazil and Mexico (Raffo et al., 2008).

Please cite this article in press as: Goedhuys, M., Veugelers, R., Innovation strategies, process and product innovations and growth: Firm-level evidence from Brazil. Struct. Change Econ. Dyn. (2011), doi:10.1016/j.strueco.2011.01.004

G Model STRECO-454; 4 No. of Pages 14

ARTICLE IN PRESS
M. Goedhuys, R. Veugelers / Structural Change and Economic Dynamics xxx (2011) xxxxxx

most of Brazils R&D is coming from the public sector, with the private sector R&D expenditure as share of revenue and numbers of R&D workers being still low2 (Peirano, 2007) and advancing on a specialisation path that requires little investment in R&D. A rst barrier to effective R&D and innovation performance in Brazil is a shortage of skills, especially in the elds of natural and exact sciences and engineering, needed for the technological development of technology intensive industries (Lugones and Suarez, 2007). Most of the recent expansion in the number of higher education degrees awarded is accounted for by private institutions, where quality concerns are present. A second barrier is access to nance. Despite macroeconomic stabilisation since the mid nineties, the cost of capital still remains high in Brazil thwarting the innovation efforts of small and medium sized rms (Prochnik and Dias de Araujo, 2007). The government has instituted several nancial incentives (De Brito Cruz and de Mello, 2006). Many are targeted at specic industries, such as aerospace and IT, concentrated in specic regions. During the economic recessions and crises of the 1990s, funding for science and technology programmes was severely cut (Cassiolato et al., 2003). The funding for innovation that was available to rms was directed to purchase capital equipment rather than for R&D. Another factor limiting the innovation potential is ICT usage. The Brazilian telecommunications equipment industry was initially a protected infant industry in Brazil. Trade barriers were erected to allow the industry room to grow. Government-sponsored R&D efforts in communications hardware and software helped to create an indigenous sector that was capable of serving the local market. Nevertheless, the use of ICT technologies is less widespread in Brazil than in countries with comparable development (Szapiro, 2003). With its history of import substitution, various sectors of the Brazilian economy remain sheltered from international competitive pressure (OECD, 2006). While the country has lowered barriers that were left over from its previous import substitution policies, it has moved to protect and foster those technology sectors that thrived under the barriers. This has included regulations on joint ventures with multinational corporations and technology licensing agreements meant to support innovative capacity building within Brazil. Despite these limitations and based on observations from innovation surveys, Arbix and De Negri (reported in De Negri and Turchi (2007)) nd strong indications that the opening up of the economy since the mid nineties has gradually led to a more entrepreneurial stance, that has improved global competitiveness of Brazilian industry. They nd innovative and product differentiating rms to combine product and process innovation more often than rms producing standardised goods; and they actively search information required for innovation abroad through FDI and alliances.

Overall, the recent fast development of Brazil provides an interesting case to examine the interplay between diffusion (buy) and creation (make) of new technologies, as centrepiece for sustaining growth, with many of the anking conditions for an innovation-growth nexus critical. Our analysis examines the relationship between the technology make and buy strategies on the one hand and the innovation and growth performance on the other hand, together with its anking conditions, at the level of individual rms operating in manufacturing sectors in Brazil. More specically, we examine which innovation strategies are most conducive to a rms successful introduction of new products and processes: technology make, technology buy or a combination of both strategies? We expect for Brazilian rms the technology buy option to be particularly important, especially when combined with a technology make strategy to improve absorptive capacity; how important successful product and/or process innovations are for rm growth; the impact, as framework conditions in the innovative and growth performance, of human capital as a crucial component of absorptive capacity, nancial constraints and international exposure. 4. Data and methodology 4.1. Data Difculties related to data availability and the measurement of innovation typically hamper rm level econometric analysis of the link between innovation and growth for developing countries. To analyse the link between technology creation and adoption, innovation and rm growth, we use the World Banks Investment Climate Survey (ICS) data collected in Brazil in 2003. The survey collected data for the periods 2000, 2001 and 2002,3 through intensive interviews with owners and managers of rms. The data collection is part of a larger programme coordinated by the World Bank that implements Investment Climate Surveys in many countries using a harmonised master questionnaire. The objective of the ICS is to obtain rm level data that allow analysing the conditions for investment and enterprise growth in the country. As such, the many aspects of the business environment that inuence the investment decisions and performance of the rms were tackled, in a number of sub-questionnaires. A set of questions was asked on the history of the rm, the background of the entrepreneur and manager, the acquisition and status of equipment and technology, the rms human resource management, innovation activities, and institutional constraints to growth and investment. The survey provides more information on a wider variety of factors affecting technological innovation and rm perfor-

80% of Brazilian researchers carry out their activities within public institutions (universities or research centres).

3 Additionally, the sales value was also asked for the year 1997, which permits investigating sales growth over a longer period of time: 19972002.

Please cite this article in press as: Goedhuys, M., Veugelers, R., Innovation strategies, process and product innovations and growth: Firm-level evidence from Brazil. Struct. Change Econ. Dyn. (2011), doi:10.1016/j.strueco.2011.01.004

G Model STRECO-454; No. of Pages 14

ARTICLE IN PRESS
M. Goedhuys, R. Veugelers / Structural Change and Economic Dynamics xxx (2011) xxxxxx 5 Table 1 Composition of the sample. Small 1029 workers Food Textiles Clothing Leather Chemicals Machinery Electronics Auto-parts Furniture Total 15 21 95 26 12 43 12 16 76 316 Medium 3099 workers 33 27 203 78 32 61 39 42 136 651 Large 100+ 72 54 126 56 34 74 23 68 89 596 All rms

mance than the Brazilian Innovation Survey 19982000,4 which was restricted to a more limited and focused set of questions on innovation. The ICS dataset allows the construction of innovative strategies, along both a make and a buy option. The innovation outcome of the rms can be assessed through self-reported success in introducing new processes as well as products. The growth performance of the rms is based on the self-reported sales data by the respondents for the years 20002002. In addition, the dataset allows for a rich set of controlling factors. Beyond the typical rm size and age variables, as well as sector and regional classication, the dataset also allows identifying the catching-up position of the rm in the technology space (technology leading or lagging); the human capital of the rm (both its work force as its management), which can be characterized by secondary and tertiary education levels; the use of ICT; the nancial constraints rms face in doing business; and the exposure to international technology and competition, investigated through multiple channels: foreign ownership, exports, imports of components, competition from imports. The main disadvantage of the dataset is the selfreported character of the data, leading to a possible subjective bias and the restriction to a cross-section dimension only, prohibiting constructing the dataset as a panel. The ICS data set of Brazil contains information on 1642 manufacturing rms, which represent a random sample, stratied on the basis of size, sector and location. The rms are selected from nine manufacturing sectors.5 Within the selected sectors, the sample gives a fair representation of the total population with respect to the size and location dimension.6 More detailed information on the sample and sampling procedure can be found in World Bank (2005). Due to missing values for some of the key variables, the number of rms used in our analysis is reduced to 1563, distributed over the different size classes and sectors as shown in Table 1. 4.2. Methodology Using the ICS information, we investigate the rm, industry and regional characteristics that can explain a rms innovation performance and its impact on rm

