Innovation and economic performance in services: a firm-level analysis

Transcription

1 Cambridge Journal of Economics 2006, 30, doi: /cje/bei067 Advance Access publication 8 August, 2005 Innovation and economic performance in services: a firm-level analysis Giulio Cainelli, Rinaldo Evangelista and Maria Savona* This paper explores the two-way relationship between innovation and economic performance in services using a longitudinal firm-level dataset which matches data from the second Community Innovation Survey, CIS II ( ), against a set of economic variables provided by the System of Enterprise Accounts ( ). The results presented show that innovation is positively affected by past economic performance and that innovation activities (especially investments in ICTs) have a positive impact on both growth and productivity. Furthermore, productivity and innovation act as a self-reinforcing mechanism, which further boosts economic performance. These findings provide empirical support for the endogenous nature of innovation in services and the presence in this sector of competition models and selection mechanisms based on innovation. Key words: Technological innovation, Economic performance, Service sector JEL classifications: O31, O33, L80 1. Introduction It is widely acknowledged that technological change and innovation are the major drivers of economic growth and are at the very heart of the competitive process. Over the last few decades, a large body of literature on economic growth has attempted to account both theoretically and empirically for such a major issue in economic theory, although from different perspectives and with different approaches. A major theoretical duel is the one between the neoclassically inspired New Growth Theory and the neo-schumpeterian evolutionary approach 1 (see Verspagen, 2005 for a recent reassessment of this debate). Manuscript received 10 March 2003; final version received 6 June Addresses for correspondence: Giulio Cainelli, University of Bari, and CERIS-CNR, Milan, Italy; cainelli@idse.mi.cnr.it; Rinaldo Evangelista, IRPPS-CNR, Rome and University of Camerino, Italy; r.evangelista@irpps.cnr.it; and Maria Savona, SPRU, Science and Technology Policy Research, University of Sussex (UK) and BETA, Bureau d Economie Théorique et Appliquée UMRCNRS7522Pôle Européen de Gestion et d Economie, Strasbourg, France; savona@cournot.u-strasbg.fr * University of Bari, and CERIS-CNR, Milan; IRPPS-CNR, Rome and University of Camerino; and SPRU, UK, and BETA, Strasbourg, respectively. The authors thank Giulio Perani (Italian National Institute of Statistics ISTAT), coordinator of a research group on Technological innovation in services, who provided the firm-level dataset used for the empirical analysis. The authors are also grateful to Daniele Archibugi, Nick von Tunzelmann, Roberto Zoboli and three anonymous referees for their valuable comments on a previous draft of this paper. 1 In the field of New Growth theory, see, among others, Romer (1990), Grossman and Helpman (1991), Bresnahan and Trajtenberg (1995), Helpman (1998), Aghion and Howitt (1992), Griliches (1984, 1995, 1998), in the Schumpeterian stream see, among others, Nelson and Winter (1982, 2002), Dosi et al. (1988), Silverberg and Soete (1994), Nelson (1995), Stoneman (1995), Freeman and Soete (1997) and Archibugi and Michie (1998). Ó The Author Published by Oxford University Press on behalf of the Cambridge Political Economy Society. All rights reserved.

2 436 G. Cainelli, R. Evangelista and M. Savona A common feature of these streams of the literature is their explicit or implicit focus on the manufacturing sector. Services for a long time have been seen as technologically backward, with innovation playing only a marginal role in explaining the aggregate performance of this sector and the competitive strategies of firms. The old debate over the long-term growth of services has been dominated since the late 1960s by Baumol s (1967) cost disease argument, according to which the growth of service activities is the main reason for the productivity slowdown that has affected the advanced countries in the last few decades. 1 It was not until quite recently, with the growth potentiality linked to the new information and communication technologies (ICTs), that this attitude began to change. Over the last decade, a new stream of contributions to the literature has in fact begun to challenge the old view of services as being technologically backward or passive adopters of technology (Miles, 1993, 1995; Miles et al., 1995; Andersen et al., 2000; Metcalfe and Miles, 2000; Gadrey and Gallouj, 2002; Tether, 2003). There is an increasing amount of empirical evidence to support this new perspective. OECD data show that service industries in the advanced countries perform up to one-third of total business R&D (BERD) and account for more than 50% of the R&D embodied in intermediate inputs (ICT hardware) and capital equipment (OECD, 2000A,B,C). The results of the second Community Innovation Survey (CIS II) confirm that innovation activities do occur in the services sector, though to differing extents and in various forms across industries (Evangelista, 2000; EUROSTAT, 2001). Although more is known about the varieties of innovation in services, investigation of its economic impact has been largely ignored, particularly in terms of firm-level analyses. The small number of firm-level studies can to some extent be explained by the difficulty involved in accessing micro-data, which in the case of services is even greater. There are also data constraints and methodological problems related to the availability of appropriate indicators to measure innovation activities in services. Those traditionally used in the manufacturing sector, e.g., R&D and patents are not at all appropriate for services (Evangelista and Sirilli, 1995; Djellal and Gallouj, 1999; Coombs and Miles, 2000). Thus, to study the relationship between technological change and economic performance in services requires different and more comprehensive measures of firms innovation activities. The CIS collected data not just on R&D, but on a much wider spectrum of firms innovation activities (OECD-EUROSTAT, 1997). Despite the potential offered by this data source, only a very few studies have so far used CIS data to explore the relationship between innovation and economic performance at the firm level. Most existing studies have focused on the manufacturing sector (Crepon et al., 1998; Klomp and van Leeuwen, 1999; Evangelista, 1999; Kremp et al., 2004). This paper explores the links between innovation and economic performance in services using longitudinal firm-level data based on CIS II ( ) and a set of economic performance indicators drawn from the Italian System of Enterprise Accounts ( ). These data are used to discover whether innovation has a real impact on the economic performance of service firms and find the extent to which innovation activities are spurred by a firm s economic performance. 1 A whole stream of literature has emerged since then, mainly concerned with de-industrialisation and productivity slowdown in the advanced economies, which has primarily been imputed by such authors to the structural change of the employment composition towards service activities (Fuchs, 1968, 1969; Petit, 1986, 2002; Cohen and Zysman, 1987; Baumol et al. 1989; Baumol, 2002; Wolff, 2002).

