Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs

G Model

ARTICLE IN PRESS

RESPOL-3393; No. of Pages 17

Research Policy xxx (2017) xxx–xxx

Contents lists available at ScienceDirect

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

Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs夽 Marco Di Cintio a , Sucharita Ghosh b,∗ , Emanuele Grassi a a b

Department of Economics, Management, Mathematics and Statistics, University of Salento, Lecce, Italy Department of Economics, The University of Akron, Akron, USA

a r t i c l e

i n f o

Article history: Received 11 November 2015 Received in revised form 17 February 2017 Accepted 18 February 2017 Available online xxx JEL classification: J63 M51 O31 F14

a b s t r a c t We study firms’ export and R&D activities and their effects on employment growth and worker flows for a sample of Italian SMEs operating in the manufacturing industry. After accounting for the under-reporting of R&D in SMEs, our quantile regressions reveal that: (i) R&D is associated with higher employment growth rates, higher hiring rates and lower separation rates; (ii) R&D-induced exports are negatively related to firm employment growth and hirings and positively related to separations; and (iii) pure exports are not a driver of employment growth and worker flows. © 2017 Elsevier B.V. All rights reserved.

Keywords: Exports R&D Firm growth Worker flows

1. Introduction Previous research has investigated how innovative activities as reflected, e.g., in R&D investments, and export behavior affects firm employment growth but has often ignored the theoretical nexus between exports and innovation. While such linkages have been recently considered when explaining employment changes at the aggregate level (Autor et al., 2015), prior research has not investigated how the innovation and export behavior of firms, when simultaneously considered, impacts employment decisions at a micro level. Thus, we complement the macro evidence by investigating the combined effects of export activities and innovation, as seen in R&D investments, on firm employment growth. Since empirical research has established several regularities about the

夽 This work was supported by Project “5x1000 per la Ricerca – 2013” (Grant number F88C13000340001). We are grateful to the Italian citizens who allocated a share of their tax payments in support of the University of Salento. We would also like to acknowledge the insightful and valuable comments provided by the editors and two anonymous referees. Of course, all remaining errors remain our own. ∗ Corresponding author. E-mail addresses: [email protected] (M. Di Cintio), [email protected] (S. Ghosh), [email protected] (E. Grassi).

interconnectedness of innovation and export performance at the firm level, ignoring such linkages may provide an incomplete picture of the effects of these variables on firm employment growth. In particular, when R&D investments stimulate exports, firm employment growth may be affected by the direct effect of R&D, an indirect effect of R&D through exports, as well as a pure export effect. In order to disentangle such channels of growth, we investigate the employment consequences of R&D and exports simultaneously after accounting for the interdependence between exports and R&D. In doing so, we rely on two existing branches of the literature; the first branch deals with the employment impact of innovation and exports separately, while the second branch focuses on the interrelationships between innovations and exports at the firm level. While the theoretical intuitions underpinning our analysis are drawn from both these branches of literature and are valid for any firm irrespective of their size, we concentrate the empirical analysis on Italian small and medium-size enterprises (SMEs). We focus on SMEs due to several reasons. First, the economic literature largely acknowledges the relevance of SMEs in the creation of new jobs (Hölzl, 2009), their ability to innovate (Romero and MartínezRomán, 2012; Forsman, 2011) and to export (Sterlacchini, 1999). Second, given the large share of SMEs active in the EU market they

http://dx.doi.org/10.1016/j.respol.2017.02.006 0048-7333/© 2017 Elsevier B.V. All rights reserved.

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model RESPOL-3393; No. of Pages 17 2

ARTICLE IN PRESS M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx

are crucial for growth and jobs (Schmiemann, 2008). Consequently, EU institutions consider policy interventions specifically targeted at SMEs to be of primary importance. Third, the Italian manufacturing sector is characterized by the presence of a large number of SMEs (Parisi et al., 2006). Thus, the results of this study would be of interest to both policy makers and academic scholars of firms’ growth, innovation and exports. The large and growing body of literature that investigates the relationship between firm employment growth and innovation finds vast, yet inconclusive, empirical evidence on the role of innovation for firms’ growth in employment (Lachenmaier and Rottmann, 2011; Hall et al., 2008, 2009; Dachs and Peters, 2014). The literature also recognizes that international trade is a key determinant of employment growth. For example, exports of goods and services from the European Union supports around 25 million jobs in Europe, suggesting the importance of international trade for job creation and, thus, firm employment growth (Sousa et al., 2012). Classical trade theory based on the Ricardian principle of comparative advantage predicts that trade leads to workers moving between industries while the new heterogeneous trade theory (Melitz, 2003; Bernard et al., 2003) focuses on the reallocation of factors of production within industries from domestic firms to the more productive export-oriented firms. Whether we use classical trade theory or heterogeneous trade theory, economists agree about the long-run growth gains from trade as resources are used more efficiently but also recognize that there are short-run adjustment costs as labor markets adjust to trade. Along with the two approaches that separately deal with firm employment growth, i.e. the innovation-growth relationship and trade-growth relationship, there is also a large body of research that examines the relationship between innovation and firms’ exports. Most of the debate in this branch of the literature focuses on the direction of causality between innovation and exports but the conclusions have been indeterminate. Since innovation translates into competitive advantages that allow firms to compete in international markets, the literature clearly suggests that firms have to achieve high productivity levels before moving to the international market. Nevertheless, causality may run in the opposite direction. Indeed, international competition in export markets might raise productivity after entry, because exporters learn from foreign competition in export markets and, as a result, improve their domestic innovation activity. In this respect, empirical evidence on reverse causality, the so-called learning-by-exporting hypothesis, is rather scarce (Greenaway and Kneller, 2007; Damijan et al., 2010; Wagner, 2007), while the evidence in favor of the self-selection hypothesis (innovation increases firms’ productivity and, thus, exports) is abundant (Wakelin, 1998; Lachenmaier and Wößmann, 2006; Becker and Egger, 2013). The latter evidence confirms the predictions of open economy growth models where innovation is a key driver of exports since firms that innovate are more likely to export as they can charge lower prices and thus obtain higher returns from foreign sales than non-innovating firms (Grossman and Helpman, 1990, 1991).1 Our study extends previous research along two dimensions. First, we contribute to a challenge that has recently emerged in the literature where, after accounting for the interdependence between exports and innovations (as measured by R&D), the employment impact of simultaneous exposure to international trade and technology advances is studied (see Autor et al., 2015). Thus, our study integrates both export activities and R&D investments into

1 Higher productivity due to innovation activities often translates into lower production costs and, in competitive markets, may determine a reduction in prices. This, in turn, translates into domestic products being more competitive in foreign markets.

an empirical model of firm growth. Our second contribution is related to the attention given to firms’ hiring and separation occurrences in the specific context of innovative-exporting firms.2 This, to the best of our knowledge, has not been studied in the earlier literature. From this standpoint, while scholars have so far thoroughly devoted their attention to various measures of firm growth, less research has been conducted on the components of the employment growth rates such as hiring and separation decisions in relation to their export and R&D activities. Theories of firm-level employment dynamics in the spirit of Jovanovic (1982) are well established and have been fruitfully used to relate heterogeneities in gross job creation/destruction to firms’ age, size and operating industry (Davis and Haltiwanger, 1992; Burgess and Lane, 2001). However, an important part of understanding employment growth at the firm level is learning about its layoff and recruiting behavior. Firms grow and contract by changing the number of hires, the number of separations, or both. These choices cannot be perceived as non-random choices and, at least in principle, can be affected by R&D and export strategies, both of which we jointly consider in our study. Thus, using a rich micro-level firm dataset, we account for the dependence of exports on R&D investments and investigate their simultaneous effects on firm employment growth. By merging three waves of the Mediocredito-Capitalia Survey of Italian Manufacturing Firms (SIMF) covering the period 1998–2006, we build a three stage empirical model based on Crépon et al. (1998) and Hall et al. (2009) where we: (i) account for the under-reporting of R&D in SMEs; (ii) estimate export intensity due to R&D; and (iii) estimate the impact of export and R&D activities on firm growth. Moreover, our empirical strategy also tests whether export and R&D activities can be associated with changes in hires, changes in separations or changes in both, thus shedding light on their impact on worker flows at the firm level. After accounting for potential endogeneity, our quantile regressions reveal that R&D is associated with higher firm employment growth rates, higher hiring rates and lower separation rates. We also find that R&D-induced exports are negatively related to firm employment growth and hirings and positively related to separations. However, pure exports are not a driver of firm growth and worker flows. Overall, the empirical evidence in our study favors policy interventions aimed at stimulating the growth of SMEs through R&D-oriented policies rather than export-oriented policies. In practice, benefits may be expected from policies that promote a productive business research environment where firms grow and enter foreign markets as successful innovators. Moreover, since employment growth is achieved both from a higher inflow of new workers and lower separations, benefits can be expected in terms of within-firm human capital growth, knowledge retention and job stability. The rest of the paper is organized as follows. Section 2 reviews the literature. Section 3 describes our research methodology. Data used in the study is described in Section 4 and Section 5 discusses the empirical results. Section 6 concludes.

2 Firm employment growth rate can be obtained from either the net worker flow or the net job flow. While worker flows are movements of workers into a firm (i.e. number of hires) and out of a firm (i.e. number of separations), job flows measure the creation and destruction of jobs at the firm level. It is also worth noting that while net worker flows coincide with net job flows and both measure firm employment growth, the magnitude of their components may differ. For instance, a firm may create three jobs and destruct one job, so that the net job flow amounts to two. At the same time, the firm could hire five workers and separate from three, so that the net worker flow is again equal to two.

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model RESPOL-3393; No. of Pages 17

ARTICLE IN PRESS M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx

2. The impact of innovation and exports on firm employment growth Our study relates to two branches of literature. The first branch investigates the impact of innovation on firm employment growth, while the second branch explores the interrelationship between exports and innovations. As explained in the previous section, our empirical analysis concentrates on SMEs. However, in this section we provide the theoretical and empirical background behind both branches of the literature using studies that are not all necessarily focused on SMEs since the conclusions apply to any firm regardless of their size. In addition, we discuss hiring and separation decisions in the specific context of innovative-exporting firms. 2.1. Innovation and firm employment growth Theorists from different traditions largely acknowledge innovation as a key driver of firm performance and growth in contemporary market economies.3 However, when growth is evaluated in terms of employment at the firm level, the theory suggests that both the kind and strength of innovation strategies are likely to produce different outcomes in firm size and labor flows, while the overall effect of innovation on firm employment is ambiguous (Van Reenen, 1997; Hall et al., 2008). Theoretical linkages between innovative processes, as reflected, e.g., in R&D investments, and firm growth suggest a positive relationship between R&D investments and firm growth which is justified based on the complementarity between labor skills and R&D expenditures in order to favor innovation success (Leiponen, 2005; Capasso et al., 2015). Through investments in R&D, firms aim to consolidate their technological leadership and increase their market shares (Klette and Griliches, 2000).4 Also, technological diversification induced by R&D entails firms to opt among different trajectories in product development, allowing them to operate in markets that grow faster and achieve better outcomes than noninnovators (Del Monte and Papagni, 2003). Potential repercussions on personnel strategies and firm employment growth can then be plausibly expected. In this respect, there is ample evidence suggesting that knowledge accumulation explains variations in firms’ outcomes (Leiponen, 2000) and that labor practices at the firm level are strongly inter-related with R&D efforts and innovation strategies (Michie and Sheehan, 1999; Di Cintio and Grassi, 2016). As R&D uses knowledge intensively and knowledge is largely embodied in workers, innovating firms should find it profitable to adapt the internal skill composition to changes in their product (process) mix. Some authors have indeed highlighted the link between hiring new workers as a tool to acquire new skills (Barney, 1991) and knowledge creation and innovation (Rao and Drazin, 2002). In addition, R&D enhances organizational learning capabilities (Dosi et al., 1995; Cohen and Levinthal, 1989) which, in turn, translates into cumulative effects on firm performance (Capasso et al., 2015). Interestingly, Leiponen (2005) examines the relationship between skills and innovation and argues that research skills do not simply improve the effectiveness of R&D activities, but have positive repercussions on the entire organization. Thus, the accumulation of knowledge within firms through investments in R&D may easily be reflected in differences in firms’ employment strategies.5

3 See the evolutionary approach developed by Nelson and Winter (1982) and the new-growth and neo-Schumpeterian approach following Aghion and Howitt (1994) and Aghion et al. (2005). 4 Although not all innovations are preceded by structured R&D activities, the development of new, or improved, products and processes are often associated to them (see Bogliacino et al., 2011). 5 From a different perspective, Bogliacino (2014) argues that net employment changes due to R&D investments depend on the balance between two effects. On

3

Studies based on output measures of innovation largely rely on the distinction between product and process innovation. Product innovation can increase employment as more labor is needed to produce new goods or improve the quality of existing goods. On the other hand, product innovation in the form of firms’ introduction of new and/or more differentiated products in an attempt to strengthen their market power and set higher prices, could lead to output and employment contractions (Lachenmaier and Rottmann, 2011). Thus, worker displacement may weaken the labor-friendly impact of innovation as mature products are replaced by newer ones (Vivarelli, 2015). Innovations may also break up obsolete worker-job matches but these can eventually be replaced by new and more productive employment opportunities (Van Reenen, 1997). This yields worker flow dynamics at the firm level that are not completely captured by net employment changes. Conversely, process innovation modifies the relative productivity of production factors and, to the extent that such innovation is labor-saving, it could reduce employment. However, several compensation mechanisms may counter-balance the adverse employment effect. As process innovation is associated with lower production costs, firms tend to increase production and their workforce via price reductions and increased demand. Yet, cost reductions may effectively command lower prices if markets are competitive. If this is not the case, higher employment opportunities can only be predicted if extra profits are re-invested in the firm in the form of new machineries (Vivarelli, 2015).6 From an empirical perspective, studies that have focused on the effects of input measures of innovations, which are typically R&D activities, on employment changes reach ambiguous conclusions.7 While Yasuda (2005) and, more recently, Bogliacino et al. (2012) and Falk (2012) find that R&D expenditures have a positive impact on growth, Brouwer et al. (1993) report a negative relationship between R&D expenditures and employment. However, after the authors refine their R&D measure as a percentage of R&D dedicated to product development, they find a positive impact on employment growth. On the other hand, Klette and Førre (1998) do not find any clear relationship between job creation and R&D intensity. These conflicting results may be related to different data sources and methodologies, and serves as a further motivation of our study to contribute to the discussion with new evidence on this topic. 2.2. Innovation and firm exports Melitz (2003) theoretically established that when a firm is exposed to trade, only the most productive firm will export and benefit from trade since they can bear the fixed costs of trade barriers. Exposure to trade will cause the least productive firms to exit the market and some productive firms will only serve the domestic market. The positive association between exporting and efficiency has also been explained as the self-selection of the more efficient firms into the export market (Clerides et al., 1998). Early studies by Wagner (1995) and Bernard and Wagner (1997) establish, using

one hand, decreasing returns of R&D investments on productivity tend to increase employment, while on the other hand, larger firms find it easier to exploit the benefits from research and, consequently, they are likely to experience higher productivity and lower employment changes (Henderson and Cockburn, 1996). 6 For extensive discussions on compensation mechanisms, see Vivarelli (1995), Bogliacino and Vivarelli (2012) and Petit (1995). 7 Studies based on output measures of innovation investigate the impact of two kinds of innovation, product innovation and process innovation, both of which can have an ambiguous impact on firm employment. While product innovation is often found to have a positive impact on growth (Lachenmaier and Rottmann, 2011; Hall et al., 2008; Dachs and Peters, 2014), process innovation has been associated not only to employment growth (Lachenmaier and Rottmann, 2011) but also to employment reductions (Dachs and Peters, 2014) and employment stability (Hall et al., 2008). A review of earlier empirical findings can be found in Pianta (2005).