120 102 424 160 78 178 74 126 301 1563

growth. Table A1 in Appendix A provides a description of all the variables used in the analysis. 4.2.1. Innovation strategies: make and/or buy We rst identify rms innovation strategies, on both the make and buy dimension. In the questionnaire, rms were asked about their major ways of acquiring new technology. Firms could report up to three important strategies. As observed in many other developing countries and in line with expectations, rms reported that the most important channel for new technology acquisition was by investing in machinery and equipment that embodied newer vintage technology, followed by in-house development of new technology.7 Hiring of key personnel came in third position. Other less frequently mentioned channels as major ways of acquiring new technology consisted of sourcing from a parent company, licensing technology from other rms, developing technology in collaboration with clients or suppliers, or universities, or from trade fairs, exhibitions and study tours. We concentrate on the two major innovation strategies, capturing the most frequently used channels: developing technology within the rm, i.e. the MAKE option versus the technology acquisition or BUY option. For the BUY option, we include acquiring new technology embodied in new machinery, in key personnel as well as licensing-in technology.8 We construct different exclusive categories for the rms innovation strategy (I): rms that only report inhouse development of technology (MakeOnly); rms that only do external technology acquisition embodied in machinery and key personnel (BuyOnly); and rms that report both own development activities and embodied technology acquisition as important channels for technology acquisition (Make&Buy). The reference category is

4 This survey was coordinated by the Brazilian Geographic and Statistical Institute (IBGE), and gathers information from 10,328 rms. Access to the data for research is restricted, in contrast to the World Bank ICS data, which are more accessible. Summary tables are available on the IBGE website, results of the survey data are discussed in De Negri and Turchi (2007). 5 Food industries (CNAE code 150), textiles (CNAE 170), clothing (CNAE 180), leather products (CNAE 190), chemical products (CNAE 240), machinery (CNAE 290), electronics (CNAE 320), auto-parts (CNAE 344), and furniture (CNAE 361). The CNAE (Classicaco Nacional de Atividades Econmicas) is the Brazilian national classication which is closely linked to the ISIC (International Standard Industrial Classication) revisions 3 and 3.1, at least up to the 2-digit level. 6 A majority of rms is from So Paulo and Minas Gerais, but in total 13 states are covered by the sample. These are Rio de Janeiro, Santa Catarina, Rio Grande do Sul, Paran, Gois, Mato Grosso, Cear, Paraba, Maranho, Bahia, Amazonas.

7 This was also observed in the Brazilian Innovation Survey 19982000: acquisition of machinery and equipment accounts for half of innovation expenditures, followed by R&D, accounting for 25% of expenditures. These proportions differ from developed economies, where R&D is relatively more important, e.g. Germany, 55%; The Netherlands, 74% (Peirano, 2007). 8 We checked the robustness of our analysis to alternative compositions of the BUY strategy. Most notably, we also concentrated the BUY strategy on the buying of equipment only, which is the most commonly chosen BUY option. Most results (unless otherwise stated) are robust to the composition of the BUY strategy.

Please cite this article in press as: Goedhuys, M., Veugelers, R., Innovation strategies, process and product innovations and growth: Firm-level evidence from Brazil. Struct. Change Econ. Dyn. (2011), doi:10.1016/j.strueco.2011.01.004

G Model STRECO-454; 6 No. of Pages 14

ARTICLE IN PRESS
M. Goedhuys, R. Veugelers / Structural Change and Economic Dynamics xxx (2011) xxxxxx

rms that have no innovation strategy of developing technology nor buying technology (NoMake&Buy).9 4.2.2. Innovative performance: introducing new products and/or processes Having identied a rms innovative strategy, we then link the technology make or buy strategy to the outcome of the innovation process: the successful introduction of a new production processes that substantially changed the way the main product is produced (PROCESS) and/or the successful introduction of new products (PRODUCT), both over the period 19982002.10 Successful innovation both the successful introduction of new products and processes is modelled following a bi-probit model, which relates the probability of being a successful innovator to the characteristics of the rm (X) and the underlying innovation strategies (I). Our main variables of interest are the set of innovation strategies (I): MakeOnly, BuyOnly, Make&Buy, with NoMake&Buy being the reference category. Our set of other independent rm characteristics includes the typical size, age, sector and regional dummies. In addition, other anking variables that are likely to inuence innovative performance are included in the equation. These are the technology position of the rm in its market (T), the absorptive capacity in the rm (AC), foreign linkages and exposure (F), competitive pressure (C) and nancial constraints (FIN). They are measured through various proxies, shown in Table A1 in Appendix A. We thus estimate the following bi-variate probit model explaining successful process and product innovations jointly, using the same set of variables as regressors.:

As a robustness test, we also estimate a multinomial logit, where we compare the impact of the various innovation strategies on different exclusive categories of combined product and process innovation outcomes: process only innovation (PROCONLY), product only innovation (PRODONLY) and combined product and process innovation (PPI), versus no innovation outcomes, which is the base case. Compared to the bivariate probit model, the multinomial logit allows coefcients to vary across exclusive combinations of innovative performance. The multinomial logit model can therefore reveal the drivers that affect the joint successful implementation of product and process innovations. The disadvantage of the multinomial logit is that it requires an Independence of Irrelevant Alternatives assumption.11 It also requires that the number of categories may not be too large and that there is sufcient variation in each category. In our dataset a large majority of our observations is in the combined product and process category, leaving the exclusive outcomes only sparsely populated. The data further indicated that the category of product only innovators was very similar to the category of combined product and process innovators. We therefore had to merge both categories. Our nal multinomial logit specication thus includes three mutually exclusive categories: the process only innovators (PROCONLY), the product innovators, irrespective of whether or not this is combined with process innovations (PRODONLY + PPI), and the non-innovators, the latter being the reference group. We estimate the following multinomial model of innova-

PRODUCT = Xi + Ii + Ti + ACi + Fi + Ci + FINi + 1i i PRODUCTi = 1 if PRODUCT 0; PRODUCTi = 0 if PRODUCT < 0 i i PROCESSi = Xi + Ii + Ti + ACi + Fi + Ci + FINi + 2i PROCESSi = 1 if PROCESSi 0; PROCESSi = 0 if PROCESSi < 0 The launching of new products may often be combined with the introduction of new production processes, leaving product and process innovations as complementary (see also Miravete and Pernias (2006) for evidence on complementarity between product and process innovations for developed countries). The bi-variate probit regression allows taking into account the correlation between PROCESS and PRODUCT explicitly: E[
1]

(1)

tion performance outcomes: Prob(Y = j) = eZ

i j

3 i eZ k k=1

j {None(1), PROCONLY(2), (2)

PRODONLY + PPI(3)} Zi

= E[

2]

= 0,

Var[

1]

= Var[

2]

= 1,

Cov[

1, 2]

is a vector of rm characteristics, comprising where beyond the innovation strategies I also (X, T, AC, F, C, and FIN). 4.2.3. Growth performance Having analyzed the innovative strategies and innovation performance of rms in a rst step, the analysis will in a second step assess the relationship between innovation

9 These rms may still be acquiring technology but from less explicit innovation strategies than the ones included in the MAKE and BUY category, such as through sourcing information from or collaboration with clients, suppliers, universities, trade fairs, exhibitions and study tours. 10 Note that the available information on process and product innovations is in terms of successful introductions. Strictly speaking, the information can therefore only be interpreted as an innovation outcome variable, not as a innovation strategy choice for process versus product innovation. Two recent papers (Srholec and Verspagen, 2008; Bodas Freitas et al., 2008) propose, though in a different context, an alternative denition of rm strategies which address the nature of innovation output itself (product and process). In this case, product and process innovation choices would be the strategies to inuence performance, rather than the make and/or buy strategies.