3 Innovation and economic performance in services 437 The paper is structured as follows. Section 2 identifies the key links between innovation and economic performance. Section 3 provides a brief description of the dataset and indicators used in the empirical analysis. Section 4 presents the model, and Section 5 presents the empirical results of the econometric analysis. Finally, Section 6 synthesises the main empirical findings and draws some conclusions. 2. The links between innovation and economic performance at the firm level The empirical literature on the relationship between innovation and economic performance has mostly focused on the economic impact of technological change, and tends to overlook the reverse relationship, that is the extent to which innovation is spurred by past economic performance. This section aims to re-establish the two-way nature of this relationship. 2.1 Mechanism A: Innovation as a determinant of economic performance (Schumpeter I) The key role played by innovation in explaining the dynamic properties of firms, industries and economic systems has been acknowledged since the origin of economic thought, as is clear from the works of Smith and Marx, and is nowadays part of the general consensus among economists. The issue was further developed by Joseph Schumpeter, who put innovation at the core of his first major contribution, The Theory of Economic Development (Schumpeter, 1934). In this work, the role of innovation is fully endogenised and conceived first and foremost as an entrepreneurial fact which is the core of competition and the dynamic efficiency of firms and industries. Whatever the primary source of scientific advance and even of technological change, it is the (successful) introduction of product, process and organisational innovations that allows firms to override the preexisting conditions of markets and industries, and to grow and gain market shares at the expense of non-innovating firms. Dynamic rather than static efficiency is what matters in the process of creative destruction brought about by innovation. Innovation allows the firm to build up monopolistic rents which tend to be progressively eroded alongside the imitative diffusion of new products and processes. The importance of this mechanism is nowadays acknowledged by neo-schumpeterian scholars and increasingly by neoclassical economists (Verspagen, 2005). We can summarise the characteristics of such a mechanism, linking firms economic performance to innovation, by labelling it Schumpeterian I (Freeman, 1982). As far as the manufacturing sector is concerned, previous studies found positive effects of innovation on economic performance and more especially on productivity (Griliches, 1995, 1998; Loof and Heshmati, 2001; Crepon et al., 1998; Klomp and van Leeuwen, 1999; Evangelista, 1999; Kremp et al., 2004). What requires to be empirically tested is whether such a mechanism governs the dynamics of firms and industries in the service sector, for which, as already mentioned, the empirical evidence is still very limited Mechanism B: Economic performance as a determinant of innovation activity Schumpeter II Another seminal contribution from Schumpeter, which has become part of our common understanding of innovation, emphasised the costly, risky and uncertain nature of innovation activities and the crucial issue of the appropriability of the economic benefits of 1 Among the few contributions including services, see van der Wiel (2001), Loof and Heshmati (2001), van Leeuwen and van der Wiel (2003) and Evangelista and Savona (2003).