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model RESPOL-3393; No. of Pages 17

ARTICLE IN PRESS M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx

4

German firm-level data, that firm growth and export performance are positively related and that exporting firms have productivity advantages of 15–20 percent compared to non-exporting firms. Since innovation is regarded as an explanatory factor of productivity premiums, a firm’s decision to invest in R&D and innovate may explain firm export behavior. Empirical studies examining the linkages between innovation and firms’ desire to export can be grouped into studies that look at innovation effort (R&D expenditures) and innovation product measures (i.e. product and process innovation). However, the results are ambiguous. Wakelin (1998) finds that non-innovating U.K. firms are more likely to export than innovating firms of the same size since innovative firms may have domestic market advantages that reduce their incentive to search out new foreign markets. Similarly, Aw et al. (2008) using firm-level data for Taiwan do not find a significant relation between firm-level R&D and the probability of firms to begin exporting. Damijan et al. (2010), using data on the Slovenian manufacturing, also find no evidence that product or process innovation drives export propensity at the firm level. However, Becker and Egger (2013), using German firm data, find that firms introducing product and process innovation increase their propensity to export by about 10 percentage points. Using Spanish manufacturing firms, Cassiman et al. (2010) also find that product innovation, but not process innovation, drives firm-level exports propensity while Caldera (2010) finds that product and process innovation impacts exports with the former having a larger impact. Finally, Dosi et al. (2015) conclude that product innovation is relatively more important than process innovation in determining the success of exporting firms. We know that R&D investments are one of the main explanatory reasons for increased productivity and that productivity impacts exports. Additionally, both R&D and exports may simultaneously affect employment growth at the firm level. Thus it is of primary importance to quantify both the amount of exports due to R&D efforts and the residual amount of exports that is not explained by R&D. 2.3. R&D, exports and firm employment growth As mentioned earlier, our contribution to the literature is twofold. First, we account for the interdependence between exports and R&D investments and the employment consequences of R&D and exports simultaneously.8 This has been previously done only at the aggregate level by Autor et al. (2015). They use 722 U.S. commuting zones to proxy for local labor markets to examine the impact of technology and international trade on aggregate employment level and find distinct effects on local labor markets. While advances in technology change the composition within occupational categories, international trade reduces overall manufacturing employment among abstract-task-intensive, manual-task-intensive and routine task-intensive occupations. Second, we study the impact of export and R&D activities on the components of firm employment growth, specifically, their hiring and separation rates. In the context of international trade Moser et al. (2010) investigated the net employment and gross job flows for a representative sample of German establishments from 1993 to 2005; however, they do not focus on firms’ hiring and separation occurrences in the specific context of innovative firms.9

8 For example, studies like Czarnitzki and Delanote (2012) and Hölzl (2009) examine only the impact of exports on firm employment growth. Czarnitzki and Delanote (2012) use a dichotomous indicator for export status in a study of young innovative companies in Flanders and find a negative and statistically significant impact of exports on firm growth. Hölzl (2009) uses the export to sales ratio and concludes that exports are important for high-growth firms. 9 The authors refer to the impact of greater international competitiveness measured by variations of the exchange rate. They find that a loss in international

Moreover, the importance of the analysis of the components of the growth rate also stems from the magnitude of the flows of workers in the specific case of SMEs. Indeed, Haltiwanger et al. (2013) document that small firms create a conspicuous number of new jobs and, thus, contribute to firm employment growth. In fact, the authors report that start-up firms account for almost 20 percent of gross job creation. Close to our study, Baumgarten (2015) relates hirings, separations, and churning rates to the establishments’ exporting and importing activities and documents only a weak link between them and measures of worker flows.10 We extend this analysis by exploring the additional variation in firms’ worker flows brought about by firms’ R&D activities. In particular, since firms make choices on how to grow and contract by changing the number of hires and the number of separations, or both, these choices can be affected by export and R&D strategies as well. Indeed, since in R&D companies knowledge is intensively used, an increase in the number of hires could reflect the need to enrich or replace the endowment of skills. Lower separations could depend upon the need to retain skills and knowledge belonging to existing employees. Exporting firms might also shape their personnel policies differently from non-exporting firms. Thus, observed hiring and separations can eventually be interpreted as the result of optimal personnel policies that take into account the commitment to R&D and internationalization pursued by firms.

3. Empirical model The empirical model is constructed in three stages. In the first stage, we take into account the issue of under-reporting of R&D in SMEs and estimate an R&D intensity equation. Predicted values are then used in the second stage to estimate export intensity due to R&D. The third stage is aimed at answering our research questions and, thus, includes the core regressions that show the impact of export activities and R&D activities on firm growth and worker flows. As suggested by Kleinknecht (1987) and confirmed by Santarelli and Sterlacchini (1990), official R&D measures for SMEs may severely underestimate their innovation activities. The presence of informal activities, the type of R&D being undertaken, or the absence of structured R&D departments are all likely to be factors influencing the declared R&D effort (Roper, 1999) and are likely to be more relevant when focusing on SMEs (Klette and Kortum, 2004). Thus, self-reported R&D expenditure often fails to adequately describe the innovative effort of SMEs. The estimates of the R&D intensity equation are thus a necessary step to obtain a better proxy of the innovative activities carried out by firms in our sample.11 To account for under-reporting of R&D in SMEs, we assume that  related to a set of independent variables, a latent variable R&Dis

competitiveness makes firms willing to adjust their workforce through lower job creation and that the low sensitivity of the job destruction rate is likely to depend on labor market rigidities. Also, firms that survive the loss in international competiveness respond by adjusting their employment mainly through a lower job creation rate rather than a higher job destruction rate. 10 Churning is defined as the amount of worker movements in excess of that required for a firm to achieve its desired employment change (Burgess and Lane, 2001). 11 A similar approach can be found in Crépon et al. Crépon et al. (1998) and Hall et al. (2009).

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model

ARTICLE IN PRESS

RESPOL-3393; No. of Pages 17

M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx

xi , and to the observed R&D intensity according to the following standard type I Tobit model:

⎧ ⎨ R&D  , R&D  >c , i i ⎩

 the regressors zi , which includes the direct effect of R&D (R&D),

where ␣ is a constant term, ˇ is a vector of coefficients to be estimated, vi is the error component and c is the threshold beyond which R&D is observed.12 It is worth noting that in this step, we also control for self-selection of firms into R&D through a Two-stage Least Square (2SLS) model and, similar to Hall et al. (2009) on the same data, we reject the hypothesis of self-selection. Consequently, we estimate the R&D equation by a Tobit regression without the inclusion of a correction term for selectivity. From this step, we take the predicted values as a firm level proxy of actual R&D intensity  (R&D). In the second stage we delve into the relationship between R&D intensity and export intensity.13 In particular, we run a type I

 Tobit model in which we regress the latent export intensity (Exp)  and a set of independent on the estimated R&D intensity (xi R&D) variables,14  +ε  i = ı + xi +  + R&D Exp i i Expi =

⎩

(1)

where  indicates the quantiles, n is the sample size, ˇ is the vector of coefficient to be estimated, ui is the error component and Quant (yi |zi )denotes the quantile of yi , conditional on the set of

 ≤c 0, R&D i

⎧ ⎨ Exp  i , Exp i > d

Koenker and Hallock (2001) and Buchinsky (1998) to estimate a model specified as: yi = zi’ ˇ + ui withQuant (yi |zi ) = zi’ ˇ (i = 1, ..., n),

 = ˛ + ˇx + v R&D i i i R&Di =

5

,

i ≤ d 0, Exp

where ı is a constant term,  and  are coefficients to be estimated, ε is the error component and d is the threshold beyond which export is observed. We use these estimates to obtain predicted values and residuals, which describe, respectively, the amount of export intensity due to the R&D effort and the residual amount of export intensity that is not explained by R&D. Predicted values and residuals are included in the last stage of the estimation procedure. The third stage of the empirical model is devoted to study the impact of export activities and R&D activities on firm growth and its components, specifically the hiring and separation rates. To capture the direct effect of R&D we include the estimated R&D values  while the indirect effect of R&D through from the first stage (R&D),

 The exports is captured by the estimated export intensity (Exp). residual obtained from the second stage captures the direct impact  res ). We will refer to this effect as the pure export of exports (Exp effect. In our study, we adopt quantile regressions to identify the impact of R&D and exports on a firm’s employment growth, hiring rate and separation rate. Quantile regressions have increasingly gained the attention of scholars in the literature based on the growth-innovation relationship, allowing numerous authors to find that, at a micro level, the effects of innovation vary substantially along the conditional distribution of the employment growth.15 In particular, we follow Koenker and Bassett Jr. (1978),

 the pure export effect indirect effect of R&D through exports (Exp),  res ) and other covariates.16 Specifically, the estimator for ˇ (Exp solves the problem: min ˇ

⎧ ⎪ ⎨  1 n

⎪ ⎩i:y ≥z’ ˇ i

|yi − zi’ ˇ | +

i 

 i:yi


⎫ ⎪ ⎬

1 −  |yi − zi’ ˇ |

⎪ ⎭

.

(2)

This methodology has several advantages over alternative strategies. First, it can be used to characterize the overall distribution of a dependent variable given a set of regressors. This allows us to quantify the effects of a variable more accurately than standard linear regression techniques based on conditional mean functions. Second, quantile regression techniques have been proved to be robust in the presence of heteroskedastic and non-normally distributed errors. Finally, the quantile regression objective function is a weighted sum of absolute deviations, so the estimated coefficients are less sensitive to outliers. 4. Data description 4.1. Survey data We use firm-level, cross-sectional data drawn from three waves of the “Survey of Italian Manufacturing Firms” (SIMF) conducted by Mediocredito-Capitalia in 2001, 2004 and 2007 which provided information on the three years prior to the interview.17 Firms are asked to complete a questionnaire eliciting information on the labor force, innovation activities, export involvement and financial characteristics. Each wave includes both a stratified sample of firms with more than 10 workers and up to 500 workers and a non-stratified sample of firms above this threshold.18 Nevertheless, since our focus is confined to SMEs, we use a threshold of 250 employees to select the estimation sample. We merge the data from the three waves and exclude observations with inconsistent or missing information. Moreover, even if each wave contains around 9000 records, exploiting the panel dimension of the data is arguable, since the sample overlapping across waves is extremely small.19 However, in a robustness check we explore the panel nature of the data and account for firm-level fixed effects.20 For our estimation procedure, which consists of the three stages described earlier, we include all observations with exploitable information (i.e. no missing data) to ensure the highest sample size and representation in each stage. Moreover, while R&D expenditure has been reported on a yearly basis, both export status and

16

The full set of covariates is described in Section 4.2. For instance, the wave released in 2001 delivers yearly information from 1998 to 2000. Nevertheless, this is not the case for some variables, such as exports, which refers to the last year of each survey. Thus, while R&D, firm employment and many other variables are available on a yearly basis, exports can only be observed in 2000, 2003 and 2006. 18 Stratification is based on industry, geographic area and firm size. 19 By merging the second and third waves, Piva et al. Piva and Vivarelli (2005) are able to build a panel of 575 manufacturing firms. 20 We acknowledge an anonymous referee for suggesting the inclusion of such analysis. 17

12 A detailed description of the independent variables used in the regression is provided in Section 4.2. 13 Also in this step we first check whether firms self-select into export activities. 14 Note that this estimation stage uses the same set of covariates used in the R&D regression. 15 See, among others, Goedhuys et al. Goedhuys and Sleuwaegen (2010) and Falk (2012).