11 The multinomial logit has the potential weakness inherent in the fact that choices between any two alternatives are assumed to be made independently of the remaining alternatives. This is problematic whenever some of the available alternatives are rather close substitutes in which case the merger of the substitutes is the most sensible solution (Kmenta, 1997, p. 259).

Please cite this article in press as: Goedhuys, M., Veugelers, R., Innovation strategies, process and product innovations and growth: Firm-level evidence from Brazil. Struct. Change Econ. Dyn. (2011), doi:10.1016/j.strueco.2011.01.004

G Model STRECO-454; No. of Pages 14

ARTICLE IN PRESS
M. Goedhuys, R. Veugelers / Structural Change and Economic Dynamics xxx (2011) xxxxxx 7

and rm performance. We use as measure of rm performance, self-reported average annual sales growth over the period 20002002.12 In line with the literature on rm growth using a learning perspective (Evans, 1987; Sleuwaegen and Goedhuys, 2002), the basic empirical model for the growth equation is a general growth function relating rm growth (g) to rm size (S) and rm age (A): g= St = g(St , A) St

where St and St are the size of a rm in end period t here 2002 and in beginning of period t 2000 respectively, and A is the age of the rm in 2003. This functional relationship is augmented with additional variables that are expected to be related to growth. Our major variables of interest are the various innovative performance indicators (IP). In addition, we add access to nance (FIN), foreign linkages (F) and absorptive capacity (AC) which are affecting growth, as anking conditions. We approximate the growth function g through a second order logarithmic expansion of a generalised function relating growth to size and age and we add the growth shifting variables. The resulting estimating equation corresponds to the following form: log(St ) log(St ) d = a0 + a1 log(St ) + a2 [log(St )]2 + a3 log(A) + a4 [log(A)]2 +a5 log(St ) log(A) + aI IP + a8 FIN + a9 F + a10 AC (3)

rm xed effects. Our data set does however not permit a panel data structure. We thus use a two step procedure in an attempt to improve our estimation while correcting for the potential biases due to endogeneity and unobserved heterogeneity. The two-step procedure uses the predicted values of the innovative performance as instrument in the growth regression (3), derived from the estimations (1) and (2). More specically, we include in the growth specication the predicted values PRODUCT and PROCESS from the biprobit estimation or the predicted exclusive categories (PROCONLY, PRODONL + PPI, and NOPPI) either derived from the bivariate estimation (1) or alternatively from the multinomial logit estimation (2).13 As an alternative, we also report the results where we relate rm growth directly to the innovation strategies (I: MakeOnly, BuyOnly, Make&Buy, and NoMake&Buy), bypassing the innovative performance outcomes. 5. The empirical analysis Before we present the results from the econometric analysis of rms innovative performance and growth in Section 5.2, we rst show some descriptive statistics on rms innovative strategies, innovative performance and growth in Section 5.1. 5.1. Descriptive statistics Table 2 summarizes the information about the rms innovation activities, more particularly who BUYs and/or MAKEs. The most frequently observed innovation strategy is the combination of Make&Buy (44%). The exclusive use of external acquisition embodied in new machinery, key personnel or through licensing (BUYonly) is the second most frequently observed observation (34%). The total proportion of sample rms that acquire technology externally (BUY) therefore sums up to 78%.14 Somewhat less but still more than half of the sample rms (56%) relies on internal development (MAKE). However, only 12% have a MakeOnly strategy. 10% of the sample rms report noMakeorBuy activities. These are rms that have no innovation strategy or resort to other less frequently observed and more informal ways for acquiring technology (such as trade fairs, exhibitions, contacts and cooperation with consultants, universities, clients, suppliers). Table 2 also shows the innovative performance of the different innovative strategies. For process innovations, two third of all rms report having successfully introduced new processes in the observed period. Companies without an innovation strategy are clearly less likely to introduce new processes (29%). Interesting to observe is the performance of the MakeOnly strategy, which is less effective as compared to an innovation strategy that uses embodied

where d stands for the number of years over which growth is measured and a are coefcient vectors. In line with this model, our dependent variable corresponds to an average annual sales growth rate over the period 20002002. The explanatory variables are as dened in Table A1 in Appendix A and include beyond innovative performance, the OVERDRAFT variable to consider nancial constraints (FIN); MNE and EXPORT for international linkages (F) and GMHIGH for the quality of human capital (AC). Our major variable of interest is the effect on sales growth of innovative performance, either through the PROCESS and PRODUCT dummy or the exclusive categories (PROCONLY, PRODONLY + PPI, and NOPPI). Simply including the innovative performance in the growth equation may result in their coefcients being biased, because of endogeneity: faster growing rms may have a larger incentive and capacity to introduce new innovations. Unfortunately our dataset does not allow a sufcient lagging of the innovative performance variable or forwarding of the growth results. Furthermore, unobserved rm-specic effects in the innovation performance regression can cause the coefcient of the innovation variables in the growth regression to be biased, if they also enter the growth error term, as an unobserved explanatory factor for growth as well. Panel data would allow including

12 Alternatively we also check robustness of our results using a longer time period for growth over the 19972002 period.

13 If we would use a perspective that interprets product and process innovations as innovation strategy choices, rather than innovation outcome variables, cf supra, our two-step procedure, can then be read as showing the impact of these strategies on growth performance, with the strategy choice being endogeneized to correct for potential biases. 14 This is consistent with the IBGE Brazilian innovation survey.

Please cite this article in press as: Goedhuys, M., Veugelers, R., Innovation strategies, process and product innovations and growth: Firm-level evidence from Brazil. Struct. Change Econ. Dyn. (2011), doi:10.1016/j.strueco.2011.01.004

G Model STRECO-454; 8 No. of Pages 14

ARTICLE IN PRESS
M. Goedhuys, R. Veugelers / Structural Change and Economic Dynamics xxx (2011) xxxxxx Frequency of occurrence 10% 12% 34% 44% 100% Process innovators 29% 60% 71% 73% 67.7*** 66% New-product innovators 34% 65% 69% 71% 31.6*** 66% Sales growth 20002002 -1.8% 1.7% 5.5% 6.9% 5.2%

Table 2 Frequency of occurrence of innovation strategies. Total No make/no buy Make only Buy only Make and buy X2
***

Signicance of X2 test at the 1%.

