Persistence of innovation in unstable environments: Continuity and change in the firm's innovative behavior

G Model RESPOL-2924; No. of Pages 11

ARTICLE IN PRESS Research Policy xxx (2013) xxx–xxx

Contents lists available at ScienceDirect

Research Policy journal homepage: www.elsevier.com/locate/respol

Persistence of innovation in unstable environments: Continuity and change in the firm’s innovative behavior Diana Suárez a,b,∗ a b

Aalborg University, Denmark General Sarmiento National University, Argentina

a r t i c l e

i n f o

Article history: Received 16 August 2012 Received in revised form 7 October 2013 Accepted 9 October 2013 Available online xxx Keywords: Persistence of innovation Unstable environments Innovative behaviors

a b s t r a c t The concept of persistence is generally used to define the positive relationship between past and present innovations, which is explained by feedback and accumulation processes triggered by the firm’s past results. This paper states that changes in the economic or institutional conditions of the environment impact on the type of profitable innovations, and past innovations might not be suitable for the new environment. As a result, firm’s innovative behavior might change, which means that the firm’s set of decisions about engaging in the seek for innovations or not and, if so, the set of investments and capabilities it allocates to innovate could be modified. Empirical evidence is provided to reject the persistence hypothesis and to show that past innovations do not necessarily impact present ones. This paper examines the relationship between past and present innovations for a group of Argentinean firms during 1998–2006, which coincides with a period of macroeconomic instability. Results suggest that persistence has to be analyzed in terms of a dynamic firm’s innovative behavior—regardless of its results—and how it allows the firm to accumulate competences and resources, which increases the odds of successfully responding to changes in the environment and continuing to innovate. © 2013 Elsevier B.V. All rights reserved.

1. Introduction The concept of innovation persistence refers to the feedbacks, accumulation, and lock-in effects that arise from innovations and put the firm in a better position to seek new innovations, with the consequent increase in the odds of continuing to achieve these (Antonelli, 1997; Geroski et al., 1997; Nelson and Winter, 1982; Phillips, 1971). The empirical literature corroborates this, although it emphasizes that persistence is confirmed when the firm’s innovative conduct is associated with explicit investments to generate technological and organizational changes (Clausen et al., 2011; Frenz and Prevezer, 2012; Le Bas et al., 2011; Peters, 2009; Raymond et al., 2010). These studies, however, implicitly condition persistence to the stability of the environments from which the empirical data was extracted (mostly European countries). This additional condition raises questions regarding the possibility of extrapolating the conclusions to unstable environments (such is the case of many middle-income countries).

∗ Correspondence to: General Sarmiento National University, Argentina, J.M. Gutierrez 1150, Los Polvorines (CP1613GSX), Buenos Aires, Argentina. Tel.: +45 9940 8235. E-mail addresses: [email protected], [email protected]

Persistence literature shares the underlying idea that the environment does not change. It assumes that what the firm did in the past is useful for the things the firm has to deal with in the present. If one relaxes the assumption about the environment, a new question emerges: what if the environment changes and past innovations are no longer suitable for the new environment? This is the question that guides this article. The hypotheses state that innovation persistence is explained by the firm’s continuity on the innovative investments (the inputs) and not only by its innovation results (the outputs). The main objective is to discuss the concept of innovation persistence in unstable environments, accepting the possibility that a firm’s innovative behavior might change. This means that faced to a change in the environment, firms might decide to continue, to stop or to initiate an innovation project. The relationship between past and present innovations will be tested in a group of Argentinean firms using data from national innovation surveys in three distinct macroeconomic environments: the 1998–2001 economic crisis, the 2002–2004 recovery period, and the 2005–2006 growth phase. Results suggest that persistence is conditioned by the performance of sustained innovative investments and the firm’s ability to respond to changes in the environment. On average, the instability of the environment fostered isolated short-term innovations, which had low impact on the firms’ capabilities and resources,

0048-7333/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.respol.2013.10.002

Please cite this article in press as: Suárez, D., Persistence of innovation in unstable environments: Continuity and change in the firm’s innovative behavior. Res. Policy (2013), http://dx.doi.org/10.1016/j.respol.2013.10.002

G Model RESPOL-2924; No. of Pages 11 2

ARTICLE IN PRESS D. Suárez / Research Policy xxx (2013) xxx–xxx

which are the sources of persistence. As such, past innovations did not increase the probabilities of future ones. Conversely, among firms with long-term innovative behavior, and given the impact of path dependence and lock-in effects, past results delayed achieving the results required to compete in the present. Finally, persistence of innovation was observed among firms that changed together with the environment, in which it had the greatest impact. Therefore, there are reasons to believe that these firms sought innovations that were suitable to the new environment without the friction caused by path dependence and lock-in effects. The remainder of this paper is as follows. Section 2 presents the theoretical framework and key empirical analysis aimed at testing persistence. It also discusses the potential and limitations of this concept when applied to unstable environments. In Section 3, methodology and data are defined. In Section 4, the model is applied and results are discussed. Finally, in Section 5, some conclusions are provided.

2. Theoretical background and empirical evidence 2.1. Innovation and persistence The concept of persistence can be traced back to Schumpeter’s (1942) creative accumulation process. This author argued that the process of technical change was associated with the existence of large firms competing in oligopolistic markets, where the development of innovations and the investments made to achieve this (R&D) triggered accumulation processes which tended to perpetuate the firm’s presence in the market. Based on these ideas, although not restricted to the oligopolistic firm, three approaches to innovation persistence became the theoretical basis of recent empirical work: path dependence, virtuous cycles of accumulation, and market power dynamics. According to the path dependence approach, the development of innovations in the past enhances a firm’s capabilities and generates opportunity costs in the present, increasing the odds of the firm deciding to carry out another innovation project, which obviously affects the likelihood of actually achieving innovations. Within a particular space and time, past decisions generate sunk costs in terms of resources (irreversibility) and set the margin for obtaining scale economies (indivisibility). Both aspects involve opportunity costs for new decisions, which are weighed up when the firm makes decisions regarding new innovative processes (Antonelli, 1997, 2008). The analysis of persistence in terms of virtuous cycles of accumulation is based on Nelson and Winter’s (1982) work. For these authors, persistence emerges from the generation of feedbacks between past innovations, present investments, and future innovations. These authors argued that the decision-making process that leads to innovation is a routine (a standard behavior) which, in the case of success, will be repeated. As a consequence, the persistence of routines impacts the firm’s innovative features, either by guiding the innovative projects or by blocking them. Successful firms (the ones that achieve innovations) stand out from the competition, create entry barriers and obtain quasi-monopoly rents, which improve their financial situation and generate surpluses to be reinvested in the quest for new innovations. The market power approach can be found in the work of Phillips (1971), Mansfield (1962), and Geroski et al. (1997), among others. According to this approach, when a firm reaches an innovation, it achieves greater market power and obtains extraordinary incomes (increases its level of resources). Past innovations thus allow future ones to be financed. The other way around, given the additional uncertainty of innovation projects, those firms that cannot generate sufficient surpluses to fund future innovations face additional