4 438 G. Cainelli, R. Evangelista and M. Savona innovation. In later work, Schumpeter argued that the increasingly scientific base of economic activities had caused innovation to become more and more costly, as a result of indivisibilities and significant economies of scale and scope (Schumpeter, 1942). In the presence of barriers to entry and weak appropriability conditions, large firms and ex ante monopolistic power might be more conducive to innovation than fully competitive markets populated by small firms (Freeman, 1982; Cohen, 1995; Freeman and Soete, 1997). Some of the insights provided by Schumpeter have important implications for the relationship between innovation and economic performance, especially in terms of its direction of causality. The funding of risky, long-term and large-scale innovation projects requires substantial financial resources and is facilitated by healthy economic track records from firms that are associated with high growth rates, large profits and healthy cashflows. 1 Although this line of reasoning mainly refers to manufacturing sectors and technologies, it might also hold for the service industries. However, innovation activities in services are believed to take place on an informal basis and be less dependent on technological breakthroughs. Both these features might reduce the importance of past economic performance as a determinant of innovation. However, innovation activities in some service sectors such as telecommunications, transports and finance are associated with the establishment of expensive technological infrastructures, which requires large financial resources and high demand. Therefore, for firms in these sectors, past economic performance might be more relevant as a basis for their overall financial commitment to innovation but, also in this case, there is no empirical evidence showing the presence and strength of such a link Schmooklerian The endogenous nature of innovation has been pointed too with reference to the role played by demand conditions on the overall pace of technological change and as an incentive for firms to invest in innovation. Markets in the early phases of their life cycle and/ or benefiting from a favourable economic environment, experience sustained growth in demand, which acts as an incentive for the entry of new firms and the growth of incumbents. Both these conditions, coupled with expectations of positive market growth, might act as an important stimulus for innovation activity. The hypothesis that technical change is mainly demand-pulled was proposed by Schmookler (1962, 1966). This hypothesis was empirically supported by the positive correlation found between cycles of inventive effort (proxied by patents, a tolerable assumption ; Schmookler, 1962, p. 119) and cycles of output across industries producing capital goods. The shape of the long-term trend of these two indicators showed that cycles of output were leading cycles of relevant patenting activity in the capital goods industries. Schmookler s claim that technical progress was dependent on economic phenomena sparked much debate about the actual determinants of technical progress. Many scholars tried to test Schmookler s hypothesis empirically at different levels of analysis (among them Scherer (1965, 1982), Mowery and Rosenberg (1979), Stoneman (1979), Walsh (1984) and, more recently, Kleinknecht and Verspagen (1990), Geroski and Walters (1995), Brower and Kleinknecht (1999)). 2 However, these contributions produced controversial results. Kleinknecht et al. and 1 See Hao and Jaffe (1990) and Cohen (1995) for a review of empirical studies. 2 Among these attempts, Kleinknecht and Verspagen tried to test the Schmooklerian hypothesis empirically at the firm-level of analysis, using Dutch firm-level CIS data. The authors re-read the Schmooklerian hypothesis as a co-presence, and mutual interaction between technology-push and demand-pull mechanisms, which in the post-schmooklerian literature had been considered to be mutually exclusive. We look at this issue in considering mechanism C.

5 Innovation and economic performance in services 439 Geroski and Walters found empirical support for the Schmooklerian hypothesis. Geroski and Walters focused on the role of demand to determine whether innovation is more likely to be pro-cyclical or counter-cyclical. 1 It emerges that the direction of the causal relationship is from variations in demand to variations in innovative activity and not the reverse. Brower and Kleinknecht reached the same conclusion, but based on crosssectional rather than panel data. Once again, all these contributions are confined to the manufacturing sector, leaving a gap in the empirical analysis of the role of market demand as an incentive for innovation activity in services. This is somewhat surprising insofar as most of the literature on innovation in services tends to emphasise the co-terminality, that is the close interaction between production and consumption of services (Miles et al., 1995; Gallouj and Weinstein, 1997) and, also, the importance of user producer links in determining the financial effort devoted to innovation by service firms. Further, some scholars have referred to the importance of distinguishing between radical and incremental innovations (Barras, 1986, 1990), with the latter expected to be more sensitive to demand and market conditions. Given that innovation in services is more likely to be incremental in nature and to consist of specific applications of a general purpose technology such as ICT (Helpman, 1998; Freeman and Soete, 1997; Freeman and Loucxã, 2001), the absence of any empirical investigation on the role of demand as an incentive for service firms to innovate is even more striking. A fairly large body of literature has in fact related the increasing importance of services in modern economies to the paradigmatic change brought about by the ICT revolution (Freeman and Soete, 1997; Freeman and Loucxã, 2001; Perez, 2002). Overall, the empirical studies of the Schmooklerian mechanism in the domain of service firms and industries are still at an embryonic stage, and generally ignore the role of demand levels and growth, and demand expectations as determinants of innovation investments and activities. The present empirical study is a first step towards filling this gap. 2.3 Mechanism C: Two-way dynamic link between innovation and economic performance (evolutionary) Mechanisms A and B above cannot be considered to be mutually exclusive. On the contrary, in a dynamic perspective, they work in tandem, reinforcing each other over time. This might be a general dynamic property of an economic system or might hold (and be particularly strong) only in certain contexts: particular sectors, markets, stages of development of industries and technologies, historical periods. In all the cases in which such a phenomenon occurs, the relationship between innovation and economic performance should be conceptualised as being two-way as well as possibly cumulative. The strength of such a mechanism could also be enhanced by the presence of increasing returns to scale, and occurring in sectors and technological regimes characterised by the Verdoorn Kaldorian laws. These latter dynamically link albeit mainly at sectoral and macroeconomic levels labour productivity performance with scale of economic activities and investments (Verdoorn, 1949; Kaldor, 1975, 1978). The presence of a two-way self-reinforcing relationship between innovation and economic performance at firm level is also fully consistent with the evolutionary approach 1 The idea of innovation as being counter-cyclical was supported by Mensch (1975), who argued that innovation activities are in fact triggered by unfavourable economic conditions which put pressure on firms to invest more effort and resources into the innovation process. According to this view, the pace of technical change accelerates in the proximity of a business cycle downturn. See also the works of Kleinknecht (1984, 1987, 1990).