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model RESPOL-3393; No. of Pages 17

ARTICLE IN PRESS M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx

6

export intensity have been reported only for the last year of each survey. This data limitation leads us to focus only on the last year of each survey when dealing with export intensity. Thus, when we deal with the problem of under-reporting of R&D, we have information on 13,114 observations. In the second stage, the sample reduces to 6649 observations. Finally, as clarified in the next section, the net employment change computed from flow variables does not always match the net employment change computed from stock variables. We decided to drop these observations and run the quantile regressions on a final sample of 5445 observations. There are several advantages in using this data. First, firm-level data provide a better representation of the process of job creation and destruction compared to aggregate data. In particular, we use annual hires and separations, as well as employment stocks, to recover measures of firm growth and labor flows. Second, since the survey includes R&D and non-R&D firms, we are able to control for self-selection into R&D. Third, the survey simultaneously delivers information on employment, R&D and exports for a sufficiently large number of private firms in the manufacturing sector, allowing us to fully conduct our econometric analysis. Despite these advantages, the SIMF poses some limitations as well. First, data on hires and separations are not as accurate as one would wish in order to relate them to specific positions or skill levels. The data contains the number of hires and separations for each year, but it is not possible to verify if these changes also reflect a change in the skill composition of each firm.21 Second, the data does not describe the process of entry and exit of firms in the manufacturing sector. For instance, when firms are no longer included in the survey, it is not possible to discern whether those firms also did not survive in the market. Third, we cannot distinguish between voluntary quits and layoffs, rather we only observe separations between workers and firms. 4.2. Data variables To depict employment dynamics, we implement the empirical model on three key dependent variables, i.e., the employment growth rate, hiring rate and separation rate. Specifically, the growth rate is defined as the yearly net employment change over initial employment. There are some discrepancies in the way data have been reported. The net employment change computed from flow variables (i.e. the difference between the yearly number of hires and separations) does not always match the net employment change computed from stock variables (i.e. the difference between employment at the end of the year and employment at the beginning of the year). To ensure consistency of firm growth rates, we compute and compare the net employment change both ways and keep all the observations with coherence between stock and flow variables. Hiring and separation rates are defined, respectively, as the yearly number of hires and separations divided by total employment. Here we stress the fact that our focus on hiring and separation rates is relatively new in this literature. The explanatory variables that are of main interest are those related to exports and R&D. We compute R&D intensity as the ratio between R&D expenditure and turnover, while export intensity is a self-reported measure of total exports over turnover. Due to the high skewness of the R&D intensity, we opt for a hyperbolic sine transformation (IHS).22 The IHS is defined as the

21 To be precise, firms were also asked to report the composition of their workforce in terms of managers, blue collars and white collars, but the response rate to this questions is extremely low. 22 Looking at the statistics for skewness, we note that the degree of skewness is 63.31 for the R&D intensity and 32.96 for the IHS transformed variable. Thus, we prefer to rely on the latter to carry out the estimates.





12 

ln yi + yi2 + 1

and, therefore, it can be interpreted in the same

way as a standard logarithmic variable but, unlike a log variable, the IHS is also defined at zero. In what follows, we describe the additional variables used in the estimation procedure. We control for several aspects common in similar studies. Table A1 in the Appendix A provides descriptions and details of each variable used in the empirical model, as well as the questions used to construct the key variables of interest. Since larger firms are expected to better exploit economies of scope and scale, we control for firm size as measured by the level of employment (Wagner, 1995; Wakelin, 1998; Cohen and Levin, 1989). Manufacturing firms are usually labor-intensive, thus we include sales per employee (Flaig and Stadler, 1994). In this way, we control for differences amongst more or less labor-intensive firms. Firms’ R&D effort, export performance and labor flows may also depend on the degree of market power (Geroski, 1990; Greenaway and Kneller, 2008). Therefore, as a measure of competitiveness, we include the share of firm sales to industry sales based on the two-digit ATECO classification.23 Based on the same classification, demand effects have been proxied with the growth rate of industry sales (Becker and Egger, 2013). Credit rationing is also expected to influence firms’ behavior. Since in the survey, firms were asked to report if credit requests were denied, we include this variable to verify whether financial barriers play a role (Bottazzi et al., 2014). A specific impact on firm R&D and export activities may also be related to the presence of foreign firms as direct competitors. As competition is fiercer, foreign firms may reduce firms’ market shares and market power. Hence, the empirical model also includes a dummy indicating if a firm perceives foreign firms as direct competitors. In this way, we control for competition both in domestic and foreign markets (Becker and Egger, 2013). Moreover, we add thirty industry dummies to capture industry specific patterns of both R&D expenditure (Cohen et al., 1987; Geroski, 1990) and internationalization (Rodríguez and Rodríguez, 2005; Bleaney and Wakelin, 2002). Finally, to partially capture differences in firms’ labor flexibility among the determinants of hiring, separation and growth rates (Moser et al., 2010), we add a dummy variable indicating if firms have used temporary agency workers. Other controls common to all the estimation stages include firm age, investments in physical capital, a dummy equal to one if a firm belongs to a group, geographic location (North-East, North-West, Centre and South), wave and time dummies.24 4.3. Summary statistics Based on the sample of 5445 observations, Table 1 shows summary statistics of growth, hiring and separation rates by export and R&D status. According to these unconditional figures, firms that engage in R&D activities, but do not trade internationally, have the highest growth rates. It seems that innovating firms are able to grow faster if they choose to sell their goods in national markets. This could be in line with the idea of limited competitive pressures in national markets compared to the competitive pressure faced in international markets. R&D can be the source of market power, which becomes stronger when the size of the market is limited. The same tables reveal that the growth and labor flow rates of non-innovative firms do not differ substantially when comparing exporters and non-exporters. In contrast, the growth and labor flow rates of exporters tend to be higher for non-innovating

23

The ATECO classification is the Italian version of the European NACE codes. The inclusion of these further controls is in line with similar studies. See, among others, Sousa et al. (2012), Wakelin (1998), Del Monte and Papagni (2003), and Hall et al. (2009). 24

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model

ARTICLE IN PRESS

RESPOL-3393; No. of Pages 17

M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx Table 1 Growth, hiring and separation rates by R&D and export status.

Table 2 Firms characteristics by R&D and export status.

Growth rate (%)

Non R&D companies R&D companies Total

Non R&D companies R&D companies Total

Non R&D companies R&D companies Total

7

R&D companies

Non-exporters

Exporters

Total

Mean SD Observations Mean SD Observations Mean SD Observations

0.03502438 0.40155485 1550 0.05407336 0.22274558 438 0.03922129 0.37 1988

0.03856296 0.7237841 1809 0.01580339 0.11577115 1648 0.02771316 0.53 3457

0.0369301 0.5970261 3359 0.02383898 0.14570343 2086 0.03191484 0.47752771 5445

Mean SD Observations Mean SD Observations Mean SD Observations

Hiring rate (%) Non-exporters 0.13092891 0.4406412 1550 0.13448088 0.22687937 438 0.13171148 0.40334511 1988

Exporters 0.13404243 0.80641291 1809 0.09113526 0.14524607 1648 0.11358798 0.59221392 3457

Total 0.1326057 0.66309928 3359 0.10023659 0.16662857 2086 0.12020498 0.53113102 5445

Mean SD Observations Mean SD Observations Mean SD Observations

Separation rate (%) Exporters Non-exporters 0.09590452 0.09547947 0.20551711 0.32741529 1550 1809 0.08040752 0.07533187 0.08583332 0.10650756 438 1648 0.09249019 0.08587483 0.1859795 0.24817178 1988 3457

Total 0.09567561 0.27785268 3359 0.07639761 0.10251463 2086 0.08829013 0.22744901 5445

Observations

size labor intensity competitiveness demand agency workers physical capital credit rationing age NW NE C foreign competitors group

Non-exporters 438

Exporters 1648

mean

SD

mean

SD

4.115321 12.7978 0.0029456 0.0498327 0.3789954 0.0420152 0.0570776 3.684901 0.3150685 0.2922374 0.2328767 0.1027397 0.2146119

0.6941607 0.7674336 0.0211195 0.0839964 0.4856917 0.0614018 0.2322562 0.7130485 0.4650744 0.4553109 0.4231477 0.3039658 0.4110223

4.637698 12.96202 0.0032668 0.0330851 0.5709951 0.0351519 0.0485437 3.815187 0.4144417 0.3276699 0.1674757 0.4035194 0.2918689

0.8032143 0.6890234 0.0227367 0.079003 0.4950843 0.0499475 0.2149773 0.6667967 0.4927749 0.4695062 0.3735134 0.4907521 0.4547603

Non-R&D companies

Observations

size labor intensity competitiveness demand agency workers physical capital credit rationing age NW NE C foreign competitors group

Non-exporters 1550

Exporters 1809

Mean

SD

Mean

SD

3.958251 12.71623 0.0039896 0.0578667 0.2980645 0.0398965 0.0587097 3.646221 0.336129 0.26 0.1954839 0.0793548 0.1612903

0.648453 0.7650096 0.0420894 0.0826091 0.4575556 0.0754583 0.2351564 0.680888 0.4725361 0.4387758 0.3967007 0.270379 0.3679172

4.304547 12.94872 0.0025515 0.0402464 0.4538419 0.029421 0.0525152 3.754119 0.3598673 0.3001658 0.2001106 0.2974019 0.973466

0.7639536 0.6942176 0.0210598 0.0801887 0.4980025 0.0479073 0.2231252 0.6776967 0.480094 0.4584566 0.4001935 0.4572413 0.3981062

5. Empirical results firms. Another interesting fact is that the standard deviations of the growth, hiring and separation rates of exporting firms are much larger than those of non-exporters. This large variability, however, might be related to differences in other firm dimensions, such as industry or regional characteristics that will be accounted for in our multivariate analysis. Table 2 clearly shows that there are sharp differences in firms’ characteristics associated with R&D and export status that must be taken into account in a multivariate setting.25 Exporting firms are on average older than non-exporting firms. Exporting firms are also substantially larger than non-exporters, especially if they also invest in R&D. Overall, firms involved both in export and innovative activities are on average larger and older. It is also interesting to highlight that exporters usually report foreign companies as their main competitors. Finally, from the geographic indicators, we find that, amongst all SMEs that simultaneously report being active both as exporters and innovators, only 9 percent belong to the South of Italy. Lastly, complete summary statistics are reported in Table A2 in the appendix, while Table A3 reports the pairwise correlation matrix.26

25 Since in the implementation of the empirical model we rely on a log–log model, in what follows all continuous variables are in logs. 26 We also check the degree of multi-collinearity with the variance inflation factors (VIF). In our analysis, all of the VIFs are lower than 2.5 and the mean VIF is 1.34. Thus, since all of the VIFs are relatively low, we can be confident that multicollinearity is not an issue for our analysis.

5.1. Accounting for the under-reporting of R&D in SMEs To check whether firms self-select into R&D, we run a Heckman two-stage selection model by using a rich set of variables aimed at capturing observable differences in R&D intensity. The first step is a selection equation estimated via Probit. Then we compute the inverse Mills ratio, which is then included in the R&D equation. The results are reported in columns (a) and (b) of Table 3 and the estimated coefficient of the selectivity term is not significant at conventional levels, which is similar to what was found by Hall et al. (2008) for the same data. We therefore conclude that selfselection is not relevant in our data. Thus, we decided to focus on a type-I Tobit model.27 The results are in column (c) of Table 3. We find that the coefficients on size is statistically significant, but the effect is quite small in magnitude, which is in line with what was found in previous studies (see, for instance, Klette and Kortum, 2004). Being part of a business group and dealing with foreign competition are both positively associated with R&D. We also find positive effects of the geographical indicators (with the reference region being the South of Italy). The estimated coefficient on credit rationing is also significant, but unexpectedly positive, probably because of the self-reported nature of this variable. However,

27 Estimates are carried out on a sample of 13,114 observations, while standard errors are bootstrapped with 399 replications. For convenience, industry, year and wave dummies have not been included in the table, but they are highly significant, both individually and jointly, with 2 -values equal to 130.88, 33.07 and 22, respectively.

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model

ARTICLE IN PRESS

RESPOL-3393; No. of Pages 17

M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx

8 Table 3 Accounting for under-reporting R&D in SMEs.

size labor intensity competitiveness demand physical capital credit rationing age NW NE C foreign competition group mills

Table 4 Export-innovation relationship.

(a)

(b)

(c)

0.358*** (0.0187) 0.136*** (0.0188) −0.726 (1.112) −0.0393 (0.0491) 0.749*** (0.145) 0.188*** (0.0521) 0.0395** (0.0170) 0.213*** (0.0402) 0.293*** (0.0404) 0.283*** (0.0440) 0.392*** (0.0279) 0.0853*** (0.0308)

−0.00700 (0.00883) −0.0477*** (0.00401) 0.356*** (0.106) −0.0168*** (0.00570) 0.0369 (0.0279) −0.000940 (0.00753) −0.00572*** (0.00216) −0.00223 (0.00752) 0.00521 (0.00913) 0.000852 (0.00923) 0.00443 (0.0102) 0.0189*** (0.00369) −0.0104 (0.0382)

0.0202*** (0.00304) −0.0159*** (0.00612) 0.196 (0.168) −0.0117 (0.00869) 0.0758** (0.0317) 0.0138** (0.00690) 0.0000861 (0.00178) 0.0153*** (0.00484) 0.0247*** (0.00647) 0.0231*** (0.00577) 0.0306*** (0.00460) 0.0173*** (0.00508)

Notes: The dependent variable is the R&D intensity. Columns (a) and (b) report the results of the Heckman two-stage selection model. Results from column (c) refer to type-I Tobit model. ***, **, * denote, respectively, significance levels at 1%, 5% and 10%. Estimates are carried out on a sample of 13,114 observations.

capital market failures in financing fast-growing firms are not new for Italy (Bottazzi et al., 2014) and the positive correlation between firms’ R&D intensity and credit rationing may be interpreted as the failure to finance firms that invest in R&D and deserve credit. We use the predicted values from this first stage to proxy the R&D intensity of firms in our sample. This R&D intensity is used to study the export-innovation relationship as well as the impact of R&D and export activities on growth, hiring and separation rates.