financial obstacles or higher entry costs as a result of the differential interest rate arising from the risk of such projects. In the three approaches described above, new innovations arise because past innovations have increased the firm’s resources and capabilities (capabilities and opportunity costs, in the terms of the path dependence approach; extra-profits, entry barriers, and routines, in the terms of the cycles of accumulation approach; and profits, in the terms of the market power approach). In all cases, the assumption behind the expected positive association between past and present innovations is that past innovations trigger new innovation projects and this leads to new results that start the process all over again. In this way, innovation persistence is the serial correlation between past and present innovations and the statistical demonstration of the binomial accumulation feedback that emerges from the interaction between the firm and the market (Malerba et al., 1997). To the extent that innovations have to be mediated by the market (they result from successful introductions of products, processes, or organizational practices), the firm will receive feedbacks from the market which will shape its innovative behavior. When facing a change in market conditions, path dependence will narrow the firm’s range of possible responses—due to sunk and opportunity costs—resources and capabilities will limit the type of innovative projects the firm can carry out, and the routines will determine how the response is taken and applied. However, the environment is more than market interactions (Lundvall, 1992; Nelson, 1994). It is the set of institutions affecting the selection process. In this sense, changes in the environment (such as a change in the economic model of growth or an economic recession) will impact not only on the last innovation project to have been implemented but on firm’s behavior as a whole and how it faces competition. If the firm has to change its innovative trajectory to face the new environment, predicting a positive correlation between past and present innovations seems difficult. Therefore, although the three persistence approaches can explain the positive association between past and present innovations within stable contexts, they fall short when explaining this relationship within unstable ones, let alone in contexts of changes in the rules of the game or profound shifts in the main trends of demand. 2.2. Empirical evidence from the literature To some extent, the recent literature on persistence can help to understand how changes in the environment could affect innovation persistence. Based on merged innovation surveys, these studies draw attention to the importance of inputs to the innovation process and how persistence is subject to specific innovative conducts. Within this literature, persistence is confirmed only in some specific types of firms. One set of studies finds a positive relationship between past and present innovations, but this is subject to simple structural characteristics of the firm. Raymond et al. (2010) analyze persistence among Dutch firms and found that it exists among firms from sectors with high and medium-high technological intensity, which implicitly correlates persistence with high R&D expenditure, higher levels of qualified human resources, and the level of technological opportunities (the definition of high-tech sectors). For the other sectors (medium-low and low technological intensity), the hypothesis of persistence is not verified. Similarly, Peters (2009) corroborates persistence among German firms but finds that capabilities, size, and access to subsidies are relevant variables to explaining continuity of innovation. Finally, Frenz and Prevezer (2012) analyze a group of British firms and find that the variables that account for the firm’s innovative behavior, together with size, sector, and age, are more important in explaining the recurrence of innovation than actual past results.

Please cite this article in press as: Suárez, D., Persistence of innovation in unstable environments: Continuity and change in the firm’s innovative behavior. Res. Policy (2013), http://dx.doi.org/10.1016/j.respol.2013.10.002

G Model RESPOL-2924; No. of Pages 11

ARTICLE IN PRESS D. Suárez / Research Policy xxx (2013) xxx–xxx

A second set of studies goes a step further and studies persistence in relation to micro-heterogeneity. This is the case of Le Bas et al. (2011) who, for the case of Luxembourgish firms, differentiate between firms where persistence is verified from firms where it is not, and find that organizational innovations are a key determinant not only to persistence, but to innovations generally speaking. In this case, persistence is linked to the search for complementary innovations and, of course, to undertaking complementary innovation activities. Clausen et al. (2011) look at the phenomenon from the opposite perspective, in terms of causality. These authors identify different combinations of activities (innovative behaviors) among Norwegian enterprises and corroborate persistence in the case of science-based and market-oriented firms, but not in the case of firms with sporadic investments in innovation or in supplierbased ones. What this empirical analysis adds to the literature is the differential impact in terms of the innovation project the firm can carry out (structural determinants such as sector, size and age) and the micro-determinants associated with the specific innovation project (the innovative behavior). Therefore, if the relationship between past and present innovations will be mediated by the firm’s innovative conduct, the relationship between persistence and changes in the environment will be also affected by the firm’s behavior. 2.3. Environmental changes, firm’s adaptation and persistence of innovation From an evolutionary perspective, as long as the firm moves forward with its innovation project, learning processes and the evolution of the environment will lead the firm to adjust its behavior (Nelson, 1991). The innovation process will trigger additional learning processes that will enhance the firm’s capabilities and increase its resources (Nelson and Winter, 1982; Penrose, 1959). As a result, the firm is more likely to achieve positive results in the future. However, the characteristics of that persistence (level of probability, type of innovation, and magnitude of the impacts) will be explained by the specificities of firm behavior (innovation activities, capabilities, resources) and how it deals with the environment (at the national, meso, and regional level). Changes in firm behavior depend on the firm’s ability to react to signals from the environment and from its own learning processes, and can take place from one moment to the next: innovation projects can be abandoned, the firm can change the innovation it is seeking, or a past innovation may no longer be suitable for the environment. In all these cases, the expected feedback process may never happen. In this respect, the possibility of a change in the firm’s innovative behavior is neglected in the persistence literature, and an automatic link between past innovations, present behavior, and future results is assumed to exist. Although the reviewed literature argues that path dependence will limit the range of the firm’s possible responses, and this could work against innovation persistence, it neglects the possibility of a change in the firm’s innovative behavior. If a firm’s innovative behavior is based on the importation of raw materials and the domestic exchange rate is devalued, the firm will most certainly have to adjust its project to the new relative prices, perhaps substituting suppliers. In cases like this, past innovations were not necessarily useful for the new environment, meaning that past results and past learning and accumulation processes might even delay the achievement of future innovations. The firm would have to adapt its innovative behavior, undertake new investments, and acquire new capabilities before it expecting future results. Conversely, if the firm was not investing in innovation, and the new environment created incentives to innovate, then present innovations would arise from path dependence without innovations, and persistence would not be confirmed. In the same example

3

H1 Type of Innovative behavior Continuingt

Innovation resultst-1 ( independent variable)

Startingt

H2 H3

Innovation resultst (dependent variable)

H4 Stoppingt

Investments in innovation resources and capabilitiest

-------------------------------------------------------

Structural characteristicst Fig. 1. Summary of the theoretical arguments and hypotheses*. *Innovation results: successful introduction to the market of a new or significantly improved product, process, or organizational or commercial technique (OECD, 2005). Base behavior: non-innovative firms (firms with zero investments in innovation).

as before, the new relative prices of a devalued currency could trigger export-oriented innovation projects, regardless of past results. Another element excluded from persistence studies is that past innovations can be a chance event that was triggered by a specific problem-solving process (Nelson and Winter, 1982) or the search for short-term benefits following opportunistic behavior (Freeman, 1974). Although learning and accumulation processes may arise, the arguments regarding the impact of past innovations on present ones are weak, to say the least. Consequently, if persistence is empirically confirmed, there would be good reason to believe that it is a spurious type of persistence (Raymond et al., 2010), caused by the variables that have been omitted from the model. To conclude, studying persistence within a context of environmental change leads one to expect that firms will modify their behavior and that this will affect the persistence of innovation. Some innovative behaviors may trigger the expected virtuous process of accumulation that leads to persistence, but other may lead to results that have little impact on future ones. As such, some innovative behaviors will lead to the firm successfully innovating, but only some of these to it innovating persistently. Within unstable environments, persistence will be the manifestation of specific innovative behaviors rather than a simple serial correlation between past and present results. From a theoretical angle, this means that the feedback and accumulation processes that the literature attributes to the persistence of innovation are in fact the result of the persistence of certain types of innovative conduct. 3. Model and methodology 3.1. The hypothesis Fig. 1 summarizes the theoretical and empirical arguments behind the hypotheses. H1 states that past innovation results impact on present innovation results (both outputs of the innovation process), subject to the firm’s investments in innovation resources and capabilities and structural characteristics. The former are the inputs of the innovation process (e.g. investments in R&D and machinery, skilled human resources, etc.). The latter are the usual control variables (size, sector and capital ownership). This is the persistence hypothesis discussed in Sections 2.1 and 2.2, where path dependence, accumulation and learning processes lead past innovations (independent variable) to increase the odds of innovating in the present (dependent variable), conditioned by