6 440 G. Cainelli, R. Evangelista and M. Savona to technological change and industrial dynamics. Such an approach, starting with the pioneering contribution of Nelson and Winter (1982), has further developed the microfoundations of the Schumpeterian model of competition and growth (Winter, 1984; Dosi, 1988; Dosi and Nelson, 1994; Dosi et al., 1995; Nelson and Winter, 2002). In an evolutionary framework, innovation is seen as the most important competitive weapon for firms in an economic and technological context characterised by high uncertainty, bounded rationality and path dependency. Such features leave room for a broad variety of (best and worst) innovative behaviours and learning processes which tend to create wide asymmetries in both the technological and economic performances of firms. Technological and economic asymmetries reflect (along with chance) differences in the level and quality of past innovation activities and competence building processes, with market forces eventually identifying the most successful. Given the highly cumulative and pathdependent nature of such processes, it is likely that asymmetries in both innovation capabilities and economic performances are not temporary, but will tend to persist and be reinforced over time. Compared with the Verdoorn Kaldorian laws, the evolutionary approach has a more explicit and robust micro-foundation. Therefore, and in line with the empirical agenda of this paper, we label the cumulative mechanism linking economic and innovation performance at the firm level Evolutionary. The lack of longitudinal firm-level data on innovation and economic performance already referred to has hampered a proper empirical testing of the evolutionary hypothesis. The presence of virtuous circles and long-lasting relationships between the innovativeness and economic performance of firms has been demonstrated so far mainly through case studies and qualitative evidence. 1 Furthermore, the literature has focused mainly on the evolutionary trajectories of manufacturing industries and firms. As far as services are concerned, we know very little about the degree of endogeneity of technological change or the relevance of models of competition and selection mechanisms based on innovation. 3. The dataset and indicators Before describing the model and the results of the econometric estimates of mechanisms A, B and C sketched above, it is worth examining the main characteristics of the dataset and indicators used in the empirical analysis. Our investigation is based on a new and original longitudinal firm-level dataset built up by matching data drawn from two different statistical sources: the Italian Community Innovation Survey (CIS II) and the System of the Enterprise Accounts (SEA). The resulting sample of this merging consists of 735 service firms with 20 or more employees for which a wide set of innovative data for the period , and a selected number of economic performance indicators for the period , are available. The statistical representativeness of our sample can be assessed by comparing it with the CIS II population in Table 1. From this table, it can be seen that our sample closely resembles the entire CIS II population in terms of both percentage of innovative firms in total firms and overall structure. The exception is the trade sector, which is slightly underrepresented in our sample. Also, our sample shows a slight bias towards innovative firms. The sector of financial services is not covered because it is not included in the SEA. The indicators used in the econometric estimations are presented in Table 2. The first group of indicators measures different dimensions of firms innovation performance and 1 Among the few exceptions, see Marsili (2001).

7 Innovation and economic performance in services 441 Table 1. A comparison between CIS II (Italy) and the sample used in the empirical analysis CIS II population Selected sample Service sectors Total firms % % Innovating firms to total firms Total firms % % Innovating firms to total firms Trade 8, Hotel & restaurants 2, Transport 2, Waste disposal Software & related R&D, engineering, technical consultancy Legal & marketing Security, cleaning, 2, other business services Post & telecommunication Financial services 1, Total 19, a Financial services are not covered by the Italian System of the Enterprise Accounts. are drawn form CIS II; the second group measures the economic performance of firms over the period The innovation performance indicators As already pointed out, compared with the technological indicators more traditionally used in this field of research, CIS II data provide us with a much richer range of information on firms innovation activities and performances. The most basic information provided by CIS is whether the firm introduced an innovation in the period covered by the survey ( ) and what type of innovation it was (product/service or process innovation). This information allows us first to link the economic performance of firms to the mere presence of innovation (INN) and second to verify whether product and process-oriented strategies (INSERV, INPRO) lead to different economic outcomes (mechanism A). The distinction between a product and a process innovation has long been recognised in the economics of innovation literature as being crucial in order to identify the different strategies of firms. Product innovations are usually associated with more radical and proactive technological strategies, which are expected to bring high economic returns. Process innovations generally prevail in traditional industries and signal the presence of a more defensive technological strategy, often associated with rationalisation and restructuring processes. Most of the empirical evidence supporting this view relates to the manufacturing industry. In the case of services, the economic outcomes of these two types of strategies might be less obvious and require proper empirical testing. In fact, in the case of services, product and process innovations are closely intertwined (Miles, 1995; Gallouj and Weinstein, 1997). Furthermore, it is argued that, in many service industries, it is the introduction of a process innovation that opens the way to improvements in the quality of the service delivered, or even to a completely new set of services (Barras, 1986, 1990).