5.2. Estimating the exports-R&D relationship We now turn to the estimation of the exports-R&D relationship. As before, we first assess self-selection into exports with a Heckman two-stage procedure. A selection into exports equation (column a, Table 4) is used to recover the inverse Mills ratio. Then, we include this ratio in a second stage regression and control the significance of the selection term (column b, Table 4). Since the Mills ratio is not significant at conventional levels, we do not correct for selection in our estimates of export intensity. Estimates of the type-I Tobit model are reported in Table 4, column (c), and show that R&D intensity has a positive and significant impact on the export intensity. This result is not surprising, since most of the literature has already highlighted the importance of innovation as a driver for international trade. Successful exporters are often innovators because innovation helps firms facing the more intense competition in international markets. Table 4 reports marginal effects on the latent variable, while the marginal effects on the actual export intensity can be obtained by multiplying the estimated coefficients by the probability that an observation is different from zero. In our case, the marginal effect of R&D on exports turns out to be 4.115. To evaluate the magnitude of this effect, we start from a reasonable change in our regressor, say 5%. By multiplying this percentage change by 4.115, we obtain 20.58%, which is the percentage increase in our dependent

R&D intensity size labor intensity competitiveness demand physical capital credit rationing age NW NE C foreign competition group mills

(a)

(b)

(c)

11.86*** (4.481) 0.359*** (0.0411) 0.342*** (0.0360) −11.40*** (2.860) 0.153 (0.306) −0.577* (0.331) 0.0507 (0.0828) 0.0553** (0.0268) 0.326*** (0.0594) 0.290*** (0.0653) 0.190*** (0.0686) 0.854*** (0.0644) −0.213*** (0.0543)

4.096*** (0.844) 0.0455*** (0.0104) 0.0695*** (0.00886) −2.357*** (0.611) 0.0986** (0.0462) −0.121** (0.0521) 0.00706 (0.0136) 0.00204 (0.00443) 0.0487*** (0.0115) 0.0294** (0.0132) 0.0389*** (0.0122) 0.135*** (0.0169) −0.0375*** (0.00985) 0.0153 (0.0327)

4.115*** (1.223) 0.0849*** (0.0114) 0.100*** (0.0103) −3.753*** (1.050) 0.116 (0.0749) −0.199** (0.0897) 0.0172 (0.0201) 0.00536 (0.00724) 0.0909*** (0.0168) 0.0704*** (0.0174) 0.0740*** (0.0193) 0.204*** (0.0152) −0.0515*** (0.0135)

Notes: The dependent variable is the export intensity. Columns (a) and (b) report the results of the Heckman two-stage selection model. Results from column (c) refer to type-I Tobit model. ***, **, * denote, respectively, significance levels at 1%, 5% and 10%. Estimates are carried out on a sample of 6649 observations.

variable.28 In our estimation sample, the average export intensity is 23%, thus we can conclude that a 5% increase in R&D intensity for the average firm would lead to an export intensity of about 27.7% (=23% + (23% * 0.2058)). As far as the other regressors are concerned, we find that age, credit rationing and our proxy for internal demand are not good predictors of exports, while the dummy for foreign competitors is positively associated with export intensity. Larger firms show higher export performance, while being part of a group is negatively associated with export intensity. Also geographic dummies indicate large disparities among Italian regions. Most of the coefficients on industry, years and wave dummies are again highly significant. From the main Tobit estimation, we compute predicted values and residuals that are later used to understand the impact of exports on growth and labor flows. Predicted values tell us the share of export intensity which is explained by R&D intensity and other control variables. Thus we use them in the next estimation step to account for any indirect effect of R&D on labor flows and firm growth. Instead, we use residuals as the component of export intensity that is not explained by R&D (we call it the pure export effect). Although we control for many factors, interpreting the estimated coefficients in the second stage as causal effects should be done with cautiousness. In particular, the variability of the R&D intensity could partly be endogenous. Section 5.4 will deal with the problem of potential endogeneity both with the Smith and Blundell (1986) two-stage procedure and an instrumental variable approach.

28

Note that also the export equation has been estimated as a log–log model.

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model

ARTICLE IN PRESS

RESPOL-3393; No. of Pages 17

M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx

9

Table 5 Direct and Indirect R&D effect and Pure Export effect. p10 Panel (a): Growth rate 7.984*** R&D intensity (0.836) −0.773*** R&D indirect effect (0.147) pure export effect −0.00697 (0.0102) size −0.0845*** (0.00820) labor intensity 0.0972*** (0.0107) agency workers −2.886** (1.141) competitiveness 0.00925* (0.00502) demand 0.0774* (0.0436) physical capital −0.111*** (0.0395) credit rationing −0.0714*** (0.0140) age −0.00140 (0.00404) NW 0.0205* (0.0109) NE −0.00185 (0.0103) −0.00565 C (0.0110) 0.0461** foreign competitors (0.0222) −0.0733*** group (0.00955) Panel (b): Hiring rate R&D intensity R&D indirect effect pure export effect size labor intensity agency workers competitiveness demand physical capital credit rationing age NW NE C foreign competitors group

−1.18e-13 (0.180) −3.80e-14 (0.0297) −2.98e-16 (0.00248) 2.03e-15 (0.00193) 1.86e-15 (0.00219) −2.62e-13 (0.168) 1.90e-16 (0.00132) 9.08e-15 (0.0117) 1.55e-15 (0.0167) 1.20e-15 (0.00241) 4.21e-18 (0.000918) 2.55e-15 (0.00239) 2.76e-15 (0.00233) 2.66e-15 (0.00236) 7.04e-15 (0.00440) −9.31e-17 (0.00216)

Panel (c): Separation rate 2.63e-13 R&D intensity (0.148) −1.51e-13 R&D indirect effect (0.0259) −9.02e-16 pure export effect (0.00233)

p20

p30

p40

p50

p60

p70

p80

p90

4.822*** (0.461) −0.363*** (0.0885) −0.0109* (0.00603) −0.0545*** (0.00477) 0.0563*** (0.00727) −1.439* (0.820) 0.00691** (0.00277) 0.0467** (0.0227) −0.0542** (0.0236) −0.0460*** (0.00941) −0.00317* (0.00171) −0.00245 (0.00573) −0.0149*** (0.00511) −0.0136** (0.00537) 0.00827 (0.0138) −0.0384*** (0.00501)

2.984*** (0.469) −0.198*** (0.0495) −0.00154 (0.00313) −0.0346*** (0.00465) 0.0299*** (0.00464) −0.429** (0.168) 0.00391*** (0.00141) 0.0247** (0.0116) −0.0286** (0.0130) −0.0296*** (0.00672) −0.00160* (0.000834) −0.00436 (0.00289) −0.0103*** (0.00279) −0.00888*** (0.00285) 0.00402 (0.00540) −0.0205*** (0.00376)

−7.24e-15 (0.119) 2.54e-15 (0.0161) 7.56e-18 (0.00144) 2.79e-17 (0.00123) −1.98e-16 (0.00131) 1.78e-14 (0.210) 3.43e-18 (0.000683) −2.56e-16 (0.00546) 2.57e-16 (0.00535) −2.54e-17 (0.00176) −1.03e-17 (0.000481) −1.31e-16 (0.00131) −8.71e-17 (0.00127) −9.71e-17 (0.00132) −3.41e-16 (0.00227) 9.03e-17 (0.00116)

1.818*** (0.283) −0.0757*** (0.0199) −0.000422 (0.00158) −0.0190*** (0.00280) 0.0141*** (0.00208) −0.0421 (0.106) 0.00169** (0.000734) 0.0170** (0.00745) −0.0217** (0.00857) −0.00551*** (0.00199) −0.00109** (0.000516) −0.00516*** (0.00170) −0.00904*** (0.00196) −0.00758*** (0.00190) −0.00260 (0.00267) −0.0106*** (0.00184)

4.792*** (0.549) −0.248*** (0.0349) −0.00361 (0.00322) −0.0589*** (0.00648) 0.0458*** (0.00444) −0.679*** (0.100) 0.00672*** (0.00158) 0.0375*** (0.0144) −0.0351 (0.0221) −0.0142*** (0.00373) −0.00517*** (0.00135) −0.0166*** (0.00358) −0.0261*** (0.00430) −0.0249*** (0.00413) −0.0000746 (0.00518) −0.0324*** (0.00362)

7.486*** (0.707) −0.348*** (0.0656) −0.00726 (0.00466) −0.0977*** (0.00711) 0.0707*** (0.00583) −1.165*** (0.150) 0.00759*** (0.00227) 0.0433** (0.0181) −0.0650* (0.0383) −0.0181*** (0.00585) −0.00814*** (0.00188) −0.0290*** (0.00556) −0.0439*** (0.00610) −0.0417*** (0.00565) −0.00793 (0.00964) −0.0506*** (0.00520)

12.00*** (1.305) −0.472*** (0.0890) −0.00411 (0.00699) −0.157*** (0.0151) 0.106*** (0.0108) −1.439 (2.371) 0.0101*** (0.00345) 0.0579** (0.0267) −0.147*** (0.0537) −0.0284*** (0.00986) −0.0122*** (0.00298) −0.0477*** (0.00749) −0.0760*** (0.00962) −0.0702*** (0.00971) −0.0298** (0.0134) −0.0769*** (0.00863)

25.36*** (3.029) −0.820*** (0.144) −0.00152 (0.00971) −0.313*** (0.0318) 0.201*** (0.0199) −3.059*** (0.455) 0.00956** (0.00451) 0.108*** (0.0414) −0.400*** (0.0995) −0.0579*** (0.0174) −0.0135*** (0.00397) −0.119*** (0.0170) −0.188*** (0.0245) −0.167*** (0.0234) −0.0937*** (0.0247) −0.153*** (0.0174)

0.907*** (0.258) −0.0306 (0.0381) −0.00406* (0.00245) −0.0121*** (0.00284) 0.0100*** (0.00339) −0.231 (0.379) 0.00926*** (0.00154) 0.00659 (0.0132) 0.0404*** (0.0128) −0.00450* (0.00259) −0.00133 (0.000909) 0.000976 (0.00261) 0.00137 (0.00248) −0.000479 (0.00245) −0.00246 (0.00519) −0.00373 (0.00250)

1.715*** (0.352) −0.0870** (0.0423) −0.00818** (0.00349) −0.0306*** (0.00423) 0.0225*** (0.00409) −0.423 (0.355) 0.0155*** (0.00190) 0.0333** (0.0144) 0.0403* (0.0215) −0.00295 (0.00393) −0.00314** (0.00133) 0.000203 (0.00326) 0.00395 (0.00360) −0.00345 (0.00346) −0.000717 (0.00596) −0.0110*** (0.00331)

2.302*** (0.369) −0.137*** (0.0442) −0.00529 (0.00411) −0.0435*** (0.00414) 0.0302*** (0.00387) −0.367*** (0.122) 0.0155*** (0.00207) 0.0295* (0.0171) 0.0426 (0.0280) −0.00458 (0.00487) −0.00591*** (0.00169) 0.000568 (0.00425) 0.00459 (0.00445) −0.00484 (0.00437) 0.00513 (0.00655) −0.0161*** (0.00365)

3.617*** (0.515) −0.198*** (0.0487) −0.00813* (0.00467) −0.0607*** (0.00566) 0.0404*** (0.00463) −0.597*** (0.140) 0.0155*** (0.00226) 0.0356** (0.0178) 0.0275 (0.0231) −0.00369 (0.00545) −0.00829*** (0.00175) −0.00550 (0.00470) −0.00451 (0.00535) −0.0120** (0.00508) 0.00567 (0.00726) −0.0231*** (0.00447)

4.408*** (0.598) −0.227*** (0.0593) −0.00990* (0.00563) −0.0736*** (0.00699) 0.0477*** (0.00572) −0.765*** (0.212) 0.0152*** (0.00277) 0.0509** (0.0230) 0.0181 (0.0290) −0.00773 (0.00547) −0.0103*** (0.00225) −0.00988 (0.00642) −0.0110 (0.00685) −0.0185*** (0.00665) 0.00456 (0.00886) −0.0244*** (0.00543)

5.496*** (0.675) −0.225*** (0.0744) −0.0123* (0.00689) −0.0957*** (0.00802) 0.0599*** (0.00690) −1.042*** (0.243) 0.0168*** (0.00314) 0.0551* (0.0307) 0.0407 (0.0458) −0.00429 (0.00801) −0.0140*** (0.00270) −0.0271*** (0.00758) −0.0261*** (0.00831) −0.0331*** (0.00791) −0.00298 (0.0112) −0.0324*** (0.00633)

9.982*** (1.500) −0.399*** (0.0917) −0.00822 (0.0112) −0.154*** (0.0172) 0.0961*** (0.0108) −1.037*** (0.250) 0.0215*** (0.00505) 0.0600 (0.0370) 0.00648 (0.0574) −0.00589 (0.0153) −0.0193*** (0.00435) −0.0671*** (0.0138) −0.0760*** (0.0158) −0.0800*** (0.0158) −0.0172 (0.0170) −0.0622*** (0.0111)

19.29*** (2.593) −0.860*** (0.203) 0.0205 (0.0164) −0.270*** (0.0278) 0.178*** (0.0199) −2.888*** (0.545) 0.0205*** (0.00724) 0.130 (0.0933) −0.202* (0.108) −0.00869 (0.0154) −0.0257*** (0.00572) −0.151*** (0.0231) −0.190*** (0.0259) −0.188*** (0.0240) −0.0289 (0.0287) −0.115*** (0.0167)

−1.174*** (0.256) 0.0648 (0.0440) 0.00251 (0.00259)

−2.474*** (0.278) 0.175*** (0.0399) 0.00223 (0.00323)

−2.638*** (0.286) 0.211*** (0.0462) 0.00359 (0.00362)

−2.655*** (0.340) 0.243*** (0.0496) 0.00573 (0.00429)

−3.037*** (0.415) 0.225*** (0.0643) 0.00621 (0.00490)

−3.416*** (0.524) 0.297*** (0.0902) 0.00545 (0.00623)

−3.290*** (0.565) 0.299*** (0.0959) 0.0134 (0.00885)

−4.891*** (0.963) 0.387** (0.164) 0.0253** (0.0129)

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model

ARTICLE IN PRESS

RESPOL-3393; No. of Pages 17

M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx

10 Table 5 (Continued)

size labor intensity agency workers competitiveness demand physical capital credit rationing age NW NE C foreign competitors group

p10

p20

p30

p40

p50

p60

p70

p80

p90

−1.28e-16 (0.00160) 1.22e-14 (0.00187) −1.40e-12 (0.0896) 2.23e-16 (0.00117) 2.40e-14 (0.00959) −1.04e-14 (0.0116) 2.11e-15 (0.00255) 6.37e-16 (0.000800) 7.45e-15 (0.00211) 5.39e-15 (0.00205) 5.68e-15 (0.00214) 2.15e-14 (0.00377) −4.01e-15 (0.00181)

0.0151*** (0.00227) −0.00948*** (0.00334) 0.133 (0.105) 0.00170 (0.00133) −0.00313 (0.0123) 0.0568*** (0.0200) 0.0103*** (0.00380) 0.00134 (0.000938) 0.0114*** (0.00296) 0.0150*** (0.00283) 0.0113*** (0.00281) −0.00167 (0.00597) 0.0102*** (0.00263)