Please cite this article in press as: Suárez, D., Persistence of innovation in unstable environments: Continuity and change in the firm’s innovative behavior. Res. Policy (2013), http://dx.doi.org/10.1016/j.respol.2013.10.002

G Model RESPOL-2924; No. of Pages 11 4

ARTICLE IN PRESS D. Suárez / Research Policy xxx (2013) xxx–xxx

these two sets of variables. Therefore, H1 can be formulated as follows: H1. The probability of innovating depends positively on having innovated in the past (persistence is verified). H1 might be verified within stable environments, where external incentives do not change and innovation persistence is correctly explained by the literature reviewed in Section 2.1. What I argue in this paper is that changes in the environment could lead to changes in the innovation incentives and firms might change their innovative behavior. To be more concrete, a devaluation of the domestic currency could allow firms to compete without innovating or, the other way around, to invest in innovations to substitute imports. This means that between t and t + 1, a firm can choose to continue with its innovation investments, to start new investments, or to stop them and this will impact on present innovation results (H2–H4). Once it is accepted that changes in the firm behavior can take place, several situations emerge just by applying a simple Cartesian product. As we shall see in Section 3.2, the panel is made up of three periods, so firm behaviors will be classified according to their behavior from t − 2 to t − 1 and from t − 1 to t as follows: • Continuous innovative firms (Cont): firms with continuous investments in innovation during two sub-periods (Continuingt in Fig. 1); • New innovative firms (New): firms which started to invest in innovation from one sub-period to the following one (Startingt in Fig. 1); • Sporadically innovative firms (Spor): firms which stopped investing in innovation from one sub-period to the following one (Stoppingt in Fig. 1); • Non-innovative firms (NI): firms with zero investments in innovation during two sub-periods (base behavior in Fig. 1). From a methodological point of view, two extreme behaviors can be defined: firms that invest in innovation and firms that do not. When these behaviors are analyzed in dynamic terms, firms face two possible choices: to stop their behavior or to continue with it. Continuous and non-innovative firms both decided to continue with the same behavior, whereas sporadic and new innovative firms decided to change theirs. When these possibilities are analyzed in terms of specific length of time (in this case, the three sub-periods between 1998 and 2006), the combination of choices gives the different possible outcomes; the eight innovative strategies. The literature on innovative behavior only looks at the final decision—or the average choice—but it does not analyze the transition from one situation to another, which is precisely the area where this paper intends to make a conceptual and methodological contribution. Based on this classification, a second set of hypothesis can be formulated, which are also schematized in Fig. 1 and discussed below. H2. The probability of innovating depends positively on having innovated in the past among firms with continuous innovative behavior (persistence is verified). H3. The probability of innovating depends positively on having innovated in the past among firms with new innovative behavior (persistence is verified). H4. The probability of innovating does not depend on having innovated in the past among firms with sporadic innovative behavior (persistence is not verified). H2 refers to firms with sustained innovative behavior (continuous firms). These firms should present higher capabilities and more

resources as a result of feedbacks from past innovation and accumulation processes. This better situation should allow them to react to the changes in the environment and to innovate persistently. Although this hypothesis could be seen as a tautology (firms that invest continuously also innovate continuously), it is not given the uncertainty attached to the innovation projects. At the same time, it is what the literature predicts and is thus worth testing. H3 is about the new innovative firms (firms that changed from a noninnovative to an innovative behavior). The assumption is that the change in the innovative behavior will impact the likelihood of achieving innovations, regardless of past results. Based on the literature, one can expect there to be a significant correlation between past and present results in firms with new innovative behavior once they start investing in innovation. H4 refers to firms with sporadic innovative behavior (isolated innovative investments), which means that the achieved innovations resulted from isolated problem-solving activities or short-term opportunistic reactions. In these cases, feedback and accumulation processes should be low and persistence of innovation should not be verified. In short, H1 is what persistence literature predicts: past innovations have a positive impact on present ones (a significant and positive coefficient in the regressions for the lagged dependent variable). H2–H4 establish the expected relationship between present and past innovations for each specific behavior (a significant and positive coefficient for continuous and new innovative firms and an insignificant coefficient among sporadic innovative ones) and aim to demonstrate that persistence depends on how past results were achieved (innovative conduct) as well as the importance of analyzing firms’ behavior from a dynamic perspective. 3.2. The data To test the hypotheses, a model that relates past and present innovations will be constructed and applied to a balanced panel of Argentinean manufacturing firms that participated in four Argentinean Innovation Surveys. The panel consists of 800 firms of different sizes, sectors and types of capital ownership. The data was gathered by the Argentinean National Institute of Statistics and Census (INDEC) using a CIS-type questionnaire (INDEC, 2010). The period under analysis covers three different environments. Between 1998 and 2002, Argentina’s GDP dropped 20%, the rate of unemployment reached 25%, and half of all families were below the poverty line. Since the second semester of 2002, Argentina has been growing again, a process that has been driven by the increase in domestic demand and the competitive shock of a 300% devaluation of the national currency. By 2005, GDP levels were above the 1998 peak, the unemployment rate had sunk to under two digits, and the increase in income and the redistribution of this had reduced total poverty levels to less than 30%. With regard to the sample, given that firms were required to have participated in all four surveys, all firms were established before 1998 (the first year of the first innovation survey), and all survived the 1998–2001 recession. This means that the sample is biased toward those firms who managed to cope better with the recession. Merged innovation surveys (a micro-data panel) have an additional bias given by the higher survival rate among firms with innovation results. The bias, however, does not undermine the usefulness of the database, as long as it allows a relatively large group of firms to be studied over time, within a period of deep environmental change. Regarding the number of observations, although the same firms are present every year, the data refers to arbitrary sub-periods allocated by the reference periods of the surveys: 1998–2001, 2004–2004, 2005, and 2006. As a result, the nine-year period was

Please cite this article in press as: Suárez, D., Persistence of innovation in unstable environments: Continuity and change in the firm’s innovative behavior. Res. Policy (2013), http://dx.doi.org/10.1016/j.respol.2013.10.002