Table 3 Innovative performance and sales growth 20002002. Frequency of occurrence No new product/no process innovation Product and process innovators Product only innovators Process only innovators Process innovators Product innovators Total 222 (14%) Sales growth 20002002 (%) 2.5

788 (50%)

6.6

267 (17%) 286 (18%) 1074 (69%) 1055 (68%) 1563

4.1 4.3 6.0 6.0 5.2

technology acquisition, exclusively or in combination with internal development. There is some evidence in favor of complementarity between Make and Buy, since the highest success rate is observed for the Make&Buy strategy (73%). But this is only slightly higher than the performance of the BuyOnly strategy (71%). On average, 66% of all companies report having successfully introduced new products in the observed period.15 The lowest frequency is again observed among the NoMake&Buy category, where 34% of rms report to have introduced new products. Again, the highest frequency of successful product innovations is among the rms in the Make&Buy category, though the disparity with the other categories is smaller than for process innovation. The successful introduction of product and process innovations is highly correlated. Table 3 reports the frequency of occurrence of the exclusive categories of rms: product innovators alone (PRODONLY), process innovators alone (PROCONLY), combined product and process innovators (PPI) or none of both (NOPPI). Half of the rms (50%) introduce simultaneously product and process

innovations.16 These companies also report the highest sales growth over the considered period (6.6%), well above the growth performance of rms that introduce only product innovations (4.1%) or only process innovations (4.3%), and signicantly higher than the sample average growth rate of 5.2%. This is suggestive of complementarity between product and process innovations. Firms who failed to introduce new innovations (either product or process) have signicantly lower sales growth (2.5%). Similarly, rms without innovation strategies (NoMake&Buy) have a lower sales performance (see Table 2). This rst descriptive view of the data suggests rst the importance of embodied technology acquisition as an innovation activity (BUY), with support for complementing technology acquisition with in-house development (MAKE) in a Make&Buy innovation strategy; second, innovative strategies are positively related to innovative performance, especially for process innovation, where the BUY option (exclusively or in combination with MAKE) proves particularly favorable; third, innovative performance is positively associated with higher sales performance, particularly the combined productprocess innovations. The next sections will examine in a multivariate analysis which factors are signicant drivers of innovative and growth performance. 5.2. Multivariate analysis results on innovative performance In this section, we analyse the determinants of product and process innovations, as indicators of innovative performance. In Section 5.3, we analyse their impact on the sales growth of rms. Table 4 reports the results on innovative performance. Columns (1) and (2) contain the bi-probit results on process, respectively, product innovation. The marginal effects of the explanatory variables are reported, which for the dummy variables represents the change in the probability resulting from a change of the binary variable from 0 to 1. Columns (3) and (4) contain the results of the multinomial logit estimation, in marginal effects. The bi-probit estimations reveal a strong and signicant effect of innovative strategies on innovative

15 There was also information on whether rms had signicantly improved products during the same period. If we dene product innovating rms as those that introduced new or signicantly improved products, almost all companies 94% report themselves as product innovators, with very little difference between the different innovation strategies, with the exception of the NoMake&NoBuy strategy, which only has three quarters of these rms being able to introduce new or signicantly improved products. The high frequency of reporting introductions of improved products makes this variable not useful for analysis. This is why we concentrate the remainder of the analysis on innovative performance on process and new product innovations only.

16 Other Brazilian evidence, using the Brazilian innovation survey, found process innovations to be the most common and relevant form of innovations (De Negri et al., 2008).

Please cite this article in press as: Goedhuys, M., Veugelers, R., Innovation strategies, process and product innovations and growth: Firm-level evidence from Brazil. Struct. Change Econ. Dyn. (2011), doi:10.1016/j.strueco.2011.01.004

G Model STRECO-454; No. of Pages 14

ARTICLE IN PRESS
M. Goedhuys, R. Veugelers / Structural Change and Economic Dynamics xxx (2011) xxxxxx 9 Bi-probit PRODUCT(1) PROCESS(2) 0.010 (0.013) 0.030 (0.021) 0.139*** (0.043) 0.249*** (0.040) 0.295*** (0.044) 0.112*** (0.028) 0.072** (0.036) 0.060** (0.030) 0.044 (0.136) 0.117* (0.063) 0.067** (0.027) 0.025 (0.023) 0.087* (0.051) 0.094*** (0.030) 0.078 (0.051) 0.065 (0.072) 0.000 (0.033) 0.048* (0.026) 0.034 (0.038) 0.028 (0.042) 0.001 (0.028) 1563 Multinomial PROCONLY(3) 0.010 (0.012) 0.005 (0.019) 0.033 (0.058) 0.026 (0.049) 0.024 (0.046) 0.006 (0.025) 0.009 (0.028) 0.046* (0.025) 0.190 (0.131) 0.043 (0.051) 0.042* (0.022) 0.005 (0.020) 0.016 (0.039) 0.031 (0.022) 0.023 (0.044) 0.037 (0.059) 0.049 (0.026) 0.007 (0.021) 0.021 (0.031) 0.060* (0.032) 0.002 (0.023) 1563 PRODUCT (=PRODONLY + PPI)(4) 0.009 (0.014) 0.013 (0.022) 0.038 (0.060) 0.105** (0.052) 0.135*** (0.052) 0.058** (0.028) 0.002 (0.034) 0.080*** (0.029) 0.265* (0.151) 0.107* (0.061) 0.009 (0.027) 0.026 (0.024) 0.100** (0.051) 0.028 (0.028) 0.005 (0.050) 0.085 (0.082) 0.040 (0.032) 0.034 (0.025) 0.053 (0.036) 0.094** (0.039) 0.037 (0.028) 1563

Table 4 Results of the bi-probit and multinomial logit analyses explaining process and product innovation.