0.0226*** (0.00298) −0.0248*** (0.00318) 0.492*** (0.0971) 0.00458*** (0.00163) −0.0110 (0.0142) 0.0930*** (0.0152) 0.0151*** (0.00399) −0.000336 (0.00115) 0.0152*** (0.00318) 0.0252*** (0.00321) 0.0180*** (0.00318) −0.00651 (0.00580) 0.0197*** (0.00288)

0.0191*** (0.00316) −0.0272*** (0.00367) 0.572*** (0.147) 0.00686*** (0.00169) −0.00177 (0.0158) 0.103*** (0.0123) 0.0184*** (0.00511) −0.000370 (0.00130) 0.0126*** (0.00421) 0.0268*** (0.00389) 0.0168*** (0.00429) −0.00885 (0.00663) 0.0236*** (0.00317)

0.0165*** (0.00408) −0.0284*** (0.00412) 0.657*** (0.154) 0.00671*** (0.00205) 0.0101 (0.0157) 0.107*** (0.0180) 0.0193*** (0.00562) −0.00163 (0.00162) 0.00994** (0.00407) 0.0247*** (0.00409) 0.0165*** (0.00445) −0.0117* (0.00700) 0.0264*** (0.00374)

0.0221*** (0.00456) −0.0306*** (0.00499) 0.581*** (0.164) 0.00668*** (0.00257) 0.000820 (0.0183) 0.102*** (0.0153) 0.0327*** (0.00743) −0.00311* (0.00184) 0.00769 (0.00496) 0.0261*** (0.00497) 0.0173*** (0.00494) −0.000689 (0.00932) 0.0295*** (0.00445)

0.0235*** (0.00618) −0.0359*** (0.00674) 0.750*** (0.217) 0.00613* (0.00318) −0.0102 (0.0270) 0.110*** (0.0187) 0.0437*** (0.0102) −0.00498** (0.00221) 0.000582 (0.00774) 0.0237*** (0.00801) 0.0111 (0.00818) −0.00969 (0.0137) 0.0340*** (0.00563)

0.0220*** (0.00660) −0.0374*** (0.00731) 1.218 (0.821) 0.00981** (0.00398) −0.00553 (0.0300) 0.0981*** (0.0212) 0.0555*** (0.0131) −0.00859*** (0.00318) −0.0114 (0.00879) 0.0132 (0.00837) 0.00593 (0.00859) −0.00456 (0.0144) 0.0365*** (0.00680)

0.0388*** (0.00953) −0.0548*** (0.0136) 1.934 (1.435) 0.00949 (0.00665) −0.0437 (0.0624) 0.110 (0.0755) 0.0559*** (0.0144) −0.0123** (0.00511) −0.0514** (0.0208) −0.0258 (0.0209) −0.0260 (0.0221) 0.00263 (0.0243) 0.0559*** (0.0137)

Notes: A quantile regression model is estimated in each panel. The dependent variable for panel (a), (b) and (c) is, respectively, the firms’ growth rate, hiring rate and separation rate. ***, **, * denote, respectively, significance levels at 1%, 5% and 10%. Estimates are carried out on a sample of 5445 observations.

5.3. Effect of R&D and exports on firm employment growth and worker flows Quantile regression results for the growth, hiring and separation rates are reported in Table 5. The main variables of interest are the R&D intensity, the R&D-induced export intensity and the export intensity not explained by R&D. In this way, we aim at capturing separately the direct impact of R&D, the indirect impact of R&D (through exports) and a pure export effect on growth and worker flows. Focusing first on the growth rate (panel a), we see that an increase in R&D intensity has a large positive impact on almost all the quantiles of the growth distribution. Moreover, the magnitude of the estimated coefficients is more pronounced when moving to the tails of the distribution. Since we have estimated a log–log model, the coefficients can be interpreted as elasticities. In particular, at the tails of the distribution, a one percent increase in R&D intensity is associated with a 7.98 and 25.36 percent increase at the 0.1 and 0.9 quantiles, respectively. The coefficient values become less pronounced when moving to the center of the distribution, thus R&D affects the shape of the conditional growth distribution especially along the tails. As far as the share of export intensity explained by R&D is concerned, i.e., the indirect effect of R&D, we find that it is negatively associated to growth, with greater effects when further away from the median. Nevertheless, the direct effect of R&D largely compensates the indirect effect, and thus the R&D intensity produces an overall improvement in the growth performance of SMEs. This is in line with most of the existing results on this topic, and thus, our findings support the idea that R&D improves the growth performance of SMEs. However, we further refine the existing results in the literature by showing that the shape of the conditional firm growth distribution is negatively affected by the indirect impact of R&D through exports, even if this impact is small in magnitude. In other words, changes of the conditional firm growth distribution driven by firms’ R&D activities are slightly more pronounced if firms are not active in international markets. Perhaps, a cautious interpretation would be that firms are able to extract a higher rent from

R&D investments when facing domestic rather than international competition. As Italian firms are often considered to be marginal innovators (Bratti and Felice, 2012; Santarelli and Sterlacchini, 1990), competition among Italian firms is largely driven by expanding the existing variety of products, especially in the manufacturing sector.29 Turning our attention to the coefficients related to the pure export effect, we see that they are also negative, small in magnitude and never significant at conventional levels. As pointed out in Section 2, previous research has only marginally tackled the exportinnovation relationship when estimating equations of firm growth. Our results suggest that there is no clear evidence in favor of a pure export effect on firm growth, while we find that R&D-driven exports negatively affects the growth distribution of firms. Overall, our results can be partly reconciled with the existing literature. First, our empirical evidence largely confirms the theoretical prediction that firms with higher R&D intensity (i.e. the most productive ones) tend to export more (Melitz, 2003). Thus, as in recent dynamic models where firms exhibit heterogeneous productivity (Melitz 2003; Grossman et al., 2006), non-R&D firms (i.e. the least productive ones) compete only in the domestic market while R&D firms engage in export activities. Second, the results on firm growth point to the importance of R&D for firm growth rather than their involvement in international trade. Thus, our study is coherent with previous research on the innovation-growth relationship (Aghion and Howitt, 1994; Aghion et al., 2005), but, at the same time, raises some doubts about the ability of exports to induce a growth in employment. In this respect, theoretical models of international trade assume that higher productivity is automatically translated into higher performance in terms of profitability and, thus, growth. Nevertheless, once firms compete in foreign markets, they tend to suffer from harsher competition. Indeed, Grazzi (2012) shows that there is no neat empirical evidence of a supe-

29 Santarelli and Sterlacchini (1990) distinguish between marginal innovation (sometimes also called soft or incremental innovation) versus radical innovation to highlight differences in the degree of product novelty.

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model RESPOL-3393; No. of Pages 17

ARTICLE IN PRESS M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx

rior performance in terms of profitability and growth for exporting firms. To explore the results more in-depth, we investigate whether the higher growth rates induced by R&D activities are compatible with the responses of hiring and separation rates to R&D. We find that the estimated coefficients in panels (b) and (c) are clearly coherent with those presented in panel (a). As R&D intensity rises, both the conditional distribution of hires and separations react in a way that is compatible with the positive effect of R&D on growth. In other words, R&D companies grow faster because of both increased hiring and decreased separations. SMEs that choose to invest in R&D tend to stabilize their workforce (lower separations) and are able to create opportunities for new jobs (higher hiring). Quantitatively, a one percent increase in R&D intensity is able to explain around a 3.62 percent increase in the hiring rate up to the median. The effect becomes even stronger at the right tail of the distribution, where estimated coefficients sharply increase up to 19.29 at the 0.9 quantile. At the same time, we observe that from the 0.6–0.9 quantiles, a unit percentage change in R&D intensity is followed by a 3–4.9 percent decrease in separations. Together, these findings suggest that innovation is a key element for reshaping firms’ hiring and firing strategies. At the firm level, workers inflows and outflows can be thought of as the result of optimal personnel policies which, in turn, can be affected by innovation strategies. Innovation strategies clearly modify the future skill composition of firms, which cannot be ignored when setting up optimal personnel policies. In particular, our results seem to support both the idea of human capital retention through lower separation rates and the need for new competences and skills in firms characterized by an innovative environment. We also find compatibility between the negative effect of the R&D-induced exports on growth and the signs of the effects on accessions and separations. Indeed, results show that the lower growth rates associated with the fraction of export intensity explained by R&D are the result of lower hiring rates and higher separation rates. Finally, in line with the results on the growth rate, we find that there is no clear evidence of a pure export effect on both hiring and separation rates. 5.4. Robustness To check the robustness of our results, we rerun the quantile regressions after excluding the pure export effect. Results are shown in Table 6 and confirm what was previously found. The exclusion of the export effect does not alter the point estimates and their statistical significance. Again, the results for the hiring and separation rates corroborate both the positive impact of R&D and the negative effect of the R&D-induced export on growth. This suggests that companies that actively engage in R&D activities outperform non-innovating companies in terms of employment growth, but this effect is slightly mitigated by the increased propensity to export once R&D is conducted. Up to now, we have ignored the fact that most Italian companies export inside the EU. Since reaching further destinations in terms of both physical and cultural distance is more difficult, we restrict the analysis to those companies that trade outside the EU. We then look at purely domestic companies that do not export at all and companies that trade outside the EU only. Results are reported in Table 7. Once again, the signs and the magnitude of the estimated coefficients point toward a positive impact of R&D intensity on employment growth and a negative impact of R&D-induced exports on growth. Also the components of the growth rate react in the expected way to increasing R&D and R&D-induced exports. Finally, exports per se do not seem to play a specific role for firms’ growth. In particular, we find that at the right tail of the growth and hiring distributions, the impact of R&D is even stronger.

11

We further check the robustness of our results along two lines of reasoning. First, we control if the results are confirmed when linking the R&D intensity to indicators of product and process innovations. Second, we exploit the panel nature of the data to make sure the results still hold when accounting for firm-level fixed effects. In the first robustness check, we relate our R&D variable to dichotomous output measures of innovation.30 This allows us to partially account for the success of R&D activities and, at the same time, examine if R&D is oriented towards product rather than process innovation. In particular, our data contains two dummy variables that reveal whether the firm produced at least one product/process innovation during the period covered by the survey. We use these dummies to select two types of innovative firms, those with positive R&D and that have only reported the introduction of product innovation, and those with positive R&D and that have only reported the introduction of process innovation. We rerun the estimates separately for the two types of firms by looking at R&D firms with product (process) innovation and companies that do not innovate at all. The results are shown in Tables 8 and 9, respectively, and corroborates what was previously found except for the indirect effect of R&D on the hiring rate, which loses significance. To explore the panel dimension of our data, we rely on the methodology suggested in Canay (2011). To account for (firm) fixed effects in quantile regressions, the author suggests a twostep estimator in which individual fixed effects are preliminarily estimated via a standard fixed-effect panel regression.31 Then, the estimated fixed effect is subtracted from the dependent variable in the quantile regression.32 As the estimator suggested by the author is asymptotically consistent for a large time span, and given the limited time dimension of our data, it is clear that the results from this check must be interpreted with care. Table 10 summarizes the estimated effects. Previous results are also broadly confirmed in this check, with the exception of the separation rates which loses significance. 5.5. Endogeneity As pointed out in Section 3, the coefficients estimated from the second stage of the empirical model could suffer from potential endogeneity bias. In particular, there is a problem of endogeneity of the R&D intensity due to the potential correlation between this measure and the error term in the export intensity equation. Indeed, exogeneity of R&D would require that firms’ R&D efforts took place independently from export decisions.33 Despite controlling for several factors, in the absence of an exogenous variation in the R&D behavior, our estimates should be interpreted with caution. To this end, we first test for potential endogeneity of the R&D intensity with the Smith and Blundell (1986) two-stage procedure. Then, even if this approach fails to reject exogeneity, we adopt an instrumental variable approach to check the robustness of the results. When conducting the Smith and Blundell test, we first regress the R&D intensity over the same regressors of the R&D equation in stage 1. The residuals from this stage are plugged into the Tobit estimation of the export intensity equation. Exogeneity is then evaluated by means of t-statistics on the coefficient of the first stage

30 We acknowledge an anonymous referee’s suggestion to explore issues related to the output of the innovation process. Unfortunately, our data does not report better measures of innovation output, such as the number of patents, the number of new products or processes and if they are radical rather than marginal innovations. 31 The estimates are carried out on an unbalanced panel of 246 firms. 32 For technical details, see Canay (2011). 33 A possible rational for this assumption could be found in the irreversibility nature of R&D investments (Pindyck, 1991; Abel et al., 1996).

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model

ARTICLE IN PRESS

RESPOL-3393; No. of Pages 17

M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx

12 Table 6 Direct and indirect R&D effect.

Panel (a): Growth rate R&D intensity R&D indirect effect Panel (b): Hiring rate R&D intensity R&D indirect effect

p10

p20

p30

p40

p50

p60

p70

p80

p90

7.920*** (0.743) −0.773*** (0.140)

4.795*** (0.459) −0.350*** (0.0965)

2.956*** (0.408) −0.194*** (0.0471)

−8.00e-15 (0.119) 2.79e-15 (0.0161)

1.807*** (0.283) −0.0744*** (0.0199)

4.769*** (0.552) −0.246*** (0.0347)

7.549*** (0.717) −0.352*** (0.0653)

11.86*** (1.426) −0.467*** (0.0867)

25.43*** (2.990) −0.814*** (0.138)

−3.33e-13 (0.180) 2.66e-14 (0.0297)

0.856*** (0.266) −0.0223 (0.0395)

1.723*** (0.355) −0.0772* (0.0422)

2.377*** (0.372) −0.132*** (0.0431)

3.519*** (0.510) −0.183*** (0.0476)

4.509*** (0.612) −0.205*** (0.0629)

5.576*** (0.688) −0.202** (0.0786)

9.866*** (1.559) −0.380*** (0.0942)

19.15*** (2.594) −0.891*** (0.196)

−1.212*** (0.248) 0.0690 (0.0439)

−2.437*** (0.278) 0.168*** (0.0393)

−2.672*** (0.282) 0.217*** (0.0456)

−2.676*** (0.345) 0.242*** (0.0488)

−2.989*** (0.387) 0.217*** (0.0607)

−3.411*** (0.546) 0.308*** (0.0899)

−3.382*** (0.575) 0.306*** (0.0994)

−4.899*** (1.088) 0.396** (0.167)

Panel (c): Separation rate 5.16e-13 R&D intensity (0.148) −2.23e-13 R&D indirect effect (0.0259)

Notes: A quantile regression model is estimated in each panel. The dependent variable for panel (a), (b) and (c) is, respectively, the firms’ growth rate, hiring rate and separation rate. ***, **, * denote, respectively, significance levels at 1%, 5% and 10%. Estimates are carried out on a sample of 5445 observations.