G Model RESPOL-2924; No. of Pages 11

ARTICLE IN PRESS D. Suárez / Research Policy xxx (2013) xxx–xxx

segmented into three sub-periods: 1998–2001, 2002–2004, and 2005–2006. In this way, the panel is made up of three observations for each case (firm). From these sub-periods, continuous variables were recalculated as annual averages and those in local currency were deflated by the producer price index, base 1998, at three digits of the International Standard Industrial Classification (ISIC) on a yearly basis. For dichotomous variables, the criterion was that of a positive response in at least one year of the sub-period.2 3.3. The model The innovative behaviors defined in Section 3.1 (continuous, new, sporadic, and non-innovative firms) will be analyzed in terms of the persistence of innovation. The variable innovation results is the same as that used in the CIS questionnaires. Firms are asked about the development of a new or significantly improved product, process, or organizational or commercial technique (OECD, 2005). For the purpose of this paper, a firm has achieved innovations if it has declared at least one innovation, regardless of the type (product, process, organization, commercialization) and the scope (new to the firm, the market, or the world). This definition is similar to that used in the reviewed literature (although in some cases only product and process innovations are considered), which will allow a comparison to be made of our findings. The rationality of this selection is based on the recognition of the potentially equal impact of technological and non-technological innovations on a firm’s performance (Le Bas et al., 2011; Lugones and Suarez, 2010; Nelson and Winter, 1982). From a practical point of view, the use of innovations in general allows us to acknowledge the fact that one specific innovation could trigger new ones in other areas of the firm: for example, if a firm has developed a new product, it will probably invest in the exploitation of that product during subsequent periods, which could in turn demand process, organizational, or commercial innovations. A probabilistic model was constructed to test the relationship between innovative behaviors and past and present results. Since the dependent variable is a binary one (did or did not innovate at time t), a dynamic random effects probit model was chosen, which also allows us to control micro-heterogeneity by means of the inclusion of unobserved effects. From a theoretical point of view, there are good reasons to believe that some unobservable effects are time-invariant and firm-specific while others are timevariant and case-invariant. As such, a random effect model better suits the characteristics of the expected micro-heterogeneity. From a methodological point of view, the selection of a random over a fixed-effects model allows the Wooldridge (2005) solution to be included to relax the assumption regarding the independence between the observed and unobserved effects.3 Wooldridge’s (2005) solution is based on the inclusion of the average values of the independent variables plus the initial value of the dependent variable. This allows the assumptions about the lack of correlation between the idiosyncratic term, the error, and the predictive variables to be relaxed. Theoretically speaking,

2 The different number of years included in each sub-period is likely to affect the reported number of innovations (it is more likely that a company has innovated when the consultation is for a period of four years than for a period of two). Unfortunately, the way this variable is addressed in the questionnaire (the question relates to the period and not to the year) makes it impossible to use a variable that spans an equal number of years. However, the rate of firms with innovation results for the whole panel does not change significantly between sub-periods (58%, 47%, and 53% for 1998–2001, 2002–2004 and 2005–2006, respectively). 3 Since there is no general transformation for eliminating the unobservable effects in binary estimations under fixed effects models, Hausman’s specification test cannot be performed to check the accuracy of the selection.

5

Wooldridge’s solution is based on the assumption that firms’ unobserved characteristics can be approximated as a linear function of their observable behavior. This implies that the firm’s innovative behavior (linkages, expenditures, qualified human resources, access to external funds, taking into account altogether) is in part the result of its unobservable characteristics.4 The firms’ characteristics will be approximated by means of the three dimensions associated with innovative behavior: innovation investments, capabilities, and resources, which coincides with the theoretical causes of persistence identified in Section 2.1 and the significant variables observed in Section 2.2. Formally, the model is written as: Innoti = ˇ0 + ˇ1 Innot−1i + Wti + wi + Ti + i + ∈ ti

(1a)

Innoti = ˇ0 + ˇ1 Inno contt−1i + ˇ2 Inno newt−1i + ˇ2 Inno sport−1i + Wti + wi + Tt + i + ∈ ti

(1b)

Wti = ˇa IIti + ˇb IBti + ˇc QHRti + ˇd linkti + ˇf ERti + ˇg Laborti

(2)

wi = KOi + MLTi + MHTi + HTi

(3)

i = ˛0 + ˛1 Inno0i +

T 

˛n Wti + εi

(4)

t=0

where innovations at time t (Innoti ) depend on innovations in t − 1 (Innot−1i ), a set of observable case- and time-variant attributes (Wti ), a set of time-invariant but also observable attributes (Wi ), and a set of unobservable and time-invariant idiosyncratic characteristics (i ). In Eq. (1b), the innovative behaviors are differentiated in order to observe the differential impact (coefficients). The common practice in the literature is to include the different behaviors as dummy variables—or to run the estimations separately—to see how different behaviors change the average level of the estimations. In this case, behaviors will be included as an interactive (multiplicative) term, which implies that the way innovations were sought and eventually achieved affects the way they impact future results. The observable characteristics (Wti ) were included based on the availability of information and the findings from Section 2.2. To control innovation investments, two indicators were constructed: the intensity (II) and the balance (IB) of the innovation expenditure. II is the ratio of expenditures on innovation activities to sales. The underlying assumption is that the higher the expenditure, the greater the firm’s commitment to the quest for technological and organizational improvements, with the consequent impact on the achievement of innovations; an assumption widely confirmed in the persistence literature (Frenz and Prevezer, 2012; Peters, 2009; Raymond et al., 2010). The balance of the innovation expenditure (IB) is another aspect that has been proven to impact the firm’s innovative dynamic (Freeman, 1974; Jensen et al., 2007; Lugones et al., 2007; Yoguel et al., 2011). The argument indicates that the incorporation of knowledge developed outside the firm (exogenous knowledge) combined with the endogenous creation of new knowledge allows the firm to absorb and transform knowledge into innovations.

4 The endogeneity problem that could arise from the interaction between innovations and innovative activities is one limitation that this solution cannot overcome. Dynamic probit models are run on the assumption of strict exogeneity of the explanatory variables. However, since dynamic models control serial correlation, even when this type of endogeneity may actually exist, coefficients have been proved to be consistent (Wooldridge, 2005).

Please cite this article in press as: Suárez, D., Persistence of innovation in unstable environments: Continuity and change in the firm’s innovative behavior. Res. Policy (2013), http://dx.doi.org/10.1016/j.respol.2013.10.002

G Model RESPOL-2924; No. of Pages 11 6

ARTICLE IN PRESS D. Suárez / Research Policy xxx (2013) xxx–xxx

Consequently, IB is a dummy variable equal to 1 if the firm combined investments in the endogenous creation of knowledge (R&D, training, or engineering and industrial design) with the exogenous acquisition of knowledge (capital goods, hardware, software, consulting, or technology transfer).5 The inclusion of capabilities as a dimension was based on the skills and linkages variables. The relative endowment of skilled human resources (QHR) was estimated as the coefficient of professionals to total employment, on the assumption that more years of formal education implies higher skill levels. Linkages between the firm and the environment is another variable included as a proxy of capabilities. Link is a dummy variable equal to 1 if the firm reported interactions with other agents of the environment (with clients, suppliers, competitors, science, technology and education institutions, public agencies, and private labs) and 0 if it did not. Theory and evidence allow us to sustain that interactions happen when the firm has crossed a threshold of minimum competencies (Cohen and Levinthal, 1990). Therefore, if the firm has established linkages, it has crossed that minimum threshold of competences and has higher skills than a firm who has not interacted. Unfortunately, the information collected during the period 2002–2004 is not comparable with the other two exercises, so it was excluded from the analysis. To approximate the availability of financial resources, access to external funding to finance innovation was included (public or private banks, development agencies, the value chain). ER assumes 1 if yes and 0 if no, for each sub-period. In the Argentinean case, the lack of resources has been signaled in all surveys as the most important obstacle to achieving innovations. Thus, if the firm has managed to overcome that barrier it will be in a better financial position. In the literature, this variable usually refers to access to public funds. In this case, the rate of access to this type of funds is remarkably low (around 2.5% of the panel), so its predictive value is also low. Since no changes were made to public policy to foster innovation during the period under analysis, nor were there significant changes among firms that accessed public funds, the use of a more general approach to the possibility of funding innovations seems more accurate. In order to control for firms’ structural characteristics, the usual variables were included: size, capital ownership, and sector. For the latter, firms were grouped into four categories based on the technological intensity of the sector: high-tech (HT), medium-hightech (MHT), medium-low-tech (MT), and low-tech (LT).6 The size control refers to the total number of people employed by the firm at the end of each sub-period (Labor). A “company with participation of foreign capital” is defined as a firm that has more than 1% of its shares owned by foreign capital in t-1 (OK).7 4. Results 4.1. Transition probabilities A simple way to obtain an initial approximation of persistence is by analyzing transition probabilities. Although such schemes lack basic controls, they are used to understand what is happening with the panel and firms’ trajectories. Fig. 2 illustrates the

5 This combination is based on Lugones’s et al. (2007) taxonomy, which arises from the weighted participation of the different innovation activities in the total efforts (see Appendix 2). A similar taxonomy is observed in Jensen et al. (2007), although based on the combination of R&D and non-R&D investments. 6 High-tech industries include ISIC classification 353, 2423, 30, 32, 33; mediumhigh includes: 31, 34, 24 (excl. 2423), 352, 359, 29; medium-low-tech: 351, 25, 23, 26–28; low-tech: 36, 37, 20–22, 15–19. (OECD, 1997). 7 On average, 75% of shares in firms owned by foreign capital ownership belong to non-national capital, so this dummy mostly refers to international firms.