LTL00 LFIRMAGE MAKEONLY BUYONLY MAKEANDBUY TECHLEADER TECHLAGGARD OVERDRAF LFHIGH LFSEC GMHIGH LTOTEXPER ICT TRAINING FORFIRMEXPER MNE EXPORT FOREIGNSUPPLIER COMPIMP COMPPRESSPF COMPPRESSPD # observations

0.009 (0.014) 0.013 (0.022) 0.116** (0.047) 0.184*** (0.044) 0.203*** (0.046) 0.061** (0.029) 0.000 (0.034) 0.084*** (0.030) 0.275* (0.050) 0.110* (0.063) 0.010 (0.027) 0.026 (0.024) 0.102** (0.051) 0.030 (0.029) 0.004 (0.050) 0.088 (0.077) 0.048 (0.033) 0.038 (0.026) 0.054 (0.038) 0.101** (0.041) 0.037 (0.029) 1563

Marginal effects reported. Standard errors in parentheses; the estimation includes eight sector dummies and 12 geographical location dummies. * Signicant at 10%. ** Signicant at 5%. *** Signicant at 1%.

performance. All innovative strategies, MakeOnly, BuyOnly and Make&Buy, have signicantly positive coefcients, a result which appears robust across alternative specications. The marginal effects indicate a lower performance of the MakeOnly strategy as compared to the BuyOnly and Make&Buy strategy. The combination of Make&Buy has the highest effect on the probability that the rm is a successful product or process innovator. This can be interpreted as evidence in favor of complementarity, but the difference with the BuyOnly strategy is only minimal. Overall the data seems to suggest that for Brazilian manufacturing rms, it is particularly the BUY strategy that is most associated with innovative performance, particularly for being successful in introducing process innovations. This result is robust to a more narrow denition for BUY, conned to buying tech-

nology embodied in equipment only. It compares well to the ndings of Peirano (2007) who concludes on the basis of the Brazilian innovation survey that the acquisition of machinery and equipment predominates the innovation activities and nds the absorption of technology to prevail over the creation of new knowledge, especially for process innovations. Technology leading rms have a signicantly higher likelihood of successfully introducing new processes and products (+11% and +6%, respectively), while technology lagging rms have a signicantly lower likelihood of introducing process innovations (7%). As this variable reects subjective self-reporting, correlated with innovativeness, we also checked the estimation without these variables on technology position. The results for the other determin-

Please cite this article in press as: Goedhuys, M., Veugelers, R., Innovation strategies, process and product innovations and growth: Firm-level evidence from Brazil. Struct. Change Econ. Dyn. (2011), doi:10.1016/j.strueco.2011.01.004

G Model STRECO-454; 10 No. of Pages 14

ARTICLE IN PRESS
M. Goedhuys, R. Veugelers / Structural Change and Economic Dynamics xxx (2011) xxxxxx

ing variables, especially the innovation strategy variables, remain unaffected and robust to the specication with or without technology position. Financial constraints are a signicant factor in the innovation process. Companies that are less nancially constrained, having overdraft facilities, are more likely to be successful innovators. The effect from nancial constraints is stronger for product innovation (+8% against +6% for process innovation). This is also found by Prochnik and Dias de Araujo (2007) on the Brazilian innovation survey data. On the different measures for absorptive capacity, the results suggest that their effect depends on the type of innovation. For process innovation it is not the proportion of employees with tertiary education that matters, but rather the share of secondary education workers and even more signicantly the provision of formal training to them. This is reminiscent of the problems in Brazils higher education system associated with being able to supply quality skills in particularly sciences and engineering. A general manager with graduate or postgraduate education even adds negatively to innovative process performance. Product innovation on the contrary, appears to be more high-skill intensive, with larger shares of highly educated employees (secondary, but also tertiary), raising the probability of being a product innovator. Raising the proportion of tertiary educated workers from zero to 20 for instance increases the probability that the rm is a product innovator by 5%.17 The use of ICT (as measured through e-mail connection) also adds to the probability of successfully introducing new products to the market (+10%), as well as new process innovations (+9%). The variables on international linkages, as well as on competition, are weak in explaining innovative performance, in various alternative specications. Only competitive pressure from foreign rms has a positive and signicant effect on the likelihood that rms develop new products, raising it by 10%. Having foreign ownership (MNE) or foreign sales (DIRECTEXPORT), or facing competition from imports, seems unrelated to innovative performance. Sourcing inputs from a foreign supplier raises the likelihood that a rm introduces new production processes, but the effect is only signicant at the 10% level. This weak effect of global market forces may be specic for Brazil, suggestive of the fact that still an important share of rms remains sheltered from global market inuences reminiscent of Brazils past industrialisation policies. The ndings are in contrast with those for other countries, such as Almeida and Fernandes (2008). The multinomial logit results conrm the signicant and strong impact from having a technology BUY strategy on the probability for successful product innovations, most typically combined with process innovations. Especially the Make&Buy strategy gives the highest return, although the difference with the BuyOnly strategy is not

statistically signicant. These favorable results are driven by the majority of cases who combine product and process innovations.18 For the category of process innovations only, there is no signicant relationship with any type of innovation strategy.19 The signicance of the innovation strategy variables in the bivariate probit results for PROCESS (column (1)) therefore seems to come from those rms that combine their process innovations with product innovations. On the other determining factors, the multinomial results conrm the importance of nancial constraints, human capital formation and competition from foreign rms for successful product innovations. In the category of rms with only process innovations, we nd a much more weaker prole of rms, as there are signicantly more rms which are nancially constrained, with lower skilled managers and conned to local sales. 5.3. Multivariate analysis results on growth performance Table 5 presents our results on sales growth. Our major variable of interest is the effect of innovative performance on sales growth. Simply including the PROCESS and PRODUCT dummy in the growth equation, as done in column (1), or the exclusive categories, PRODONLY, PROCONLY, PPI, as done in column (2), may result in their coefcients being badly estimated, because of endogeneity and unobserved heterogeneity. As explained in the previous section, we use a two step procedure in an attempt to improve our estimation. The two-step procedure uses the predicted values of the innovative performance as instrument in the growth regression. Column (3) uses the bivariate estimations reported in columns (1) and (2) of Table 4, to construct predicted values for PB PROCESS and PB PRODUCT or predicted values for PB PRODONLY, PB PROCONLY and PB PPI in column (4). Column (5) of Table 5 reports the results as derived from the multinomial logit presented in Table 4, columns (3) and (4) on the predicted exclusive categories: process innovators alone (PM PROCONLY), product and/or process innovators (PM PRODONLY/PPI) or none of both. Finally, we estimate a reduced form version of the model explaining growth directly by the innovation strategies: MakeOnly, BuyOnly and Make&Buy (column (6)). The results on growth performance conrm the significance of negative rm size and age effects as predicted from the learning models. Small rms grow signicantly faster in Brazil than larger rms, but with signicant evidence of non-linearity in the size effect, as indicated by the squared term. Also younger rms grow signicantly faster, especially the very youngest, but no signicant interaction effect between size and age prevails. Our main variable of interest in the growth equation is the innovation performance of the company. Do rms that introduce successfully new product and pro-

The mean proportion of workers with higher education is 0.08, with a standard deviation of 0.09, making this example coincide with the effect about one standard deviation more and less than the mean proportion of high skilled workers. For secondary education, the mean proportion among the labour force equals 0.23, SD 0.20. The effect here is a 2% increase.

17

18 In the (not-reported) multinomial logit results for the 4 exclusive categories, disentangling PRODONLY from PPI, the marginal effects results for the PPI category on the innovation strategy are even stronger in favor of the BUY strategy, particularly the Make&Buy option. 19 The insignicance of the marginal effects for the innovation strategy variables in the PROCONLY category also holds in the multinomial logit results for the 4 exclusive categories, disentangling PRODONLY from PPI.