Table 7 Direct and indirect R&D effect and pure export effect (extra-EU exporters).

Panel (a): Growth rate R&D intensity R&D indirect effect pure export effect Panel (b): Hiring rate R&D intensity R&D indirect effect pure export effect

p10

p20

p30

p40

p50

p60

p70

p80

p90

7.501*** (2.620) −0.674** (0.332) −0.127 (0.204)

6.357*** (1.394) −0.455*** (0.122) −0.0870*** (0.0320)

4.737*** (1.210) −0.336*** (0.0947) −0.0405 (0.0300)

3.643*** (0.895) −0.218*** (0.0712) −0.0193 (0.0179)

5.184*** (1.199) −0.266*** (0.0834) −0.0236 (0.0192)

9.552*** (1.773) −0.437*** (0.0879) −0.0217 (0.0136)

12.74*** (1.479) −0.361*** (0.0833) −0.0382** (0.0181)

20.76*** (2.684) −0.433*** (0.151) −0.0465** (0.0211)

36.13*** (6.656) −0.777** (0.345) −0.0543* (0.0295)

0.743* (0.400) −0.0695 (0.0573) 0.00402 (0.00954)

1.849*** (0.481) −0.0911 (0.0809) 0.00708 (0.0108)

2.487*** (0.593) −0.110 (0.0787) 0.00542 (0.0113)

3.893*** (0.788) −0.175** (0.0761) 0.00128 (0.0101)

5.474*** (1.421) −0.236** (0.106) −0.00993 (0.0133)

7.148*** (1.671) −0.340*** (0.131) −0.0163 (0.0137)

9.985*** (1.717) −0.347* (0.194) −0.0383 (0.0242)

15.55*** (3.326) −0.482** (0.239) −0.0373 (0.0411)

29.29*** (4.724) −0.517 (0.453) 0.0546 (0.0673)

−1.947*** (0.593) 0.124 (0.115) 0.0179 (0.0165)

−3.308*** (0.522) 0.218*** (0.0783) 0.0272 (0.0174)

−4.156*** (0.540) 0.317*** (0.0812) 0.0290 (0.0181)

−4.377*** (0.681) 0.412*** (0.0951) 0.0249 (0.0337)

−4.526*** (0.876) 0.455*** (0.118) 0.0452** (0.0192)

−4.569*** (1.069) 0.445*** (0.144) 0.0694*** (0.0186)

−3.117*** (1.084) 0.335 (0.226) 0.0595 (0.0396)

−3.253** (1.376) 0.0897 (0.222) 0.0790 (0.0681)

Panel (c): Separation rate R&D intensity 1.23e-13 (0.431) R&D indirect effect −3.61e-14 (0.0714) pure export effect 4.24e-16 (0.0151)

Notes: A quantile regression model is estimated in each panel. The dependent variable for panel (a), (b) and (c) is, respectively, the firms’ growth rate, hiring rate and separation rate. ***, **, * denote, respectively, significance levels at 1%, 5% and 10%. Estimates are carried out on a sample of 2105 observations.

residuals. In particular, if we cannot reject the null hypothesis, the R&D intensity is an endogenous regressor and the standard errors for the R&D equation are not valid. Instead, if the t-test confirms exogeneity, then the second stage coefficients should not differ in magnitude and significance levels from those obtained in the Tobit regression that does not include the first stage residuals among its regressors. The procedure delivers a large p-value (0.141) for the residuals and the estimated coefficients are in line with those presented in Table 4. Thus, we conclude that exogeneity is not rejected.34 Despite this procedure being heavily used in applied work which makes us confident that our results do not suffer from endogeneity bias, we still assess the robustness of our results through the estimation of an IV Tobit model. The use of IV regressions is not new to the export-innovation literature (e.g. Lachenmaier and Wößmann, 2006; Czarnitzki and Wastyn, 2010) and is based on the need to find variation in innovation activities that is exogenous to export

34

performance. In our study, we use the yearly average amount of government financial incentives to firms’ R&D expenditure as the instrument. In the survey, firms were asked to report the percentage of R&D financed through direct subsidies and tax incentives, and the number of laws to which they applied in order to benefit from public financial incentives. We divide the amount of financial incentives by the number of laws to obtain the yearly average amount of public financial incentives received by firms. We test the relevance of the chosen instrument by looking at the F-value of the instrumental variable which, according to Staiger and Stock (1997), should exceed the value of 10. From the first stage regression, R2 and adjusted-R2 are around 0.72, so there is considerable precision in our IV estimation. Results indicate that the instrument has a positive and significant effect on R&D intensity and the F-statistic is 47.31, which is considerably larger than the rule of thumb of 10. Therefore, we conclude that the instrument is relevant, and the IV regressions will not suffer from a possible weak instrument bias. At the second stage, we find that the instrument has a positive and significant impact on export intensity. Moreover, the mini-

Complete tables are available upon request.

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model

ARTICLE IN PRESS

RESPOL-3393; No. of Pages 17

M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx

13

Table 8 Direct and indirect R&D effect and pure export effect (product innovation).

Panel (a): Growth rate R&D intensity R&D indirect effect pure export effect Panel (b): Hiring rate R&D intensity R&D indirect effect pure export effect

p10

p20

p30

p40

p50

p60

p70

p80

p90

11.12*** (1.469) −1.106*** (0.225) −0.00547 (0.0206)

6.202*** (0.874) −0.524*** (0.141) −0.00840 (0.00904)

3.468*** (0.945) −0.205** (0.0898) 0.00465 (0.00514)

0.230 (0.328) −0.00842 (0.0270) 0.000132 (0.00232)

1.433*** (0.457) −0.0300 (0.0300) 0.000943 (0.00251)

4.750*** (0.851) −0.109** (0.0539) 0.00174 (0.00408)

7.772*** (1.252) −0.165* (0.0921) 0.00466 (0.00869)

11.86*** (2.882) −0.218* (0.130) 0.00771 (0.0109)

26.05*** (3.528) −0.567*** (0.181) 0.000170 (0.0123)

1.11e-14 (0.418) 4.46e-15 (0.0603) −2.37e-17 (0.00554)

1.292*** (0.435) −0.0289 (0.0542) −0.00307 (0.00454)

2.082*** (0.546) −0.0857 (0.0603) −0.00577 (0.00555)

2.169*** (0.714) −0.0625 (0.0717) −0.00141 (0.00696)

3.391*** (0.942) −0.0996 (0.0845) −0.00484 (0.00749)

4.740*** (1.087) −0.118 (0.0987) −0.000774 (0.0115)

6.911*** (1.234) −0.228** (0.111) 0.00706 (0.0120)

7.708*** (1.828) −0.236 (0.170) 0.00149 (0.0158)

13.88*** (5.207) −0.474 (0.343) 0.0517 (0.0373)

−1.148*** (0.438) 0.0152 (0.0576) −0.000226 (0.00451)

−2.493*** (0.461) 0.150** (0.0628) 0.00152 (0.00574)

−3.268*** (0.548) 0.217*** (0.0706) 0.00180 (0.00663)

−4.165*** (0.638) 0.332*** (0.0972) 0.000844 (0.00730)

−4.463*** (0.747) 0.355*** (0.102) −0.000363 (0.00786)

−5.776*** (0.867) 0.473*** (0.139) 0.00895 (0.0113)

−6.907*** (1.172) 0.425*** (0.146) 0.0338*** (0.0129)

−9.144*** (2.330) 0.563* (0.321) 0.0482** (0.0204)

Panel (c): Separation rate R&D intensity 1.74e-13 (0.376) R&D indirect effect −2.76e-14 (0.0534) pure export effect −1.30e-16 (0.00495)

Notes: A quantile regression model is estimated in each panel. The dependent variable for panel (a), (b) and (c) is, respectively, the firms’ growth rate, hiring rate and separation rate. ***, **, * denote, respectively, significance levels at 1%, 5% and 10%. Estimates are carried out on a sample of 2323 observations.

Table 9 Direct and indirect R&D effect and pure export effect (process innovation). p10

p20

p30

p40

p50

p60

p70

p80

p90

10.37*** (2.353) −0.953*** (0.241) −0.0169 (0.0290)

5.976*** (1.210) −0.418*** (0.145) −0.000990 (0.00864)

3.664*** (1.088) −0.254** (0.100) 0.00443 (0.00430)

−8.30e-13 (0.253) 3.01e-12 (0.0285) 7.14e-15 (0.00248)

0.988** (0.461) −0.0307 (0.0297) 0.000273 (0.00237)

5.629*** (1.358) −0.147** (0.0632) −0.00203 (0.00509)

9.123*** (1.379) −0.192* (0.101) −0.00125 (0.00838)

14.40*** (2.207) −0.259*** (0.0746) 0.00368 (0.0136)

29.51*** (5.303) −0.487** (0.192) 0.0173 (0.0149)

3.25e-14 (0.433) −1.35e-13 (0.0587) −3.04e-16 (0.00543)

1.157** (0.503) −0.0611 (0.0557) −0.00396 (0.00438)

2.253*** (0.513) −0.0966* (0.0584) −0.00894* (0.00489)

3.118*** (0.828) −0.124* (0.0725) −0.0116* (0.00648)

3.786*** (0.686) −0.171** (0.0782) −0.00994 (0.00704)

5.667*** (1.222) −0.187* (0.102) −0.0135 (0.0102)

7.623*** (1.555) −0.237 (0.148) −0.00788 (0.0141)

10.43*** (2.110) −0.314* (0.181) −0.00122 (0.0151)

20.74*** (7.690) −0.465 (0.330) 0.0324 (0.0363)

Panel (c): Separation rate 8.83e-13 R&D intensity (0.425) −3.09e-12 R&D indirect effect (0.0564) −7.08e-15 pure export effect (0.00520)

−1.333*** (0.478) 0.0251 (0.0602) 0.00149 (0.00435)

−2.276*** (0.531) 0.0792 (0.0613) −0.00481 (0.00490)

−3.264*** (0.718) 0.196** (0.0883) −0.00681 (0.00647)

−3.590*** (0.852) 0.276*** (0.0991) −0.00620 (0.00785)

−3.653*** (0.862) 0.273*** (0.101) −0.0123 (0.00760)

−4.434*** (1.058) 0.358** (0.141) −0.0162* (0.00893)

−5.313*** (1.094) 0.361** (0.156) 0.00921 (0.0162)

−9.985*** (1.977) 0.797*** (0.308) 0.0253 (0.0187)

Panel (a): Growth rate R&D intensity R&D indirect effect pure export effect Panel (b): Hiring rate R&D intensity R&D indirect effect pure export effect

Notes: A quantile regression model is estimated in each panel. The dependent variable for panel (a), (b) and (c) is, respectively, the firms’ growth rate, hiring rate and separation rate. ***, **, * denote, respectively, significance levels at 1%, 5% and 10%. Estimates are carried out on a sample of 2230 observations.

mum chi-square statistic (Amemiya, 1978; Newey, 1987; Lee 1992) shows a p-value of 0.138, implying that our instrument is indeed an exogenous regressor.35 The IV Tobit coefficients are then used, as before, to obtain predicted values and residuals, which describe, respectively, the export intensity due to the R&D effort and the export intensity which is not explained by R&D. Then we study the impact of exports and R&D on firm growth, hiring and separation rates in order to confirm our previous results. Quantile regression results for the growth rate and its components are reported in Table 11. Once again an increase in R&D intensity has a significant and large positive impact on the growth distribution. We notice that the magnitude of the estimated coeffi-

35

Complete tables are available upon request.

cients is slightly more pronounced. Also, the negative effect of the R&D-induced exports on growth is confirmed and, with the exception of some cases, exports per se do not play a specific role for firms’ growth. As in previous estimates, these results are fully compatible with the effects found for the hiring and separation rates.

6. Conclusion Innovation, firm employment growth and exports have long been at the center of a challenging scientific debate. Our study takes an in-depth look at labor market dynamics at the firm-level and provides insight into how exposure to international trade and technology advances may shape employment growth patterns, as well as the related hiring and separation decisions. Specifically, the contribution of our paper is twofold. On one hand, we build and

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model

ARTICLE IN PRESS

RESPOL-3393; No. of Pages 17

M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx

14

Table 10 Direct and indirect R&D effect and pure export effect (panel and firm fixed effects).