8 442 G. Cainelli, R. Evangelista and M. Savona Table 2. List of variables used in the econometric estimates Acronym Variable Innovation performance indicators INN Dichotomous variable equal to 1 for firms which have introduced at least one innovation in INPROC Dichotomous variable equal to 1 for firms which have introduced at least one process innovation in INSERV Dichotomous variable equal to 1 for firms which have introduced at least one service innovation in RD-DES R&D, Design, Know How expenditure per employee (Log variable) ICT ICT (software) expenditure per employee (Log variable) INV Capital equipment and ICT hardware expenditure per employee (Log variable) TOTEXP Total innovative expenditure per employee (Log variable) Economic performance indicators SALES Average annual growth rate of sales PROD Average level of productivity (sales per employee) (Log variable) Sector dummies NACE two and three digit classification equivalent TRADE Trade and repair of motor-vehicles (50), Wholesale trade (51), Retail trade (52) HOTELS Hotels and Restaurants (55) TRANSP Land transport (60), Sea transport (61), Air transport (62), Travel and transport agencies (63) WASTE Waste and disposal (90) COMP Software and related (72) R&DCONS R&D (73), Engineering (74.2) and Technical consultancy (74.3) LEGMKT Legal and Accounting (74.1) and Marketing (74.4) OTHBUS Security (74.6), Cleaning (74.7) and Other Business (74.8) Size dummies D20 99 Firms with more than 20 and less than 100 employees D Firms with more than 100 and less than 250 employees D250 Firms with more than 250 employees Along with R&D, the CIS takes into account other fundamental sources of innovation, such as activities related to the design of new services, software development, the acquisition of know-how, investment in new machinery (ICT hardware) and training. Firms were asked to provide quantitative figures on the financial resources devoted to these different activities. These data are particularly important in the case of services, since several studies have already shown that R&D activities and assets play only a marginal role in this sector of the economy and patents are rarely taken out by service firms to protect their innovative output from imitation (Evangelista, 2000; EUROSTAT, 2001). In most service sectors, innovation activities are incremental in nature, require substantial human capital investment and rely upon the acquisition and internal development of ICT. Thus, we built four additional innovation performance indicators which capture: the overall innovative efforts of firms (i.e., total innovation expenditure per employee: TOTEXP); the resources devoted, out of total innovation expenditures, to: (i) R&D, design activities and the acquisition of know-how (RD-DES); (ii) the development or acquisition of new software (ICT); and (iii) innovative investments in capital equipment (INV). These four

9 Innovation and economic performance in services 443 indicators allow us to identify which of these different innovation inputs are the most important in explaining the economic performance of firms (mechanism A) and what kinds of innovation activity are spurred by firms past economic performance and demand factors (mechanisms B1 and B2). 3.2 The economic performance indicators The economic performance indicators used in our econometric investigation are in line with most of the empirical literature referred to in the previous section. We employ two particular economic performance indicators: (i) the average growth rate of sales at current prices over the two sub-periods and , expressed in natural logarithms (SALES); and (ii) the ratio between sales at current prices and number of employees, used as a proxy for labour productivity at current prices. 1 The latter was computed as the natural logarithm of the average values of the ratio in the sub-periods and (PROD9395 and PROD9698). While the rationale behind the use of (i) is straightforward, we need to justify our use of the ratio between sales and the number of employees. This indicator is used to measure both the impact of innovation on the firm s economic performance (mechanism A) and the impact of economic performance on innovation (mechanism B). Innovation can have a positive impact on the sales per employee ratio through either enlarging the numerator or decreasing the denominator. The introduction of new or improved services allows firms to increase their sales in quantitative terms or via a price increase for the service delivered; the introduction of process innovations increases the ratio by reducing the labour content of the service produced and delivered. Using the ratio between sales and employees also seems an obvious way to capture the impact of economic performance on innovation. It is a good proxy for the total amount of resources that a firm has available to finance its innovation activity. Moreover, the use of a level indicator turns out given the time-span of the data at our disposal to be a more reliable proxy for structural differences in economic performance across firms. In fact, the level of productivity tends to capture not only the firm s static efficiency, but also its dynamic efficiency, which in turn results from the technological investments made in the past. In other words, the innovative activity of a firm is likely to be reflected in its level of productivity rather than in the short-term rate of growth of this variable, which is affected by the state of the business cycle or by the contingent behaviours of firms. 3.3 Dummy variables The last group of indicators in Table 2 includes a set of dummies. These were selected to capture sector-specific technological regimes as well as structural differences between sectors and firm-size classes in terms of funding and conducting innovation activities, and also in terms of economic performance. Great care was taken in the empirical identification of the sectoral dummies which were identified on the basis of earlier work that used the full set of data provided by CIS to explore the different dimensions of innovation in services (Evangelista, 2000; Savona, 2002; Evangelista and Savona, 2003). 2 1 The economic performance indicators such as sales/employees and sales growth are expressed in terms of current prices; thus they may be subject to price change effects. In order to account for this, we should need appropriate sectoral deflators, which unfortunately were not available. However, the use of constant prices is not relevant here, because the time span considered in the analysis is quite short. 2 In some instances, the choice of sectoral dummies was dictated by the small number of cases observed in some industries. The choice of the size dummies was based on a purely numerical criterion, that is, we attempted to preserve homogeneity in the distribution of firms across the different size classes.