Yes 221 Yes 470 No 249

Yes 97 No 320 No 223 1998-2001

2002–2004

Yes 152 No 69 Yes 157 No 92 Yes 53 No 44 Yes 55 No 168 2005-2006

Fig. 2. Firms with innovation results—Transition probabilities—Total Panel. “Yes (No)” means that the firm has (has not) achieved innovations. Inside the boxes: number of firms; inside the brackets: probabilities as compared to initial condition (1998–2001). Source: own elaboration based on INDEC (2010).

probability of having innovated subject to past results. This scheme, albeit preliminary, shows that the persistence of innovation has a decreasing trend: between 2002 and 2004 (hereinafter t − 1) 47% of firms with innovation results had innovated in the previous period (1998–2001, hereinafter t − 2), and between 2005 and 2006 (hereinafter t) the ratio dropped to 32%. Another interesting fact is the apparent correlation between the initial condition and the final situation: regardless of the result in t − 1, if the firm was an innovator in t − 2, the probability of sustaining this situation in t was 66%. The other way around, almost seven out of ten of firms that were non-innovative in t − 2 remained so at the end of the period—again, regardless of their situation in t − 1. In this sense, persistence seems to exist in negative terms (not being an innovator is maintained over time) but not in positive ones (achieving innovations in the previous period does not seem, prima facie, to be associated with innovations in the following period). Table 1 presents the same transition probabilities but distinguishes firms according to the innovative behaviors defined in Section 3.1: continuous, sporadic, and new innovative firms. The rate of firms where persistence is confirmed is provided between brackets. As expected, higher levels of both innovation and persistence are observed among the continuous firms, which contrasts with the low levels registered among the sporadic and noninnovative firms. Among the new innovative firms, results at the beginning of the period (t − 2 and t − 1) resemble the levels of noninnovative and sporadic firms but they get close to the values of continuous firms at the end (t). 4.2. Econometric estimations Table 2 presents the estimation of the model. Results show that persistence is not corroborated for the general panel (Model a), while the innovative behaviors show, as expected, positive and significant impacts on innovations. When the innovative behaviors are differentiated (Model b), persistence is confirmed with a positive and significant sign for the continuous and new innovative firms,

Please cite this article in press as: Suárez, D., Persistence of innovation in unstable environments: Continuity and change in the firm’s innovative behavior. Res. Policy (2013), http://dx.doi.org/10.1016/j.respol.2013.10.002

G Model RESPOL-2924; No. of Pages 11

ARTICLE IN PRESS D. Suárez / Research Policy xxx (2013) xxx–xxx

Table 1 Innovation persistence and strategies (% of firms). Sub-period

Inno

1998–2001

2002–2004

2005–2006

Innovative 94.15

Continuous 48.31 (48)

80.49 84.85

51.22 (50) Sporadic 42.42 (45)

85.11 Non-Innovative 31.67

34.04 (30) New 55.00 (71)

4.00 14.93

44.00 (100a ) NI 47.76 (40)

9.86

0.70 (0)

Continuous 81.54 (40) Sporadic 2.44 (0) New 63.64 (24) NI 4.26 (5) Continuous 76.67 (53) Sporadic 4.00 (0) New 52.24 (20) NI 0 (0)

Source: Own elaboration based on INDEC (2010). Obs.: 800. Inno: at least one innovation during the sub-period, % of firms. In brackets: % of firms that innovated in the current and previous sub-periods. a Only 1 firm innovated in 1998–2001.

but not among the sporadic group. As a result, the findings only partially corroborate what it is predicted by the persistence literature: past innovations impact future ones only when different innovative behaviors are included and only for new and continuous innovative firms. Regarding the innovative dimensions, both models present similar results, namely those predicted by the literature: firms with a dynamic innovative profile (higher intensity, balanced investments, qualified human resources, linkages, and access to external funds) have higher probabilities of achieving innovations. Although the average innovation expenditure (m II) is just a proxy of the

7

firm’s unobserved characteristics, its positive and significant coefficient is an indication of the importance of sustained investments in these activities (that is, continuous innovative behavior). With regard to the firm’s capabilities, linkages in the last subperiod have a positive and significant impact on the probability of innovating but, surprisingly, the level of qualified human resources does not. Although this result coincides with a similar study (Huergo and Moreno, 2011), it contradicts the findings of several authors based on numerous empirical studies, both for developed and developing countries (Lugones et al., 2007; Jensen et al., 2007; Yoguel et al., 2011, among others). Further studies would certainly shed light on this lack of correlation. Regarding the unobservable characteristics, the impact of the initial condition remains significant, which confirms the correlation between the initial and the final sub-period. However, while the unobserved characteristics of firm () have a positive and significant impact for the whole panel, they lose significance when the behaviors are included. This change signals the usefulness of the concept of innovative behavior—and how it was estimated—as long as it reduces the impact of unobserved effects. 4.3. Robustness check In order to check the sensibility of the coefficients to the selected model, three additional estimations were performed (Table 3). In order to simplify the comparisons, only model b was included (RE2). RE1 corresponds to the estimation of a dynamic random effect probit model without the innovative dimensions (investments, linkages, and external resources). In this case, the only change is the significance and sign of past innovations among sporadic firms. This change, however, reinforces our hypothesis about the isolated nature of innovations among this group of firms. PA corresponds to a dynamic population-averaged probit model, which means that the probability of persistence of each case is compared to the sample’s probability. ZI corresponds to the

Table 2 Random effects probit model—Dep. Variable: Inno. Model a—Eq. (1a)

Model b—Eq. (1b)

Coef.

Std. Err.

−0.11371

0.18622

Marg. Eff.

Structural equation Linnot Linno contt Linno newt Linno spot IIt IBt QHRt Linkt ERt MLT MHT HT KO t−2 Labort ˇ0

2.22172** 0.56286* −0.58931 1.25230* 0.51484* 0.01713 0.08722 0.17263 −0.09346 0.19970** 0.00050 −1.09824*

1.08117 0.11684 0.42623 0.13997 0.14228 0.10957 0.10728 0.19072 0.12424 0.10136 0.00067 0.12070

0.66912 0.16952

Individual heterogeneity Inno0 m II m QHR m labor Ln u u 

0.40128** 3.50856** 1.09146 0.00001 −1.29614 0.52305 0.21481**

0.16054 1.57693 0.77593 0.00065 0.71281 0.18642 0.12023

0.12085 1.05667

0.37716 0.15505

0.06014

Coef.

0.21226** 0.60831* −0.11816 1.71586*** 0.46439* −0.50871 1.14088* 0.41943* 0.00586 0.07615 0.13155 −0.08718 0.15254*** 0.00044 −0.99970* 0.20512** 2.95395** 0.91206 −0.00001 −11.60932 0.00301 0.00001

Std. Err.