Panel (a): Growth rate R&D intensity R&D indirect effect pure export effect Panel (b): Hiring rate R&D intensity R&D indirect effect pure export effect

p10

p20

p30

p40

p50

p60

p70

p80

p90

13.89*** (2.103) −1.526*** (0.350) −0.0116 (0.0162)

11.79*** (2.429) −1.227*** (0.432) −0.0103 (0.0107)

10.17*** (2.394) −1.057** (0.447) −0.00784 (0.0129)

7.734*** (1.828) −0.619** (0.278) −0.00590 (0.0100)

8.755*** (1.932) −0.714*** (0.273) −0.00358 (0.0116)

11.12*** (2.229) −0.881*** (0.317) −0.00163 (0.0130)

13.97*** (3.492) −0.993*** (0.333) −0.0111 (0.0198)

18.38*** (3.971) −1.105** (0.455) −0.0168 (0.0268)

34.98*** (7.676) −1.991*** (0.320) −0.0425** (0.0209)

3.159** (1.366) −0.240 (0.220) −0.0216* (0.0123)

5.359*** (1.139) −0.297* (0.172) −0.0143 (0.0128)

6.238*** (1.528) −0.434** (0.213) −0.00749 (0.0161)

8.179*** (1.447) −0.624*** (0.204) 0.0106 (0.0169)

10.14*** (1.996) −0.730*** (0.233) 0.00973 (0.0157)

11.94*** (2.110) −0.849*** (0.259) 0.00712 (0.0231)

14.87*** (3.158) −0.935** (0.371) 0.00757 (0.0460)

13.39*** (3.895) −0.694 (0.502) 0.0196 (0.0366)

19.78*** (4.659) −0.908** (0.412) 0.0449 (0.0276)

−1.578 (1.719) −0.0298 (0.291) −0.00368 (0.0121)

−3.054** (1.190) 0.180 (0.203) 0.0127 (0.0116)

−2.555* (1.360) 0.146 (0.235) 0.00835 (0.0133)

−1.793 (1.526) 0.150 (0.274) 0.0115 (0.0130)

−2.063 (2.921) 0.114 (0.432) 0.0137 (0.0194)

−3.225 (4.867) 0.298 (1.161) 0.0284 (0.0175)

−5.046 (5.194) 0.721 (0.685) 0.0315 (0.0252)

−6.221 (7.478) 1.251 (1.295) 0.0448 (0.0442)

Panel (c): Separation rate R&D intensity −0.399 (1.408) R&D indirect effect −0.0649 (0.262) pure export effect −0.00366 (0.0139)

Notes: A quantile regression model is estimated in each panel. The dependent variable for panel (a), (b) and (c) is, respectively, the firms’ growth rate, hiring rate and separation rate. Each model explores the panel dimension of the data by relying on the Canay (2011) methodology. ***, **, * denote, respectively, significance levels at 1%, 5% and 10%. Estimates are carried out on an unbalanced panel of 246 firms.

Table 11 Direct and indirect R&D effect and pure export effect (IV Tobit model).

Panel (a): Growth rate R&D intensity R&D indirect effect pure export effect Panel (b): Hiring rate R&D intensity R&D indirect effect pure export effect

p10

p20

p30

p40

p50

p60

p70

p80

p90

39.73*** (1.809) −3.038*** (0.156) −0.0151** (0.00642)

27.19*** (1.361) −1.980*** (0.0963) −0.00743* (0.00417)

21.91*** (1.539) −1.527*** (0.108) −0.00598 (0.00364)

18.85*** (1.954) −1.271*** (0.135) −0.00425 (0.00334)

18.81*** (1.559) −1.234*** (0.104) −0.00445 (0.00351)

21.77*** (1.679) −1.361*** (0.105) −0.00648* (0.00380)

24.97*** (1.614) −1.520*** (0.101) −0.00808 (0.00499)

34.54*** (2.144) −1.954*** (0.111) −0.00924 (0.00654)

46.17*** (3.887) −2.380*** (0.159) −0.00725 (0.00960)

1.701*** (0.289) −0.131*** (0.0237) −0.00125 (0.00182)

5.107*** (0.527) −0.375*** (0.0384) −0.00586** (0.00274)

7.248*** (0.605) −0.493*** (0.0434) −0.00533 (0.00352)

9.936*** (0.918) −0.674*** (0.0620) −0.00974** (0.00427)

12.44*** (0.937) −0.808*** (0.0631) −0.00915* (0.00472)

15.71*** (1.173) −1.035*** (0.0769) −0.00933** (0.00465)

20.71*** (1.601) −1.343*** (0.104) −0.00983 (0.00731)

27.59*** (1.913) −1.715*** (0.0937) −0.00997 (0.00999)

40.02*** (3.978) −2.336*** (0.217) 0.0102 (0.0159)

−4.979*** (0.541) 0.324*** (0.0397) 0.00352 (0.00269)

−6.471*** (0.573) 0.402*** (0.0451) 0.00377 (0.00312)

−7.033*** (0.660) 0.440*** (0.0503) 0.00500 (0.00359)

−7.485*** (0.767) 0.495*** (0.0552) 0.00479 (0.00406)

−8.837*** (0.990) 0.586*** (0.0775) 0.00578 (0.00492)

−9.792*** (1.308) 0.664*** (0.0993) 0.00869 (0.00586)

−10.84*** (1.796) 0.764*** (0.129) 0.00831 (0.00798)

−12.19*** (2.151) 0.844*** (0.161) 0.0221 (0.0139)

Panel (c): Separation rate −2.32e-12 R&D intensity (0.276) 3.93e-14 R&D indirect effect (0.0231) −4.35e-15 pure export effect (0.00232)

Notes: A quantile regression model is estimated in each panel. The dependent variable for panel (a), (b) and (c) is, respectively, the firms’ growth rate, hiring rate and separation rate. ***, **, * denote, respectively, significance levels at 1%, 5% and 10%. Estimates are carried out on a sample of 5445 observations.

estimate an empirical model that simultaneously disentangles the role of exports and R&D to firm employment growth. On the other hand, we shed light on firms’ hiring and separation occurrences in the specific context of innovative-exporting firms. It is important to understand how firms’ R&D and export activities affect their hiring of new workers, their separations from existing firms, or both, since it improves our understanding of the factors behind the hiring and separation processes. After all, firms grow and contract by changing the number of hires, the number of separations, or both and these choices can be affected by R&D and export strategies. We use establishment-level, cross-section data drawn from three waves of the “Survey of Italian Manufacturing Firms” for the period 1998–2006. After controlling for self-selection and endogeneity, our quantile regressions reveal that R&D is associated with

higher growth rates, higher hiring rates and lower separation rates; R&D-induced exports are negatively related to firm growth and accessions and positively related to separations; and pure exports are not a driver of growth and worker flows. The fact that pure exports are not associated with firm growth is not new in the literature (Grazzi, 2012). Although this represents a partial departure from standard models of trade à la Melitz (2003), it can be explained by the harsher competition faced by firms once they compete in the foreign market. The major findings of our study have interesting policy and managerial implications. Our results support the idea that, while there is no market failure argument to justify public interventions to firms that are already growing fast (Hölzl, 2009), R&D public policies could be effective to foster export activities. Public inter-

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model RESPOL-3393; No. of Pages 17

ARTICLE IN PRESS M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx

ventions can serve as a policy tool for stimulating R&D spending and, through their positive effect on exports, have both welfare and growth implications. The welfare implications stem from customer needs improving through international trade and the growth implications arise from the positive effects on firms’ employment (Hall et al., 2008). In this respect, however, our results indicate that pure exports do not affect growth and worker flows, while the negative effect of R&D-induced exports on firms’ growth is small in magnitude. Therefore, to stimulate firms’ employment growth, our results indicate that R&D-oriented policies should be preferred to export-oriented policies. Thus the main challenge for policy makers is to stimulate a creative environment for business development where firms grow and enter foreign markets as successful innovators. These policy implications turn out to be particularly relevant in the context of SMEs, where investment decisions, as pointed out in Esteve-Pérez and Rodríguez (2013), are often constrained by limited access to financial resources. From a managerial point of view, firms that engage in R&D activities pursue different optimal personnel policies compared to non R&D firms. In particular, the need to increase the endowment of skills pushes the hiring rate upwards as knowledge is largely embodied in workers. At the same time, knowledge retention appears to be essential for innovative companies, as the separation rate is lower compared to that of non-innovative companies. When combined, these effects allow innovative firms to protect their intellectual property more effectively and, at the same time, allow workers to enjoy greater job stability. As pointed out by Buch et al. (2009), companies active in international markets may be able to balance demand risks or, as noted by Baumgarten (2015), labor exiting exporting firms could be lower to the extent that wages are higher in these companies. Furthermore, SMEs planning to increase their presence in foreign markets should accept the challenge to increase their R&D effort by availing of national and international funding opportunities.

15

Table A2 Summary statistics.

growth rate (log) hiring rate (log) separation rate (log) R&D intensity (log) export intensity (log) size (log) labor intensity (log) competitiveness (log) demand (log) agency workers physical capital (log) credit rationing age (log) NW NE C foreign competitors group

Mean

SD

Min

Max

0.0214909 0.1038772 0.0831465 0.0098886 0.2330024 4.29158 12.87442 0.0032091 0.043866 0.4389348 0.0351506 0.0534435 3.736319 0.3660239 0.2964187 0.1915519 0.2517906 0.2170799

0.1502963 0.1834354 0.1326936 0.0631893 0.2716043 0.7859941 0.7274862 0.0290496 0.0814254 0.4963026 0.058829 0.2249369 0.6813798 0.4817603 0.4567198 0.3935584 0.4340815 0.4122953

−0.7766703 0 0 0 0 2.998223 7.838312 0.0000118 −0.2637302 0 0 0 0 0 0 0 0 0

3.93221 3.970648 2.949471 3.229893 0.8813736 6.214612 18.48152 0.8813736 0.5361266 1 1.375348 1 5.945427 1 1 1 1 1

Questions related to growth, hiring and separation rates: - Indicate the total number of employees on December 31, 2000. - Indicate the total number of employees for 1999 and 1998. - How many employees were hired by the company in 2000? In 1999? In 1998? - How many employees in 2000 have ceased employment with the company due to retirement, dismissal and other causes? In 1999? In 1998? Questions related to R&D intensity: - Indicate the total amount spent on research and development (internally and externally acquired) in 2000, 1999 and 1998. - Indicate the total turnover in 2000, 1999 and 1998.

Appendix A. List of questions used to construct the key variables used in the study These questions were asked in the 2001 wave of the Mediocredito-Capitalia Survey of Italian Manufacturing Firms. The same questions were asked in the 2004 and 2007 waves of the survey.

Question related to export intensity: - Indicate the amount of exports as a share of total turnover for the year 2000.

Table A1 Variables description. Variable

Description

growth rate hiring rate separation rate R&D intensity export intensity size labor intensity competitiveness demand agency workers physical capital credit rationing age NW NE C S (base category) foreign competitors group

yearly net employment change divided by total employment lagged one year yearly number of hires divided by total employment lagged one year yearly number of separations divided by total employment lagged one year ratio between R&D expenditure and total turnover total exports revenues over total turnover firm total employment ratio between total employment and total turnover firm’s market share: ratio between firm total turnover and industry sales (ATECO-2-digits) growth rate of industry sales (ATECO-2-digits) dummy equals to 1 if the firm reports to make use of temporary agency workers ratio between investments in physical capital and total turnover dummy equals to 1 if the firm reports to receive less credit than that requested number of years of firm’s business activity dummy equals to 1 if the firm is located in the North-West of Italy dummy equals to 1 if the firm is located in the North-Est of Italy dummy equals to 1 if the firm is located in the Centre of Italy dummy equals to 1 if the firm is located in the South of Italy dummy equals to 1 if the firm reports that its main competitors are from abroad dummy equals to 1 if the firm belongs to a corporate group

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model RESPOL-3393; No. of Pages 17

ARTICLE IN PRESS

1.0000 −0.4932* −0.3699* 0.0827* −0.0126 1.0000 0.1595* −0.0227* −0.0668* 0.0621* −0.0546* 1.0000 −0.0211 −0.0060 −0.0505* 0.0005 0.0070 −0.0063 1.0000 0.0067 0.0021 0.0123 −0.0100 −0.0253* −0.0138 −0.0039 1.0000 0.0017 −0.0226* 0.0498* 0.1046* 0.0993* −0.0675* 0.0914* 0.1411* 1.0000 −0.0721* 0.0601* −0.0091 −0.0584* −0.0499* −0.0060 0.0258* −0.1375* −0.0460* 1.0000 0.0068 0.0274* 0.0020 0.0114 −0.0115 −0.0134 0.0310* −0.0190 0.0150 0.0102 1.0000 0.0597* 0.0354* 0.0600* −0.1259* −0.0619* 0.0242* −0.0058 0.0266* 0.0048 0.0144 0.1257* 1.0000 −0.0143 0.0440* −0.0988* 0.2407* 0.0196 −0.0058 0.1539* 0.0050 0.0307* −0.0590* 0.1908* 0.3123* 1.0000 0.2865* 0.1075* −0.0135 −0.0931* 0.1416* −0.0591* −0.0162 0.0663* 0.0565* 0.0230* 0.0218 0.3129* 0.0991* 1.0000 0.0364* 0.0681* −0.1735* −0.0019 −0.0019 0.0382* 0.0231* −0.0091 −0.0122 −0.0085 0.0255* −0.0098 0.0145 0.0624* 1.0000 −0.0085 −0.0170 −0.0772* 0.0199 0.0100 0.0547* −0.0304* 0.0203 0.0691* −0.0254* −0.0588* −0.0108 0.0050 0.0188 −0.0085 1.0000 0.6253* −0.0116 −0.0336* −0.0289* 0.0526* 0.0037 0.0963* 0.0053 0.0609* 0.0332* −0.0962* −0.0713* 0.0108 −0.0134 −0.0166 −0.0173

18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1

1.0000 0.7099* −0.1021* −0.0069 −0.0241* 0.0330* 0.0471* −0.0047 0.0711* 0.0335* 0.0592* −0.0180 −0.0971* −0.0373* 0.0234* −0.0206 −0.0374* −0.0144 1 growth rate 2 hiring rate 3 separation rate 4 R&D intensity 5 export intensity 6 size 7 labor intensity 8 competitiveness 9 demand 10 agency workers 11 physical capital 12 credit rationing 13 age 14 NW 15 NE 16C 17 foreign competitors 18 group

Table A3 Pairwise correlation matrix.