Please cite this article in press as: Goedhuys, M., Veugelers, R., Innovation strategies, process and product innovations and growth: Firm-level evidence from Brazil. Struct. Change Econ. Dyn. (2011), doi:10.1016/j.strueco.2011.01.004

G Model STRECO-454; No. of Pages 14

ARTICLE IN PRESS
M. Goedhuys, R. Veugelers / Structural Change and Economic Dynamics xxx (2011) xxxxxx 11 Effect of product and process inno.; exclusive categories (2) Predicted values from biprobit Predicted values from biprobit; exclusive categories (4) Predicted values from multinomial; exclusive categories (5) Reduced form estimation

Table 5 Innovation and rm sales growth (20002002). Effect of product and process innovation Dependent variable: sales growth 20002002 LSSAL SLSSAL LFIRMAGE SLAGE SALES00AGE PRODUCT PROCESS PRODONLY PROCONLY PPI
PB PRODUCT PB PROCESS PB PRODONLY PB PROCONLY PB PPI PM PROCONLY PM PPI + PRODONLY

(1)

(3)

(6)

0.181*** (0.042) 0.005*** (0.002) 0.314*** (0.069) 0.030*** (0.010) 0.006 (0.005) 0.057 (0.036) 0.021 (0.014)

0.181*** (0.042) 0.005*** (0.002) 0.314*** (0.069) 0.029*** (0.010) 0.006 (0.005)

0.183*** (0.042) 0.005*** (0.002) 0.320*** (0.068) 0.030*** (0.009) 0.006 (0.005)

0.184*** (0.042) 0.005*** (0.002) 0.320*** (0.068) 0.030*** (0.009) 0.006 (0.005)

0.180*** (0.042) 0.004*** (0.002) 0.323*** (0.068) 0.029*** (0.009) 0.007 (0.005)

0.175*** (0.042) 0.005*** (0.002) 0.316*** (0.069) 0.030*** (0.010) 0.006 (0.005)

0.018 (0.023) 0.019 (0.022) 0.041** (0.019) 0.357*** (0.111) 0.014 (0.083) 0.279 (0.180) 0.071 (0.176) 0.348*** (0.077) 0.035 (0.135) 0.339*** (0.061) 0.044 (0.029) 0.060** (0.026) 0.078*** (0.025) 0.068*** (0.015) 0.029** (0.014) 0.059 (0.037) 0.047** (0.019) 1.795*** (0.300) 1503 0.14

MAKEONLY BUYONLY MAKE&BUY OVERDRAF GMHIGH MNE EXPORT CONSTANT # observations R-squared 0.065*** (0.015) 0.030** (0.014) 0.057 (0.037) 0.047** (0.019) 1.837*** (0.300) 1503 0.14 0.065*** (0.015) 0.031** (0.014) 0.058 (0.037) 0.046** (0.019) 1.881*** (0.300) 1503 0.14 0.034** (0.016) 0.020 (0.015) 0.075** (0.037) 0.034* (0.019) 1.768*** (0.297) 1503 0.15 0.034** (0.016) 0.020 (0.015) 0.076** (0.037) 0.033* (0.019) 1.818*** (0.311) 1503 0.16 0.038** (0.017) 0.022 (0.015) 0.072* (0.037) 0.037* (0.019) 1.752*** (0.300) 1503 0.15

The estimation includes eight sector dummies and 12 geographical location dummies. Standard errors in parentheses. * Signicant at 10%. ** Signicant at 5%. *** Signicant at 1%.

Please cite this article in press as: Goedhuys, M., Veugelers, R., Innovation strategies, process and product innovations and growth: Firm-level evidence from Brazil. Struct. Change Econ. Dyn. (2011), doi:10.1016/j.strueco.2011.01.004

G Model STRECO-454; 12 No. of Pages 14

ARTICLE IN PRESS
M. Goedhuys, R. Veugelers / Structural Change and Economic Dynamics xxx (2011) xxxxxx

cess innovations also reap the benets of this in terms of larger sales growth? When we separately include PROCESS and PRODUCT in the growth equation, the results (not reported) are positive and signicant for each of them. But when including product and process innovation jointly as explanatory variables (column (1)), none of the innovation performance variables show up as being signicantly related to sales growth. The high occurrence of rms doing both product and process innovation is responsible for this effect, and requires the investigation of exclusive innovation categories: process innovators alone (PROCONLY), product innovators alone (PRODONLY), combined product and process innovators (PPI) with none of both the reference category (column (2)). Only those rms that combine successful product innovations with process innovations realize a signicantly higher sales growth, supportive of the complementarity between both types of innovations. These results are conrmed in the two-step estimations: only the bivariately predicted PRODUCT innovations, rather than the PROCESS innovations, are signicantly related to sales growth (column (3)). This is due to the rms combining product and process innovations as the results using the bivariate predictions to construct exclusive categories, indicate (column (4)). Also the multinomial predictions support the lack of effects on sales growth from introducing only process innovations, reminiscent of the cost cutting/restructuring story behind pure process innovations. Only the introduction of product innovations, exclusively, but mostly done in combination with process innovations, translates into signicantly higher sales growth (column (5)). When directly relating the innovation strategies to rms growth, rather than the innovation performance (column (6)), we nd that the technology BUY strategy is signicantly related to rm growth, particularly, although not signicantly more, when combined with a technology MAKE strategy. An innovation strategy of MakeOnly fails to signicantly translate in higher sales growth. This is not so surprising, as both the BuyOnly and Make&Buy strategies were found to positively effect the successful introduction of product innovations, particularly the combined product and process innovations (see Table 4) and the latter were found to translate into higher sales growth (columns (1)(5) in Table 5). With respect to other anking conditions for growth, the results conrm the strong, signicant and robust effect of short term credit constraints on growth. Also prior exporting experience is systematically and robustly related to higher growth rates. Other variables on foreign linkages and absorptive capacity are not consistently signicant across specications. 6. Conclusions Using data from the World Banks Investment Climate Survey (ICS) data collected in Brazil in 2003 for manufacturing sectors, this paper tries to contribute to the literature on technological progress and development by bringing on board a micro-econometric perspective on the factors determining innovative performance and rm growth. Our focus on Brazil allows analysing the case of a country in fast