Marg. Eff.

0.09587 0.15734 0.15761 0.99292 0.10045 0.39130 0.10585 0.12212 0.09248 0.09076 0.16110 0.10519 0.08771 0.00060 0.08413

0.066273 0.189932

0.08979 1.35436 0.66827 0.00058

0.064044 0.922315

0.535745 0.144997 0.356219 0.13096

0.047626

Source: Own elaboration based on INDEC (2010). Obs.: 2400. ***, ** and * indicate significance at 99%, 95% and 90%, respectively. Obs.: 800. m : time-average of the corresponding variable. T-3 and link t-2 were omitted due to collinearity, Marginal effects based on Delta method (Inno0 = 1 and ␮i =0). Estimations based on Gauss–Hermite, 12 quadrature points. Quadrature checks, no variations over 1%.

Please cite this article in press as: Suárez, D., Persistence of innovation in unstable environments: Continuity and change in the firm’s innovative behavior. Res. Policy (2013), http://dx.doi.org/10.1016/j.respol.2013.10.002

G Model RESPOL-2924; No. of Pages 11

ARTICLE IN PRESS

8

D. Suárez / Research Policy xxx (2013) xxx–xxx

Table 3 Robustness check on persistence—Dynamic probit—Dep. Variable: Inno. RE1 Coef. Structural equation 0.31815*** Linno contt 0.56477*** Linno newt −0.36999** Linno spot IIt IBt QHRt Linkt ERt 0.12583 MLT MHT 0.26067*** HT 0.42170*** KO 0.09561 −0.35103*** t−2 0.00097* Labort −0.51899*** ˇ0 Individual heterogeneity Inno0 0.42545*** m II −0.00046 m QHR m labor Ln u −12.37319 u 0.00206  0.00000

RE2

PA

ZI

Std. Err.

Coef.

Std. Err.

Coef.

Std. Err.

Coef.

Std. Err.

0.08735 0.14898 0.15147

0.21226** 0.60831*** −0.11816 1.71586* 0.46439*** −0.50871 1.14088*** 0.41943*** 0.00586 0.07615 0.13155 −0.08718 0.15254 0.00044 −0.99970***

0.09587 0.15734 0.15761 0.99292 0.10045 0.39130 0.10585 0.12212 0.09248 0.09076 0.16110 0.10519 0.08771 0.00060 0.08413

0.26646*** 0.66721*** −0.07507 1.7452* 0.46414*** −0.50072 1.13816*** 0.41567*** 0.00506 0.07480 0.13056 −0.08857 0.14111 0.00044 −0.99889***

0.09572 0.15938 0.15748 1.05044 0.10049 0.39258 0.10515 0.12194 0.09205 0.08965 0.15897 0.10502 0.08810 0.00059 0.08288

0.21164*** 0.50170*** −0.21938 0.20173 0.23913*** −0.06143 0.77872*** 0.22355*** 0.02127 0.09142 0.12651 −0.02184 0.20088** 0.00001 −1.57917***

0.05617 0.09551 0.17413 0.22621 0.04973 0.05741 0.08198 0.05430 0.06832 0.06109 0.07870 0.05833 0.08418 0.00017 0.08508

0.08979 1.35436 0.66827 0.00058

0.17738** 2.87883** 0.89712 −0.00001

0.08873 1.48212 0.68496 0.00057

0.23868*** 1.56399** 0.02531 0.00016

0.06854 0.77255 0.12242 0.00018

−1.97873*** −5.19055***

0.03127 0.02911

0.08684 0.08384 0.14257 0.09375 0.06892 0.00059 0.06892 0.08122 0.00058

0.20512** 2.95395** 0.91206 −0.00001 −11.60932 0.00301 0.00001

Inflated variable Inno0 Constant

Source: Own elaboration based on INDEC (2010). Obs.: 2400. ***, ** and * indicate significance at 99%, 95% and 90% level. RE/PA/ZI: random effect/population-averaged/zero inflated. m time-average value of the corresponding variable. Link t-2 omitted due to collinearity. RE1 & RE2: Gauss–Hermite estimation, 12 quadrature points, no variations over 1%.

estimation of a zero-inflated model where a Poisson regression is run first, to differentiate firms with and without innovations at the initial moment, and a probit estimation is carried out afterwards, this time, among firms where the inflated variable is equal to 1 (Inno0 ). In both estimations (PA and ZI), the results confirms the initial findings, with the same sign and similar coefficients among the analyzed variables, which also account for the robustness of the random effect dynamic probit model run initially. 4.4. Discussion of the findings and implications for the hypothesis To return to the hypotheses, the positive relationship between past and present innovations (H1) is rejected. This result is consistent with what was discussed in Section 2.3, in the sense that in unstable environments additional effects of past innovations are not necessarily observed. In terms of the path dependence and capability accumulation approaches, the results suggest that given a change in the environmental conditions, accumulated resources, and capabilities may no longer be useful or sufficient for the new environment (the 1998–2001 recession combined with an appreciated exchange rate required different capabilities than the 2005–2006 growth phase with a devalued currency). In terms of the market power approach, changes in the market, in the institutions regulating the competition and in the income-elasticity of the demand could alter the distribution of the market share, allowing new firms to enter and forcing others to leave. The environment based on liberal market conditions that lasted up to 2002 was very different from the populist 2002–2006 administration, under which strong support was provided to local industries together with restrictions on imports. The positive relationship between the continuity of the innovative behavior and the persistence of innovation (H2) is confirmed, although marginal effects show that the highest persistence levels

are observed among the new innovative group. The lower coefficient among the continuous group can be explained by the lock-in effect of past innovation projects. However, as the rest of the determinants of innovations are controlled (investments, capabilities, and resources), the positive and significant sign of past innovations means that previous innovations have triggered additional feedbacks and accumulation processes that increased the probability of achieving innovations in the present associated, theoretically, with higher levels of capabilities and previously acquired resources. For example, during the 1990s, some firms improved their productivity via the combination of external technology with endogenous knowledge creation (Lugones et al., 2007). These firms were in a better position to expand their sales to meet local demand once the recession had been overcome and the domestic market was driving the economic growth. However, the change in the relative prices in favor of the labor component impacted the possibilities of continuing to import technology, and process innovations have to be based on domestic investments. Finding a new supplier of capital goods is not automatic, especially when local producers of these goods almost disappeared during the previous decade. Therefore, these firms had to adapt their behavior to the new context, which delayed the achieving of positive results. Regarding sporadic innovative firms, the lack of impact on persistence resembles the characteristics of non-innovative firms and confirms H4. Among these firms, there is no a connection between past and present results and although they do innovate, innovations are isolated events without additional effects in terms of the firm’s future innovative performance. This is the case with those firms that exported based only on a devalued exchange rate during 2002–2004, without further investments in genuine productivity gains (Porta and Bianco, 2007). When local inflation eroded price-competitiveness of the exchange rate, these firms had not accumulated enough competences to continue exporting. Another

Please cite this article in press as: Suárez, D., Persistence of innovation in unstable environments: Continuity and change in the firm’s innovative behavior. Res. Policy (2013), http://dx.doi.org/10.1016/j.respol.2013.10.002