10 444 G. Cainelli, R. Evangelista and M. Savona The basic descriptive statistics of the indicators used in the econometric estimates are presented in Table The econometric analysis 4.1 Specification of the model In order to test mechanisms B and A empirically, already discussed in Section 2, we estimated the following two reduced-form equations, respectively Y i;t ¼ a 0 þ a 1 X i;tÿ1 þ a# 3 Z i þ e i;1 X i;t þ1 ¼ b 0 þ b 1 Y i;t þ b# 3 Z i þ e i;2 where Y i,t denotes the innovative performance of firm i at time t, and X i,tÿ1 and X i,tþ1 respectively, denote the economic performance of firm i at time tÿ1 and tþ1, and Z i is a vector of sector and size dummies. 1 e i,j is a normally distributed error term. Equation (1) aims to test whether growth and productivity differentials across firms in the period are associated with differentials in the propensity to innovate in the same period, and the amount of resources devoted to innovation in The hypothesis underlying equation (1) is in line with mechanisms B1 and B2 discussed in Section 2. More particularly, firms with higher levels of productivity or those experiencing faster (than average) growth rates are expected to be more profitable and to have greater financial resources. Both these factors would be expected to act as an incentive to innovate (mechanism B1). We also assume that high growth rates (in sales) and labour productivity levels hint at the presence of a demand-pull incentive to innovate (mechanism B2). We estimated seven different specifications of equation (1), each using a different innovation indicator from those listed in Table 2. These included: the probability that a firm will introduce an innovation; probability of it being a process innovation; probability of it being a service innovation; total innovation expenditure per employee; innovation expenditure (per employee) devoted to: (i) R&D, design activities, acquisition Table 3. Economic and innovation indicators descriptive statistics Variables N obs. Mean Std. dev. Min. Max. Innovation performance indicators TOTEXP ÿ RD-DES ÿ ICT 204 ÿ ÿ INV ÿ Economic performance indicators SALES ÿ SALES ÿ PROD PROD ð1þ ð2þ 1 Given the nature of the dataset, we are not able to take fixed effects into account in our investigation. Therefore, as already stated, we paid close attention to the empirical identification of sectoral and size dummies in order to reduce the degree of unobserved heterogeneity.

11 Innovation and economic performance in services 445 of know-how; (ii) development or acquisition of software; and (iii) acquisition of new capital equipment. The reasons for choosing these indicators were discussed in the previous section. Here, it suffices to restate that the use of the indicators listed above allows us to explore in depth the endogenous nature of technological change in services and, in particular, to identify which type of innovation (product or process) and type of innovation activity is spurred by firms economic performance and demand factors. The economic performance indicators used in equation (1) are rate of growth of sales in the periods (SALES9395) and (SALES9698), expressed in logarithms (SALES), and labour productivity in the periods (PROD9395) and (PROD9698). Equation (2) estimates the impact of firms innovation activities on their economic performance (mechanism A). The aim is to verify what really boosts the productivity and economic growth of service firms. In other words, to find out whether just being an innovator is what matters, or whether it is the type of innovation introduced and the specific knowledge input used that is important. As explanatory variables, we use in separate estimations all the innovation indicators listed in Table 2. Finally, we would expect there to be a virtuous circle between innovation, economic performance and enhanced competitiveness, which according to mechanism C discussed in Section 2 would boost innovation through a dynamic self-reinforcing mechanism. In order to test empirically for the presence of such a mechanism, we estimated the following equation X i;t þ1 ¼ g 0 þ g 1 ^Y i;t þ g 3 Z i þ e i;3 In equation (3), Ŷ i,t is the Y i,t variable estimated in equation (1) and can therefore be interpreted as firms innovation activity induced by their past economic performance. By this means, we intend to account for the cumulative effect of past economic performance through induced innovation on the economic performance in the subsequent period. In other words, in estimating equation (3), we aim to verify whether the evolutionary metaphor (mechanism C) is effective to depict models of competition and selection mechanisms in services. Also, in this case, the use of different innovation indicators will allow us to identify the technological factors sustaining the long-term performance of firms, and the kind of knowledge inputs that have lasting effects on the growth and productivity of service firms. It would be interesting from this point of view to compare the role of ICT vis à vis other types of knowledge inputs (R&D and Design) as alternative determinants of technological and economic asymmetries among service firms. 4.2 Some econometric issues Before describing the empirical findings of our analysis, it is worth discussing two statistical and econometric issues related to the characteristics of the database used in the econometric analysis. These issues are: (i) the lag structure between innovation and economic performance and the causality direction between these variables, and (ii) the potential bias related to the sample selection problem. First, the characteristics of the database used in the empirical analysis, and described in Section 3, are not those of a panel. While the indicators of economic performance refer to the whole time-span ( ), matching with the CIS II only allows us to dispose of innovation indicators for the year This constrains the possibility of using a proper lag structure between innovation and performance. However, we believe that, given the constraints related to the characteristics of the database, which do not allow us to test for Granger causal links, the data are adequate to conduct a sound test for the existence of ð3þ