development, where the interplay between diffusion (buy) and creation (make) of new technologies, shapes innovation and growth performance. With a large internal market and mixed legacy of protectionism and international openness, it is also a specic case to analyse the importance of international linkages for innovation and growth. There are a number of interesting ndings that appear to be robust across specications. First, like in many developing countries, also in Brazil, innovation strategies involve mostly technology acquisition, by acquiring knowhow embodied in machinery and equipment, exclusively but also substantially, in combination with own internal development. Secondly, innovative strategies contribute signicantly to being able to introduce successfully new innovations. For both process and product innovation the technology buy strategy is dominant, with some weak evidence on complementarities for rms combining make and buy strategies. The option of relying on internal development only is clearly less effective. The positive link between an innovation make and/or buy strategy and innovative performance, only holds for product innovations, either exclusively or in combination with process innovations. None of the innovative strategies are of signicant effect for rms which are only involved in process innovations. What is needed in terms of human capital is also surprising: a large share of workers with secondary education is very important for process innovations. It does not require a tertiary skilled workforce to be a successful process innovator in Brazil. Product innovation is more high-skill intensive, needing higher shares of highly (university) educated workers, a highly educated management and ICT usage for communicating with the market. This suggests that product innovation, in comparison to process innovation, is a more complex process with multiple inputs requiring more advanced knowledge inputs and absorptive capacity. Product innovation also translates into superior sales growth rates. This is particularly so when it is combined with process innovation. Process innovation alone without the introduction of new products runs the risk of being ineffective to boost higher growth. It is indeed possible that the benets of more cost efcient production are only reaped after an initial period of restructuring, beyond what we can measure with our data set. Alternative measures, such as productivity, productivity growth, or protability, may capture the benecial inuence of process innovation more rapidly. Another robust factor driving innovative and growth performance is access to nance. This supports policy interventions to alleviate the nancial constraints, by improving nancial market functioning and/or providing nancial incentives. Finally, our results on international linkages are somewhat mixed. International openness is important for stimulating innovative and growth performance, but this openness works particularly through competition and access to markets, but not necessarily as a mechanism for technology absorption improving innovative performance. Again this can be related to the specics of the Brazilian case, which has for a long time protected local champions in its policies.

Please cite this article in press as: Goedhuys, M., Veugelers, R., Innovation strategies, process and product innovations and growth: Firm-level evidence from Brazil. Struct. Change Econ. Dyn. (2011), doi:10.1016/j.strueco.2011.01.004

G Model STRECO-454; No. of Pages 14

ARTICLE IN PRESS
M. Goedhuys, R. Veugelers / Structural Change and Economic Dynamics xxx (2011) xxxxxx 13

Appendix A.

Table A1 Denition of variables. Sales growth 20002002 PROCESS PRODUCT Firm characteristics (X) LFIRMAGE LTL00 Innovation strategy (I) MAKEONLY Average annual sales growth over the period 20002002, calculated by (ln(sales2002) ln(sales2000))/2 =1 if the rm successful introduced new technology that has substantially changed the way the main product is produced, in 19982002 =1 if the rm successfully developed a major new product line in 19982002 Age of the rm, in logarithmic terms Size of the rm, measured by number of employees in 2000, in log 2.68 (0.80) 3.91 (1.15) 0.12 0.34

=1 if rm reports in-house development as major way of acquiring new technology, not embodied in machinery, hiring key personnel, licensing BUYONLY =1 if rm reports new technology embodied in machinery, the hiring of key personnel or licensing of technology as major ways of acquiring new technology; not developed in-house MAKEANDBUY =1 if rm reports both in-house development and embodied in machinery or hiring key personnel or licensing as major ways of acquiring new technology The technology position of the rm in its market (T) TECHLEADER =1 if the rm reports that his technology is more advanced than that of its main competitors TECHLAGGARD =1 if the rm reports that his technology is less advanced than that of its main competitors Absorptive capacity (AC), as measured through various proxies LFHIGH Proportion of the labour force with higher education [0,1] LFSEC Proportion of the labour force with secondary education as highest level attained [0,1] GMHIGH =1 if the rm has a general manager with a graduate or postgraduate degree or diploma of tertiary college LTOTEXPER Total number of years of experience of the manager working in this industry, in log terms ICT =1 if the rm uses a email to interact with clients and suppliers TRAINING =1 if the rm offers formal training to its employees Foreign linkages (F) FORFIRMEXPER =1 if the manager has previously acquired working experience in the same industry, in a foreign rm MNE =1 if the rms principal shareholder is a foreign company EXPORT =1 if the rm is a direct exported prior to 2000 FOREIGNSUPPLIER =1 if the rm sources its main supply or inputs from foreign owned rms Competition (C) COMPIMP =1 if for the rms main product the main competitor is an imported product COMPPRESSPF =1 if rm reports that the most important inuence/pressure to develop new products are foreign competitors COMPPRESSPD =1 if rm reports that the most important inuence/pressure to develop new products are domestic competitors Financial constraints (FIN) OVERDRAFT =1 if the rm has an overdraft facility with a formal bank

0.46

0.24 0.15

0.08 (0.09) 0.23 (0.20) 0.51 2.46 (0.69) 0.92 0.67 0.08 0.04 0.27 0.40 0.13 0.09 0.24

0.74

References
Abramovitz, M., 1986. Catching up, forging ahead and falling behind. Journal of Economic History 46 (2), 386406. Aghion, P., Howitt, P., 1998. Endogenous Growth Theory. MIT Press, Cambridge, MA, 694 pp. Aitken, B., Harrison, A., 1999. Do domestic rms benet from foreign direct investment? Evidence from Venezuela. American Economic Review 89 (3), 605618. Almeida, R., Fernandes, A.M., 2008. Openness and technological innovations in developing countries: evidence from rm-level surveys. Journal of Development Studies 44, 701727. Bastos, F., Nasir, J., 2004. Productivity and the investment climate: what matters most? World Bank Policy Research Working Paper 3335. Bell, M., Pavitt, K., 1995. The development of technological capabilities. In: Haque, I. (Ed.), Trade, Technology and International Competitiveness. World Bank, Washington, DC. Benavente, J.M., 2006. The role of research and innovation in promoting productivity in Chile. In: Hall, B., Mairesse, J. (Guest Eds.), Empirical Studies of Innovation in the Knowledge Driven Economy. Economics of Innovation and New Technologies 15, 301315.

Blalock, G., Gertler, P., 2009. How rm capabilities affect who benets from foreign technology. Journal of Development Economics 90, 192199. Blmstrom, M., 1986. Foreign investment and productive efciency: the case of Mexico. Journal of Industrial Economics 37, 315328. Blomstrm, M., Kokko, A., 1998. Multinational corporations and spillovers. Journal of Economic Surveys 12 (3), 247277. Blmstrom, M., Persson, H., 1983. FDI and spillover efciency in underdeveloped economy: evidence from Mexico. World Development 11, 493501. Bodas Freitas, A.M., Clausen, T., Fontana, R., Verspagen, B., 2008. Formal and informal external linkages and rms innovative strategies. A cross-country comparison, UNU-MERIT Working Paper 2008-065. Cassiolato, J., Lastres, H., Maciel, M.L., 2003. Systems of Innovation and Development. Edward Elgar Publishing Ltd, Cheltenham, UK. Chudnovsky, D., Lpez, A., Pupato, G., 2006. Innovation and productivity in developing countries: a study of Argentine manufacturing rms behaviour (19922001). Research Policy 35, 266288. Coe, D., Helpman, E., Hoffmaister, A., 1997. North-south R&D spillovers. The Economic Journal 107, 134149. Cohen, W.M., Levinthal, D.A., 1989. Innovation and learning: the two faces of R&D. Economic Journal 99, 569596.