1.0000 −0.3159* 1.0000 −0.0152 −0.0437* 1.0000 0.0113 −0.0129 0.1082* 1.0000

M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx

16

References Abel, A.B., Dixit, A.K., Eberly, J.C., Pindyck, R.S., 1996. Options, the value of capital, and investment. Q. J. Econ. 111 (3), 753–777. Aghion, P., Howitt, P., 1994. Growth and unemployment. Rev. Econ. Stud. 61 (3), 477–494. Aghion, P., Bloom, N., Blundell, R., Griffith, R., Howitt, P., 2005. Competition and innovation: an inverted-U relationship. Q. J.Econ. 120 (2), 701–728. Amemiya, T., 1978. The estimation of a simultaneous equation generalized probit model. Econometrica 46 (5), 1193–1205. Autor, D.H., Dorn, D., Hanson, G.H., 2015. Untangling trade and technology: evidence from local labour markets. Econ. J. 125 (584), 621–646. Aw, B.Y., Roberts, M.J., Yi Xu, D., 2008. R&D investments, exporting, and the evolution of firm productivity. Am. Econ. Rev. 98 (2), 451–456. Barney, J., 1991. Firm resources and sustained competitive advantage. J. Manage. 17 (1), 99–120. Baumgarten, D., 2015. International trade and worker flows: empirical evidence for Germany. Rev. World Econ. 151 (3), 1–20. Becker, S., Egger, P., 2013. Endogenous product versus process innovation and a firm’s propensity to export. Empir. Econ. 44 (1), 329–354. Bernard, A.B., Wagner, J., 1997. Exports and success in German manufacturing. Weltwirtschaftliches Arch. 133 (1), 134–157. Bernard, A.B., Eaton, J., Jensen, B., Kortem, S., 2003. Plants and productivity in international trade. Am. Econ. Rev. 93 (4), 1268–1290. Bleaney, M., Wakelin, K., 2002. Efficiency, innovation and exports. Oxford Bull. Econ. Stat. 64 (3), 3–15. Bogliacino, F., Vivarelli, M., 2012. The job creation effect of R&D expenditures. Aust. Econ. Pap. 51 (2), 96–113. Bogliacino, F., Piva, M., Vivarelli, M., 2011. R&D and Employment: Some Evidence from European Microdata. IZA DP 5908 http://ftp.iza.org/dp5908. pdf. Bogliacino, F., Piva, M., Vivarelli, M., 2012. R&D and employment: an application of the LSDVC estimator using European microdata. Econ. Lett. 116 (1), 56–59. Bogliacino, F., 2014. Innovation and employment: a firm level analysis with European R&D Scoreboard data. EconomiA 15 (2), 141–154. Bottazzi, G., Secchi, A., Tamagni, F., 2014. Financial constraints and firm dynamics. Small Bus. Econ. 42 (1), 99–116. Bratti, M., Felice, G., 2012. Are exporters more likely to introduce product innovations? World Econ. 35 (11), 1559–1598. Brouwer, E., Kleinknecht, A., Reijnen, J.O., 1993. Employment growth and innovation at the firm level. J. Evo. Econ. 3 (2), 153–159. Buch, C.M., Döpke, J., Strotmann, H., 2009. Does export openness increase firm-level output volatility? World Econ. 32 (4), 531–551. Buchinsky, M., 1998. Recent advances in quantile regression models: a practical guideline for empirical research. J. Hum. Resour. 33 (1), 88–126. Burgess, S., Lane, J., 2001. Churning dynamics: an analysis of hires and separations at the employer level. Labour Econ. 8 (1), 1–14. Caldera, A., 2010. Innovation and exporting: evidence from Spanish manufacturing firms. Rev. World Econ. 146 (4), 657–689. Canay, I.A., 2011. A simple approach to quantile regression for panel data. Econ. J. 14 (3), 368–386. Capasso, M., Treibich, T., Verspagen, B., 2015. The medium-term effect of R&D on firm growth. Small Bus. Econ. 45 (1), 39–62. Cassiman, B., Martinez-Ros, E., Golovko, E., 2010. Innovation, exports and productivity. Int. J. Ind Organiz 28 (4), 372–376. Clerides, S.K., Lach, S., Tybout, J.R., 1998. Is learning by exporting important? Micro-dynamic evidence from Colombia, Mexico, and Morocco. Q. J. Econ. 113 (3), 903–947. Cohen, W.M., Levinthal, D.A., 1989. Innovation and learning: the two faces of R&D. Econ. J. 99 (397), 569–596. Cohen, W.M., Levin, R.C., Mowery, D.C., 1987. Firm size and R & D intensity: a Re-examination. J. Ind. Econ. 35 (4), 543–565. Crépon, B., Duguet, E., Mairesse, J., 1998. Research, innovation and productivity: an econometric analysis at the firm level. Econ. Innov. New Technol. 7 (2), 115–158. Czarnitzki, D., Delanote, J., 2012. Young Innovative Companies: the new high-growth firms? Ind. Corp. Change 22 (5), 1315–1340. Czarnitzki, D., Wastyn, A., 2010. Competing Internationally: on the Importance of R&D for Export Activity. ZEW-DP 10-071 https://www.econstor.eu/bitstream/ 10419/41430/1/636382338. pdf. Dachs, B., Peters, B., 2014. Innovation, employment growth and foreign ownership of firms. Res. Policy 43 (1), 214–232. ˇ Polanec, S., 2010. From innovation to exporting or vice Damijan, J.P., Kostevc, C., versa? World Econ. 33 (3), 374–398. Davis, S.J., Haltiwanger, J., 1992. Gross job creation, gross job destruction, and employment reallocation. Q. J. Econ. 107 (3), 819–863. Del Monte, A., Papagni, E., 2003. R&D and the growth of firms: empirical analysis of a panel of Italian firms. Res. Policy 32 (6), 1003–1014. Di Cintio, M., Grassi, E., 2016. Uncertainty, flexible labour relations and R&D expenditure. Metroeconomica, http://dx.doi.org/10.1111/meca.12127. Dosi, G., Marsili, O., Orsenigo, L., Salvatore, R., 1995. Learning, market selection and the evolution of industrial structures. Small Bus. Econ. 7 (6), 411–436. Dosi, G., Grazzi, M., Moschella, D., 2015. Technology and costs in international competitiveness: from countries and sectors to firms. Res. Policy 44, 1795–1814. Esteve-Pérez, S., Rodríguez, D., 2013. The dynamics of exports and R&D in SMEs. Small Bus. Econ. 41 (1), 219–240.

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006

G Model RESPOL-3393; No. of Pages 17

ARTICLE IN PRESS M. Di Cintio et al. / Research Policy xxx (2017) xxx–xxx

Falk, M., 2012. Quantile estimates of the impact of R&D intensity on firm performance. Small Bus. Econ. 39 (1), 19–37. Flaig, G., Stadler, M., 1994. Success breeds success. The dynamics of the innovation process. Empir. Econ. 19 (1), 55–68. Forsman, H., 2011. Innovation capacity and innovation development in small enterprises. A comparison between the manufacturing and service sectors. Res. Policy 40 (5), 739–750. Geroski, P.A., 1990. Innovation, technological opportunity, and market structure. Oxford Econ. Pap. 42 (3), 586–602. Goedhuys, M., Sleuwaegen, L., 2010. High-growth entrepreneurial firms in Africa: a quantile regression approach. Small Bus. Econ. 34 (1), 31–51. Grazzi, M., 2012. Export and firm performance: evidence on productivity and profitability of Italian companies. J. Ind. Compet. Trade 12, 413–444. Greenaway, D., Kneller, R., 2007. Firm heterogeneity, exporting and foreign direct investment. Econ. J. 117 (517), F134–F161. Greenaway, D., Kneller, R., 2008. Exporting, productivity and agglomeration. Eur. Econ. Rev. 52 (5), 919–939. Grossman, G., Helpman, E., 1990. Comparative advantage and long-run growth. Am. Econ. Rev. 80 (4), 796–815. Grossman, G., Helpman, E., 1991. Innovation and Growth in the Global Economy. MIT Press, Cambridge, MA. Grossman, G.M., Helpman, E., Szeidl, A., 2006. Optimal integration strategies for the multinational firm. J. Int. Econ. 70 (1), 216–238. Hölzl, W., 2009. Is the R&D behaviour of fast-growing SMEs different? Evidence from CIS III data for 16 countries. Small Bus. Econ. 33 (1), 59–75. Hall, B.H., Lotti, F., Mairesse, J., 2008. Employment, innovation, and productivity: evidence from Italian microdata. Ind. Corp. Change 17 (4), 813–839. Hall, B.H., Lotti, F., Mairesse, J., 2009. Innovation and productivity in SMEs: empirical evidence for Italy. Small Bus. Econ. 33 (1), 13–33. Haltiwanger, J., Jarmin, R.S., Miranda, J., 2013. Who creates jobs? Small versus large versus young. Rev. Econ. Stat. 95 (2), 347–361. Henderson, R., Cockburn, I., 1996. Scale, scope, and spillovers: the determinants of research productivity in drug discovery. RAND J. Econ. 27 (1), 32–59. Jovanovic, B., 1982. Selection and the evolution of industry. Econometrica 50 (3), 649–670. Kleinknecht, A., 1987. Measuring R&D in small firms: how much are we missing? J. Ind. Econ. 36 (2), 253–256. Klette, T.J., Førre, S.E., 1998. Innovation and job creation in a small open economy-evidence from Norwegian manufacturing plants 1982–92. Econ. Innov. New Technol. 5 (2–4), 247–272. Klette, T.J., Griliches, Z., 2000. Empirical patterns of firm growth and R&D investment: a quality ladder model interpretation. Econ. J. 110 (463), 363–387. Klette, T.J., Kortum, S., 2004. Innovating firms and aggregate innovation. J. Polit. Econ. 112 (5), 986–1018. Koenker, R., Bassett Jr., G., 1978. Regression quantiles. Econometrica 46 (1), 33–50. Koenker, R., Hallock, K., 2001. Quantile regression: an introduction. J. Econ. Perspect. 15 (4), 43–56. Lachenmaier, S., Rottmann, H., 2011. Effects of innovation on employment: a dynamic panel analysis. Int. J. Ind Organ. 29 (2), 210–220. Lachenmaier, S., Wößmann, L., 2006. Does innovation cause exports? Evidence from exogenous innovation impulses and obstacles using german micro data. Oxford Econ. Pap. 58 (2), 317–350. Lee, L.F., 1992. Amemiya’s generalized least squares and tests of overidentification in simultaneous equation models with qualitative or limited dependent variables. Econometric Rev. 11 (3), 319–328. Leiponen, A., 2000. Competencies, innovation and profitability of firms. Econ. Innov. New Technol. 9 (1), 1–24. Leiponen, A., 2005. Skills and innovation. Int. J. Ind Organ. 23 (5–6), 303–323. Melitz, M.J., 2003. The impact of trade on intra-industry reallocations and aggregate industry productivity. Econometrica 71 (6), 1695–1725.

17

Michie, J., Sheehan, M., 1999. HRM practices, R&D expenditure and innovative investment: evidence from the UK’s 1990 workplace industrial relations survey (WIRS). Ind. Corp. Change 8 (2), 211–234. Moser, C., di Mauro, B.W., Urban, D., 2010. International competitiveness, job creation and job destruction − an establishment-level study of German job flows. J. Int. Econ. 80 (2), 302–317. Nelson, R.R., Winter, S.G., 1982. An Evolutionary Theory of Economic Change. Harvard University Press, Cambridge, Mass. Newey, W.K., 1987. Efficient estimation of limited dependent variable models with endogeneous explanatory variables. J. Econ. 36 (3), 231–250. Parisi, M.L., Schiantarelli, F., Sembenelli, A., 2006. Productivity, innovation and R&D: micro evidence for Italy. Eur. Econ. Rev. 50 (8), 2037–2061. Petit, P., 1995. Employment and technological change. In: Stoneman, P. (Ed.), Handbook of the Economics of Innovation and Technological Change. North Holland, Amsterdam, pp. 366–408. Pianta, M., 2005. Innovation and employment. In: Fagerberg, J., Mowery, D., Nelson, R. (Eds.), Handbook of Innovation. University Press Oxford, Oxford, pp. 568–598. Pindyck, R.S., 1991. Irreversibility, uncertainty, and investment. J. Econ. Lit. 29 (3), 1110–1148. Piva, M., Vivarelli, M., 2005. Innovation and employment: evidence from Italian microdata. J. Econ. 86 (1), 65–83. Rao, H., Drazin, R., 2002. Overcoming resource constraints on product innovation by recruiting talent from rivals: a study of the mutual fund industry, 1986–94. Acad. Manage. J. 45 (3), 491–507. Rodríguez, José L., Rodríguez, R.M.G., 2005. Technology and export behaviour: a resource-based view approach. Int. Bus. Rev. 14 (5), 539–557. Romero, I., Martínez-Román, J.A., 2012. Self-employment and innovation. Exploring the determinants of innovative behavior in small businesses. Res. Policy 41 (1), 178–189. Roper, S., 1999. Under-reporting of R&D in small firms: the impact on international R&D comparisons. Small Bus. Econ. 12 (2), 131–135. Santarelli, E., Sterlacchini, A., 1990. Innovation, formal vs. informal R&D, and firm size: some evidence from Italian manufacturing firms. Small Bus. Econ. 2 (3), 223–228. Schmiemann, M., 2008. Enterprises by Size Class-overview of SMEs in the EU Statistics in Focus, No. 31/2008. Eurostat, Luxembourg. Smith, R.J., Blundell, R.W., 1986. An exogeneity test for a simultaneous equation Tobit model with an application to labor supply. Econometrica 54 (3), 679–685. Sousa, N., Rueda-Cantouche, J., Arto, I., Andreoni, V., 2012. Extra-EU Exports and Employment. European Commission, DG Trade Chief Economist Note http:// ftp.jrc.es/EURdoc/JRC71120. pdf. Staiger, D., Stock, J.H., 1997. Instrumental variables regression with weak instruments. Econometrica 65 (3), 557–586. Sterlacchini, A., 1999. Do innovative activities matter to small firms in non-R&D-intensive industries? An application to export performance. Res. Policy 28 (8), 819–832. Van Reenen, J., 1997. Employment and technological innovation: evidence from UK manufacturing firms. J. Labor Econ. 15 (2), 255–284. Vivarelli, M., 1995. The Economics of Technology and Employment: Theory and Empirical Evidence. Edward Elgar Publishing, London. Vivarelli, M., 2015. Innovation and employment. IZA World Labor 154, 1–10. Wagner, J., 1995. Exports, firm size, and firm dynamics. Small Bus. Econ. 7 (1), 29–39. Wagner, J., 2007. Exports and productivity: a survey of the evidence from firm-level data. World Econ. 30 (1), 60–82. Wakelin, K., 1998. Innovation and export behavior at the firm level. Res. Policy 26 (7–8), 829–841. Yasuda, T., 2005. Firm growth, size, age and behavior in Japanese manufacturing. Small Bus. Econ. 24 (1), 1–15.

Please cite this article in press as: Di Cintio, M., et al., Firm growth, R&D expenditures and exports: An empirical analysis of italian SMEs. Res. Policy (2017), http://dx.doi.org/10.1016/j.respol.2017.02.006