G Model RESPOL-2924; No. of Pages 11

ARTICLE IN PRESS D. Suárez / Research Policy xxx (2013) xxx–xxx

example is the above-mentioned case of firms that based their innovative behavior on the isolated incorporation of machinery. The introduction of equipment automatically leads to a process innovation. However, there are now a priori reasons to expect further innovations unless the firm decides to invest in complementary improvements. The highest level of persistence is observed among new innovative firms, and the positive and significant sign confirms H3. New innovative firms started the innovation process looking for innovations that were suitable for the new environment and since they were non-innovative during the previous phase, they faced low (and even no) lock-in or path-dependence effects. This is the case for those firms that started exporting based on a the new relative prices after the 2002 devaluation and decided to carry on with an innovative behavior capable of exploiting the exogenous advantage (the devalued exchange rate) with genuine productivity improvements oriented toward sustaining the competitive advantage. Unlike continuous innovative firms, these enterprises initiated their innovative projects knowing the relative prices. Unlike sporadic innovative firms, they took advantage of the competitiveness shock and used it not just to export but also to seek genuine productivity gains, for instance, scale economies. As was the case with continuous firms, once they became part of into the innovator’s club, the way these firms sought technological and organizational improvements created additional feedbacks and accumulation processes which increased their odds of persisting with innovation. With regard to the reviewed empirical evidence, the partial confirmation of persistence agrees with those analyses where persistence was confirmed only for certain groups of firms. For Raymond et al. (2010), there is genuine persistence only for medium-high- and high-tech companies; for Clausen et al. (2011), only when the innovation strategies are based on science or market-oriented firm behavior; for Le Bas et al. (2011), when organizational innovations are achieved. In other words, persistence is confirmed when the analysis is conditioned to a particular type of innovation process. In terms the disagreements, our findings suggest that persistence is more linked to the firm’s innovative behavior than to the results it has achieved. If one assumes that even in stable environments firms can change their behavior, the average persistence for the whole panel—as is the case in Peters (2009) and Frenz and Prevezer (2012)—would be explained by the combination of spurious persistence among firms with sporadic innovative behavior and genuine persistence among firms with more coherent and articulated innovative dynamics. At the same time, and coming back to the research questions that motivated this article, when comparing this study with those made by other colleagues, firms that have faced decades of macroeconomic stability (Dutch companies in Raymond et al., German firms in Peters, and Norwegian ones in Clausen et al.) are compared with firms which went through periods of great turbulence and changes in the rules of the game. Consequently, it does not seem appropriate to condone the use of similar empirical approaches or to expect the innovative dynamic and the firms’ innovative paths to fall within the same theoretical developments. In this respect, our findings suggest that if one wants to apply the concept of persistence to stable and unstable environments, then we have to assume that the innovative behavior of some firms leads them to achieve isolated positive results while that of others allows them to innovate persistently. 5. Conclusions The aim of this paper was to review the concept of innovation persistence in relation to unstable environments, accepting the possibility of changes in firms’ innovative behavior. The hypotheses

9

stated that firms with continuous (sustained) innovative behavior will be more likely to continue achieving positive results, while firms with sporadic innovative behavior will show no persistence at all. In between these behaviors, some firms could have reacted to the changes in the environment and decided to seek a competitive advantage based on innovation. Within these firms, persistence should appear together with the change in the environment. To test these hypotheses, firms were classified according to the frequency of their innovation expenditure, distinguishing between those making sustained investments (continuous behavior), those making isolated investments (sporadic behavior), and those that changed from non-innovative to innovative behavior (new innovative firms). The triple relationship between past and present innovations and firm behavior was then tested with a dynamic random effect probit model, controlling both the unobservable effects and the initial condition. The model was applied to a group of Argentinean manufacturing firms using data from the national innovation surveys for the period 1998–2006, which coincides with a period of macroeconomic instability. The results suggest that in unstable environments, an automatic and linear relationship between past innovations, present innovative behaviors, and future results cannot be assumed. The analysis based on the continuity of the innovation expenditures showed that persistence is linked to sustained behavior but it is not exclusively determined by it. More years of expenditure do not increase the odds of persistence when the firm is operating in an unstable environment. In this respect, path dependence could delay the achievement of results since the firm has to adjust its behavior to the new scenario. Conversely, the persistence levels among new innovative firms show path independence and raise questions about the ability firms have to respond quickly to changes in the environment. The relationship between the firm and the environment is intrinsically dynamic: the environment can change and the firm can react to those changes. As such, a more dynamic understanding of the innovative behavior of the firm is required. Finally, a limitation of this study—and also an unsolved problem of persistence literature—is the assumption about the time that needs to elapse between the implementation of an innovation project and its impacts in terms of the feedbacks and accumulation processes that lead to new innovations. The time that R&D investments take to impact innovations is probably different from the time that takes new machinery to impact production. In the literature, the estimated lags seem to be explained more by the availability of information than by a theoretical framework about the time window considered in the analysis. Evidently, an average delay is better than no delay at all, and only the increase in the quantity and quality of available information will allow for more complex approaches to the relationship between past actions and present results. However, our results suggest that the combination of different innovative investments impacts persistence, which means that firms pursuing only R&D activities or acquiring only external embodied knowledge have different probabilities of persistence than firms combining different investments. This reinforces the importance of analyzing innovative investments in terms of the specific innovative projects in order to partially acknowledge the specific timing each project could have.

Acknowledgements This paper is part of the author’s PhD thesis, entitled: “Innovative strategies in unstable environments: the case of Argentinean firms”, adscript at Aalborg University, Denmark, and Quilmes National University, Argentina. The author wants to thank B. Johnson, B.Å. Lundvall, G. Yoguel, F. Barletta, M. Pereira, V. Roberts and G. Montes Rojas for their comments. This paper was also enriched

Please cite this article in press as: Suárez, D., Persistence of innovation in unstable environments: Continuity and change in the firm’s innovative behavior. Res. Policy (2013), http://dx.doi.org/10.1016/j.respol.2013.10.002

ARTICLE IN PRESS

G Model RESPOL-2924; No. of Pages 11 10

D. Suárez / Research Policy xxx (2013) xxx–xxx

Table A distribution of the innovative behaviors (number of firms). 1998–2004

2005–2006

Continuous New Sporadic Non-innovative Total

Continuous

New

Sporadic

Non-innovative

Total

325 120 0 0 445

0 0 33 67 100

39 25 0 0 64

2 0 47 142 191

366 145 80 209 800

Source: Own elaboration based on INDEC (2010)

Table B Correlation matrix. Innoi

IIi

t−2

t−1

t

Equal weights were assigned to each category (0.25) in order to analyze the values obtained by different groups of firms: the strategies go from the perfectly balanced firm (index equal to 1) to the perfectly unbalanced (values tending to 0). Based on this index, firms were ranged between balanced (an index higher than 0.5) and biased (and index lower than 0.5). A dummy variable (rather than a continuous one) was selected because of the different extra-innovation elements affecting the distribution of the innovation expenditure (knowledge vs. capital intensive sectors, labor-capital composition, scale). With this dummy variable, we can still look at more balanced firms without penalizing them for not having perfectly balanced behavior.