Please cite this article in press as: Goedhuys, M., Veugelers, R., Innovation strategies, process and product innovations and growth: Firm-level evidence from Brazil. Struct. Change Econ. Dyn. (2011), doi:10.1016/j.strueco.2011.01.004

G Model STRECO-454; 14 No. of Pages 14

ARTICLE IN PRESS
M. Goedhuys, R. Veugelers / Structural Change and Economic Dynamics xxx (2011) xxxxxx Markusen, J.R., Venables, A.J., 1999. Foreign direct investment as a catalyst for industrial development. European Economic Review 43, 335356. Miravete, E., Pernias, J., 2006. Innovation complementarity and scale of production. Journal of Industrial Economics 54, 129. Narula, R., Dunning, J., 2000. Industrial development, globalisation and multinational enterprises: new realities for developing countries. Oxford Development Studies 28 (2). Nelson, R. (Ed.), 1993. National Systems of Innovation: A Comparative Study. Oxford University Press, Oxford. Nelson, R., Winter, S.G., 1982. An Evolutionary Theory of Economic Change. Harvard University Press, Cambridge, MA. OECD, 2006. Economic Surveys: Brazil, vol. 18. Pakes, A., Ericson, R., 1998. Empirical implications of alternative models of rm dynamics. Journal of Economic Theory 79, 145. Peirano, F., 2007. Technological change in the manufacturing sectors of Argentina and Brazil: an analysis based on innovation surveys 2007. In: De Negri, J., Turchi, L.M. (Eds.), Technological Innovation in Brazilian and Argentine Firms. IPEA, Brasilia. Prez, P., Dutrnit, G., Barceinas, F., 2005. Actividad Innovadora y Desempeno Econmico: un anlisis economtrico del caso mexicano. In: Albornoz, M., Ratto, D. (Eds.), Indicadores de Ciencia y Tecnologa en Iberoamerica. Agenda. , pp. 299318. Prochnik, V., Dias de Araujo, R., 2007. A comparison of the technological strategies of the less productive Argentine and Brazilian rms. In: De Negri, J., Turchi, L.M. (Eds.), Technological Innovation in Brazilian and Argentine Firms. IPEA, Brasilia. Raffo, J., Lhuillery, S., Miotti, L., 2008. Northern and southern innovativity: a comparison across European and Latin American countries. The European Journal of Development Research 20, 219239. Romer, P., 1994. The origins of endogenous growth. Journal of Economic Perspectives 8, 322. Sleuwaegen, L., Goedhuys, M., 2002. Growth of rms in developing countries, evidence from Cte dIvoire. Journal of Development Economics 68, 117135. Solow, R., 1956. A contribution to the theory of economic growth. Quarterly Journal of Economics 70 (1), 6594. Srholec, M., Verspagen, B., 2008. The voyage of the beagle in innovation systems land. Explorations on sectors, innovation, heterogeneity and selection, TIK Working Paper on Innovation Studies 20080220. University of Oslo. Szapiro, M., 2003. Downgrading local capabilities in IT: the telecom innovation system in Campinas. In: Cassiolato, J., Lastres, H., Maciel, M.L. (Eds.), Systems of Innovation and Development. Edward Elgar Publishing Ltd, Cheltenham, UK. Veugelers, R., Cassiman, B., 2004. Foreign subsidiaries as channel of international technology diffusion: some direct rm level evidence from Belgium. European Economic Review 48, 455476. World Bank, 2005. Brazil Investment Climate Assessment, vol. II. Mimeo, The World Bank, Washington, DC, 278 pp. World Bank, 2008. Global Economic Prospects 2008, Technology Diffusion in the Developing World. The World Bank, Washington, DC, 201 pp.

Crpon, B., Duguet, E., Mairesse, J., 1998. Research, innovation and productivity: an econometric analysis at the rm level. Economics of Innovation and New Technology 7, 115158. De Brito Cruz, C., de Mello, L., 2006. Boosting innovation performance in Brazil. OECD Economic Department Working Paper 532. De Negri, J., Turchi, L.M. (Eds.), 2007. Technological Innovation in Brazilian and Argentine Firms. IPEA, Brasilia. De Negri, F., Araujo, B.C., De Negri, J.A., 2008. Channels of international technology transfer in Brazilian industrial rms. In: Paper Presented at the Second MEIDE Conference , Beijing. Dollar, D., Hallward-Driemeier, M., Menigstae, T., 2003. Investment Climate and Firm Performance in Developing Countries. Mimeo the World Bank Development Research Group, 42 pp. Eifert, B., Gelb, A., Ramachandran, V., 2005. Business environment and comparative advantage in Africa: evidence from the investment climate data. Center for Global Development, Working Paper 56. Enos, J., 1992. The Creation of Technological Capabilities in Developing Countries. Pinter, London. Evans, D.E., 1987. The relationship between rm growth, size and age: estimates for 100 manufacturing industries. Journal of Industrial Economics 35, 567582. Fagerberg, J., Srholec, M., 2008. National innovation systems, capabilities and economic development. Research Policy 37, 14171435. Fagerberg, J., Srholec, M., Knell, M., 2007. The competitiveness of nations: why some countries prosper, while others fall behind. World Development 35 (10), 15951620. Grg, H., Greenaway, D., 2003. Much ado about nothing? Do domestic rms really benet from FDI? Mimeo. Grossman, G., Helpman, E., 1991. Innovation and Growth in the Global Economy. MIT Press, Cambridge, MA. Hoekman, B., Smarzynska Javorcik, B. (Eds.), 2006. Global Integration and Technology Transfer. Palgrave Mac Millan, New York. Hoekman, B.K., Maskus, K., Saggi, K., 2005. Transfer of technology to developing countries: unilateral and multilateral policy options. World Development 33 (10), 15871602. Jovanovic, B., 1982. Selection and the evolution of industry. Econometrica 50, 649670. Katz, J. (Ed.), 1987. Technology Generation in Latin America. MacMillan, London. Kim, L., 1997. Imitation to Innovation: The Dynamics of Koreas Technological Learning. Harvard Business School Press, Cambridge. Kmenta, J., 1997. Elements of Econometrics. The University of Michigan Press, 786 pp. Kokko, A., 1994. Technology, market characteristics and spillovers. Journal of Development Economics 43, 279293. Lall, S., 1992. Technological capabilities and industrialisation. World Development 20, 165186. Lugones, G., Suarez, D., 2007. National innovation systems in Brazil and Argentina: key variables and available evidence. In: De Negri, J., Turchi, L.M. (Eds.), Technological Innovation in Brazilian and Argentine Firms. IPEA, Brasilia.

Please cite this article in press as: Goedhuys, M., Veugelers, R., Innovation strategies, process and product innovations and growth: Firm-level evidence from Brazil. Struct. Change Econ. Dyn. (2011), doi:10.1016/j.strueco.2011.01.004