12 446 G. Cainelli, R. Evangelista and M. Savona structural associations between innovation and past and future economic performance. In this sense, the estimates of our equations should be regarded as purely descriptive, and not as causality tests between the independent and the dependent variables. In other words, as we shall show, the empirical tests suggest that firms that performed better in the past tend to carry out more innovative activities (equation (1)) and that firms that were engaged in innovation activities in the past tend to perform better in the future (equation (2)). In order to perform a true causality test between innovation and performance, a panel dataset would be needed. The second econometric issue concerns the (potential) presence in our data of a sample selection bias. In order to overcome this potential bias, we estimated equations (1), (2) and (3) using the Heckman two-step procedure. The first step consists of estimating a Probit model of a dummy variable. In our case, the latter takes the value 1 if the service firm has introduced a technological innovation and 0 otherwise, and is explained by a set of variables available for all the firms in the sample (innovative and non-innovative). 1 The residuals of this regression were used to construct a selection bias control factor, which is equivalent to the Inverse Mill s Ratio (Greene, 2000). This factor accounts for the effects of all unmeasured characteristics which are related to the selection variable. The Inverse Mill s Ratio is then introduced as an extra explanatory variable in the second stage of the Heckman procedure. The second step of the procedure consists in estimating the maximum likelihood of equations (1), (2) and (3) using the selection bias control factor as an additional independent variable. In this way, we obtain efficient and consistent estimates of the unknown coefficients of the equations. 5. The empirical results In this section, we present the results of the empirical estimation of the model described in the previous section. 5.1 From economic performance to innovation (B mechanisms) Table 4 presents the results of a set of robust Logit regressions estimating the impact of economic performance on respectively the probability of introducing an innovation (INN) [1], the probability of introducing a process innovation (INPRO) [2] and the probability of introducing a service innovation (INSERV) [3]. Each specification in turn considers the effects on the binary dependent variable of the average growth rate of sales over (SALES9395) [a] and the average level of labour productivity for the same period (PROD9395) [b]. The Logit models also include the complete set of sectoral and size dummies. In Table 4 (and all subsequent tables) the statistical significance of the variables under investigation has been measured in terms of t-ratios, corrected for the potential presence of data heteroscedasticity. Table 4 shows that the best performing firms in terms of both sales growth and labour productivity levels in the period are more likely to introduce innovations in that same period (estimation 1). However, these will be process innovations (estimation 3). Past economic growth (SALES9395) seems to be a greater stimulus for innovation than productivity levels (PROD9395). The coefficients of the sectoral dummies reveal the presence of wide differences across industries in the average propensity for firms to innovate, which are associated with different levels of technological opportunity. As 1 The independent variables used in the first step are the following: a constant term, two size dummies, a geographical dummy (North-West) and a dummy for whether or not the firm belongs to a business group.

13 Innovation and economic performance in services 447 Table 4. The impact of economic performance on the propensity to innovate Explanatory var. Dependent variables [1] [2] [3] INN INPROC INSERV [a] [b] [a] [b] [a] [b] Estimation Method Logit Logit Logit Logit Logit Logit Constant 0.445** ÿ2.555** 0.436** ÿ [0.188] [0.606] [0.250] [0.801] [0.246] [0.809] SALES ** 1.695** [0.501] [0.605] [0.661] PROD ** ÿ0.075 [0.095] [0.119] [0.121] TRADE Ref. Ref. Ref. Ref. Ref. Ref. HOTELS ÿ0.638** ÿ0.386 ÿ0.214 ÿ1.232** ÿ1.360** [0.360] [0.393] [0.553] [0.601] [0.664] [0.690] TRANSP ÿ ** 0.489* ÿ0.401 ÿ0.537 [0.206] [0.249] [0.293] [0.353] [0.297] [0.363] WASTE ÿ1.028** ÿ [0.507] [0.532] [0.944] [0.938] [0.837] [0.873] COMP 2.149** 2.872** 0.949** 1.030** 1.207** 1.096** [0.487] [0.497] [0.410] [0.452] [0.366] [0.408] RDCONS 1.238** 1.772** ** 1.590** [0.402] [0.403] [0.481] [0.472] [0.515] [0.526] LEGMKT * 1.251* ÿ0.112 ÿ0.227 [0.518] [0.614] [0.675] [0.697] [0.620] [0.654] OTHBUS ÿ1.193** ÿ0.410 ÿ0.160 ÿ0.039 ÿ0.213 [0.248] [0.333] [0.401] [0.497] [0.392] [0.513] D20 99 ÿ1.175** ÿ1.236** ÿ1.051** ÿ0.946** ÿ0.559 ÿ0.514 [0.222] [0.223] [0.326] [0.315] [0.343] [0.345] D ÿ0.340* ÿ0.415** ÿ0.464* ÿ0.432* ÿ0.157 ÿ0.126 [0.182] [0.188] [0.259] [0.261] [0.254] [0.257] D250 Ref. Ref. Ref. Ref. Ref. Ref. N obs Pseudo R ** Significant at 5%; * significant at 10%; robust standard errors in brackets. Equation [1] estimates on total sample; equations [2] and [3] on the sub-sample of innovative firms. expected, the software industry (COMP) and the S&T-based business services (RDCONS) show positive and much higher coefficients compared with the more traditional service sectors (Hotels and restaurants, Transport, Other business), though such differences mainly refer to service innovations. Also large firms were found more likely to innovate than small firms, though this finding holds with reference to process innovations only. Table 5 reports the robust Heckit estimations for the impact of past economic performance on firms financial commitment to innovation, and particularly on the amount of resources devoted to R&D and other disembodied technological inputs (design and know-how (RD-DES), software development and acquisition (ICT) and investments