IBi

t−2

t−1

t

QHRi

t−2

t−1

t

Linki

t−2

t−1

t

ERi

t−2

t

Innoi

t−2 t−1 t

1 0.177* 0.325*

1 0.201*

1

IIi

t−2 t−1 t

0.277* 0.149* 0.183*

0.074 0.141* 0.146*

0.133* 0.229* 0.342*

1 0.181* 0.113*

1 0.289*

1

IBi

t−2 t−1 t

0.629* 0.388* 0.365*

0.133* 0.183* 0.240*

0.299* 0.490* 0.645*

0.209* 0.099* 0.096*

0.119* 0.300* 0.192*

0.124* 0.211* 0.221*

1 0.421* 0.382*

1 0.609*

0.232 0.08 0.162*

0.07 0.072 0.111*

*

0.178 0.092* 0.116*

0.056 0.002 0.035

*

0.217 0.341* 0.191*

*

0.132 0.111* 0.103*

*

0.272 0.036 0.182*

*

QHRi

t−2 t−1 t

*

0.200 0.144* 0.210*

0.201 0.09 0.167*

1 0.459* 0.614*

1 0.444*

1

Linki

t−2 t

0.539* 0.270*

0.122* 0.110*

0.248* 0.503*

0.161* 0.103*

0.133* 0.206*

0.119* 0.243*

0.441* 0.230*

0.287* 0.445*

0.252* 0.472*

0.229* 0.199*

0.086 0.096*

0.179* 0.199*

1 0.225*

1

ERi

t−2 t−1 t

0.325* 0.098* 0.193*

0.075 0.086 0.054

0.077 0.136* 0.271*

0.288* 0.014 0.056

0.064 0.174* 0.091*

0.079 0.149* 0.310*

0.332* 0.056 0.164*

0.162* 0.236* 0.199*

0.116* 0.147* 0.226*

0.048 0.051 0.078

0.001 0.133* 0.102*

0.025 0.067 0.065

0.256* 0.055 0.133*

0.052 0.136* 0.179*

t−2

t−1

1 0.112* 0.120*

1 0.342*

t

1

1

Source: Own elaboration based on INDEC (2010). Pearson’s correlations. Obs.: 800. * Significant at 99%.

by the exchanges and contributions from the Globelics, IKE, and SIDPA seminars. Appendix 1. Descriptive statistics See Tables A and B. Appendix 2. Innovation balance index (based on Lugones et al., 2007) The IB index is formally written: IBi = 1 −

n  |(gj AIj ∗ ˛j ) − 1|

n

where 0 < IBi ≤ 1.

j=1

Category (j)

Description

Weight (␣)

a b c d

Research and development (internal and external) Engineering and industrial design + training Capital goods + hardware Technology transfer + consulting + software

0.25 0.25 0.25 0.25

where i is the identifier of the firm, j identifies each category of expenditure (a to b), g is the expenditures in each category (j), AI is the cumulative total expenditures on innovation activities (in constant prices 1998), ˛ is the weighting coefficient for each j (in this case 0.25), n is the total number of categories analyzed (n = 4). The index is reduced to the group of firms that invested in innovation, regardless of the results. For the non-innovative firms, the IB was set to zero.

References Antonelli, C., 1997. The economics of path-dependence in industrial organization. Int. J. Ind. Organ. 15, 643–675. Antonelli, C., 2008. Localized Technological Change. Towards the Economics of Complexity. Routledge, London and New York. Clausen, T., Pohjola, M., Sapprasert, K., Verspagen, B., 2011. Innovation strategies as a source of persistent innovation. Ind. Corp. Change 22, 33–72. Cohen, W., Levinthal, D., 1990. Absorptive capacity: a new perspective on learning and innovation. Adm. Sci. Q. 35, 128–152. Freeman, C., 1974. The Economics of Industrial Innovation. Penguin Modern Economic Text, Great Britian. Frenz, M., Prevezer, M., 2012. What Can CIS data tell us about technological regimes and persistence of innovation? Ind. Innovat. 19, 285–306. Geroski, P.A., Van Reenen, J., Walters, C.F., 1997. How persistently do firms innovate? Res. Policy 26, 33–48. Huergo, E., Moreno, L., 2011. Does history matter for the relationship between R&D, innovation, and productivity? Ind. Corp. Change 20, 1335–1368. ˜ Empresarial. INDEC, Buenos Aires. INDEC, 2010. Base Integrada de Desempeno Jensen, M.B., Johnson, B., Lorenz, E., Lundvall, B.Å., 2007. Forms of knowledge and modes of innovation. Res. Policy 36, 680–693. Le Bas, C., Mothe, C., Nguyen Thi, T.U., 2011. Technological Innovation Persistence: Literature Survey and Exploration of the Role of Organizational Innovation. SSRN eLibrary. Lugones, G., Suarez, D., 2010. STI indicators for policy making in developing countries: An overview of experiences and lessons learned, UNCTAC, Conference Room Paper, Multi-Year Expert Meeting on Enterprise Development Policies and Capacity-Building in Science, Technology and Innovation, Ginebra. Lugones, G., Suarez, D., Le Clech, N., 2007. Innovative Behaviour and Its Impact on Firms” Performance, Micro Evidence on Innovation in Developing Countries. UNU-MERIT, Maastricht. Lundvall, B.Å., 1992. National System of Innovation: Towards a Theory of Innovation and Interactive Learning. Pinter, London. Malerba, F., Orsenigo, L., Peretto, P., 1997. Persistence of innovative activities, sectoral patterns of innovation and international technological specialization. Int. J. Ind. Organ. 15, 801–826.

Please cite this article in press as: Suárez, D., Persistence of innovation in unstable environments: Continuity and change in the firm’s innovative behavior. Res. Policy (2013), http://dx.doi.org/10.1016/j.respol.2013.10.002

G Model RESPOL-2924; No. of Pages 11

ARTICLE IN PRESS D. Suárez / Research Policy xxx (2013) xxx–xxx

Mansfield, E., 1962. Entry, Gibrat’s law, innovation, and the growth of firms. Am. Econ. Rev. 52, 1023–1051. Nelson, R., 1991. The role of firm differences in an evolutionary theory of technical advance. Sci. Public Policy 18, 347–352. Nelson, R., 1994. The co-evolution of technology, industrial structure, and supporting institutions. Ind. Corp. Change 3, 47–63. Nelson, R., Winter, S., 1982. An Evolutionary Theory of Economic Change. The Belknap Press of Harvard University Press, Cambridge. OECD, 1997. Revision of the high-technology sector and product classification. In: Hatzichronoglou, T. (Ed.), OECD, Science, Technology and Industry Working Papers, 1997/2. OECD, 2005. Oslo Manual – 3rd ed. Guidelines for Collecting and Interpreting Innovation Data, First edition 1992. OECD. Penrose, E., 1959. The Theory of the Growth of the Firm. Oxford University Press, Oxford. Peters, B., 2009. Persistence of innovation: stylised facts and panel data evidence. J. Technol. Trans. 34, 226–243.

11

Phillips, A., 1971. Technology and Market Structure: A Study of the Aircraft Industry. Heath, Lexington, Mass. Porta, F., Bianco, C., 2007. Especializacion productiva e insercion internacional. Evidencias y reflexiones sobre el caso argentino. PNUD. Raymond, W., Mohnen, P., Palm, F., van der Loeff, S., 2010. Persistence of innovation in Dutch manufacturing: is it spurious? Rev. Econ. Stat. 92, 495–504. Schumpeter, J., 1942. Capitalism, Socialism and Democracy. Harper and Brothers, New York. Wooldridge, J.M., 2005. Simple solutions to the initial conditions problem in dynamic, nonlinear panel data models with unobserved heterogeneity. J. Appl. Econometrics 20, 39–54. Yoguel, G., Robert, V., Pereira, M., Barletta, F., 2011. The effects of feedbacks on firms’ productivity growth. In: Micro, macro, and meso determinants of productivity growth in Argentinian firms, IX Globelics Conference, Buenos Aires, Argentina.

Please cite this article in press as: Suárez, D., Persistence of innovation in unstable environments: Continuity and change in the firm’s innovative behavior. Res. Policy (2013), http://dx.doi.org/10.1016/j.respol.2013.10.002