14 Table 5. The impact of economic performance on the innovation intensity Explanatory var. Dependent variables [1] [2] [3] [4] TOTEXP RD-DES ICT INV [a] [b] [a] [b] [a] [b] [a] [b] Estimation method Heckit Heckit Heckit Heckit Heckit Heckit Heckit Heckit Second stage eq. Constant ÿ2.216** 2.176** ÿ ÿ3.407** ÿ1.590** ÿ4.201** [2.389] [0.646] [0.557] [1.052] [0.317] [0.539] [0.616] [1.122] SALES [0.411] [0.423] [0.549] [0.458] PROD ** 0.445** 0.624** 0.430** [0.090] [0.132] [0.079] [0.122] TRADE Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. HOTELS ÿ1.033** ÿ0.057 ÿ1.378** ÿ0.732 ÿ0.811** ÿ [0.368] [0.361] [0.519] [0.583] [0.256] [0.279] [0.504] [0.513] TRANSP ÿ0.648** ÿ1.078** ÿ0.400 ÿ1.013** ÿ0.098 ÿ [0.244] [0.261] [0.396] [0.459] [0.226] [0.238] [0.322] [0.383] WASTE ÿ ÿ0.773** ** [0.546] [0.506] [0.292] [0.289] [0.596] [0.612] COMP 0.952** 1.892** 1.515** 2.240** ** ** [0.270] [0.274] [0.392] [0.461] [0.268] [0.297] [0.317] [0.347] R&DCONS 2.226** 3.046** 3.197** 4.111** ** 1.123** 1.394** [0.410] [0.477] [0.470] [0.510] [0.277] [0.275] [0.478] [0.527] LEGMKT ** ÿ ** ** [0.332] [0.287] [0.594] [0.671] [0.589] [0.434] [0.331] [0.372] 448 G. Cainelli, R. Evangelista and M. Savona

15 OTHBUS ÿ1.688** ÿ0.176 ÿ2.040** ÿ0.913* ÿ1.910** ÿ0.302 ÿ0.938** [0.271] [0.311] [0.499] [0.545] [0.332] [0.335] [0.344] [0.434] D ** 2.781** 2.815** 1.435** 1.295** 0.699** 0.686* [1.131] [0.290] [0.552] [0.539] [0.305] [0.272] [0.382] [0.415] D ** 0.790** 0.667** 0.510** [0.454] [0.206] [0.405] [0.402] [0.251] [0.228] [0.261] [0.248] D250 Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. N obs Censored obs Uncensored obs Likelihood ÿ ÿ ÿ ÿ ÿ692.8 ÿ ÿ ÿ ** Significant at 5%; * significant at 10%; robust standard errors in brackets. Innovation and economic performance in services 449

16 450 G. Cainelli, R. Evangelista and M. Savona in technologically new capital equipment (INV)). The analysis of these coefficients shows that labour productivity in the period is positively related to all types of innovation expenditure made in However, a comparison among the different elasticity coefficients in Table 5 reveals that highly productive firms are more likely to re-invest their revenues in internal development or acquisition of software. The growth rate of sales in does not seem to have a statistically significant positive impact on innovation. The coefficients associated with the variable PROD9395 (specification b) provide the value of coefficient a 1 in equation (1). This coefficient shows values ranging from for capital equipment expenditure, to for ICT expenditure per employee. The coefficient a 1 can then be compared with the value of the elasticity g 1 (equation 3), to test for the presence of a cumulative effect (mechanism C), which will be discussed in Section 5.3. To sum up, past economic performance does affect both the propensity for service firms to innovate and the amount of resources devoted to innovation activities. Somewhat surprising is the result that past economic performance has an impact on process innovation rather than on the introduction of new services. This might be a peculiarity of services. It has already been pointed out that, in some of service sectors, process innovation takes the form of heavy investment in costly technological infrastructures (both tangible and intangible), while service innovations might consist of quality improvements carried out on a more continuous basis. Our results provide indirect support for the hypothesis that high growth rates, large profits and substantial cash flows might be a precondition for process innovation activity in services. Further, our estimates provide support for the hypothesis that past economic performance strengthens firms commitment to make investments in ICT both hardware and software. The endogenous nature of innovation seems therefore to have a process-oriented connotation, and this is likely to be a peculiar feature of services. However, it should be recalled that, in the case of services, process innovation strategies do not necessarily follow a cost-cutting objective. Both process innovations and ICTs could be introduced to enhance the quality and performances of the services delivered. Indeed, sales growth rates and levels of labour productivity measured as a ratio between sales and number of employees might be considered as a proxy for final demand, with high rates of sales growth and labour productivity levels being a symptom of favourable and sustained demand conditions. Although, as pointed out above, our analysis is not able to prove a Granger causality between economic performance and innovation activity, we can nevertheless argue that the presence of a positive structural association between the two variables does support the idea of a Schmooklerian type of mechanism in operation in service firms, which implies that favourable and sustained conditions of demand are a positive incentive to innovate and increase the amount of innovation expenditure. These findings for services are in line with most of the empirical evidence in the post- Schmooklerian tradition, discussed in Section 2, but hitherto exclusively confined to manufacturing activities (see Kleinknecht and Verspagen, 1990; Geroski and Walters, 1995; Brower and Kleinknecht, 1999, among others). Further, the stronger link found between past economic performance and the level of innovation expenditure devoted to ICTs compared with other types of innovation expenditure, is in line with most of the empirical literature on innovation in services. According to this body of work, service innovation is mainly incremental in nature and more likely to be related to specific applications of ICT as a general purpose technology (Helpman, 1998; Freeman and Soete, 1997) and arguably highly dependent on the positive response of destination markets as well as favourable demand conditions.