Innovation and job creation: a sustainable relation?

This study compared the employment growth patterns of innovative and non-innovative firms, focusing on whether or not there are systematic differences between these two categories in the persistence of the jobs they create. To this end, a unique longitudinal dataset of 3304 Spanish firms over the period 2002–2009 and a semi-parametric quantile regression approach was used. The empirical results indicate that, ceteris paribus, innovative, smaller and younger firms are more likely to experience high employment growth episodes than non-innovative firms. More interestingly, among those firms that contribute more to yearly job creation (e.g. high-growth firms), only innovative companies are able to sustain high growth over time (in contrast to non-innovative firms). In addition, among declining firms, non-innovators tend to deteriorate faster in terms of economic performance than innovators.


Introduction
The aim of the analysis presented in this study is to contribute to the discussion on the growth of firms and the related Gibrat's law (1931) and to verify if there are differences in the speed and pattern of growth of innovative and non-innovative firms. 1 Although Gibrat's law predicts that the growth rate of a firm is unpredictable and stochastic in nature (Geroski et al. 1997), the existing empirical evidence on year-to-year growth patterns at the firm level is still the subject of discussion, and empirical evidence on innovative firms is quite limited. In his review of Gibrat's legacy, Sutton (1997) found that half a century of testing had revealed a series of statistical regularities that were incompatible with the view that firms' growth rate is random. Most remarkably, he pointed out that small firms generally appeared to grow faster than large ones, and that growth rates were serially correlated. In addition, while early empirical studies into the growth of firms encountered positive autocorrelations ranging from 30 to 33 % (Ijiri and Simon 1967;Singh and Whittington 1975;Dunne and Hughes 1994), 2 more recent studies-relying on longer time series-have found more diverse annual autocorrelation patterns (Coad 2007), 3 in spite of the fact that one would expect persistence to prevail when measured over a shorter time horizon. 4 According to Coad (2009), there are several reasons why these mixed results may emerge. He argued that serial correlation changes with two characteristics of the firm, namely its size and its growth rate. Stated simply, there is no 'one size fits all' serial coefficient. In a previous study (Coad 2007), he also showed that autocorrelation coefficients are systematically affected by firms' sizes and that small firms are often subject to a negative correlation of annual growth rates, whereas larger firms display a positive correlation. Ciriaci et al. (2014) found similar results for Spanish innovative firms' employment, sales and innovative sales growth. 5 1 According to the Community Innovation Survey (CIS) definition, innovative firms are those that have answered positively to at least one of the following four questions: (1) During the period 1998-2000, has your enterprise introduced on the market any new or substantially improved products? (2) During the period 1998-2000, has your enterprise introduced any new or substantially improved production processes? (3) By the end of 2000, did your enterprise have any ongoing innovation activities? (4) During the period 1998-2000, did your enterprise have any innovation activities that were abandoned? 2 These studies generally consider growth serial correlation over periods of 4-6 years. 3 For simplification, hereafter the focus will be on those studies analysing positive relative growth rates. 4 For instance, Chesher (1979), Wagner (1992), Geroski et al. (1997), Weiss (1998), Bottazzi et al. (2002) and Bottazzi and Secchi (2003) found positive serial correlations for UK quoted firms, German manufacturing firms, Austrian firms, the worldwide pharmaceutical industry and US manufacturing. Negative serial correlation has been observed, in contrast, for German firms by Boeri and Cramer (1992), for quoted Japanese firms by Goddard et al. (2002) and for Italian and French manufacturing firms by Bottazzi et al. (2007Bottazzi et al. ( , 2011. Finally, a number of studies did not find any significant autocorrelation in firms' growth rates [e.g. Almus and Nerlinger (2000) analysing German start-ups, Bottazzi et al. (2002) for selected Italian manufacturing sectors, Geroski and Mazzucato (2002) for the US automobile industry and Lotti et al. (2003) for Italian manufacturing firms]. 5 They also highlighted, however, that smaller firms tend to outperform larger ones if growth in employment and sales is considered, while, in the case of high innovative sales growth episodes, larger firms tend to outperform smaller firms. When it comes to the ability to experience high growth innovation episodes, therefore, larger firms have a competitive advantage.
In the light of all the current policy efforts seeking to stimulate both job creation and innovation through research and development (R&D) spending, it appears to be fairly relevant to know to what extent innovative firms have a growth advantage and whether or not the jobs created by innovative and non-innovative firms are persistent. The present study contributes to this policy debate and to the aforementioned empirical literature using an original panel of 3304 Spanish firms over the period [2002][2003][2004][2005][2006][2007][2008][2009]. We thus compare the employment growth patterns of innovative 6 and non-innovative firms to verify whether or not being innovative determines a growth advantage and whether or not 'innovation' helps to create (enduring) jobs over time. We also test whether or not being concurrently young, innovative and relatively small (i.e. being a 'young innovative company '-YIC 7 -Veugelers 2008;Schneider and Veugelers 2010) implies an employment growth advantage. 8 To these ends, we use a semi-parametric quantile regression approach, which allows us to investigate if there are asymmetries in downsizing and upsizing processes (i.e. negative and positive employment growth rates, respectively), namely analysing how employment serial correlation of innovative and non-innovative firms changes along their growth rate distribution. In contrast to standard least squares regression techniques, 'which provide summary point estimates that calculate the average effect of the independent variables on the average firm' (Coad and Rao 2008), this econometric model makes the most of sample heterogeneity and is particularly suitable when outliers (i.e. high-growth and declining firms) are of particular interest.
The paper is organised as follows: This introduction is followed by a section providing a review of the relevant literature on firms' growth, with a focus on if (and why) innovative and non-innovative firms are characterised by different employment growth patterns. The database and the applied methodological approach are both presented in Sect. 3. Section 4 illustrates the empirical results, while Sect. 5 concludes the paper and discusses the policy implications of the obtained results.

Literature
The so-called Gibrat's law (1931) is a seminal reference when dealing with the statistical characteristics of firms' growth dynamics. It is also sometimes referred to as the 'law of proportionate effect' because its basic underlying hypothesis is that a firm's growth is independent of size and is simply driven by several small idiosyncratic events (Bottazzi et al. 2002). 9 6 Innovative firms are defined for the purposes of this study as firms that, during the period 2002-2009, had introduced products/processes new to the market and/or new to the firm, and those that stated that they have invested constantly in R&D. 7 We use the European definition of YICs as specified in Article 35 of the General Block Exemption Regulation (GBER). According to this definition, a YIC has fewer than 250 employees, is younger than 6 years old and spends at least 15 % of its revenues on R&D. 8 For the sake of completeness and as a robustness check, we tested the same hypotheses for sales growth. 9 Gibrat's law can be considered the first attempt to explain, in stochastic terms, the systematically skewed pattern of the distribution of firms' size within an industry (Aitchison andBrown 1957). Eurasian Bus Rev (2016) 6:189-213 191 The simplicity of its formulation has led to several waves of studies, but an allencompassing review of them would go beyond the scope of our research. Notwithstanding, it is worth highlighting that several influential surveys on firms' intra-industry dynamics (e.g. Sutton 1997;Geroski 1998) concluded independently that the empirical evidence does not always support Gibrat's general hypothesis. 10 In a recent survey, Santarelli et al. (2006) concluded that one cannot consider Gibrat's law either generally valid or systematically rejected, as the results strongly depend on the sector analysed and the size composition of the firms in the sample. 11 All in all, the theoretical propositions and the corresponding empirical evidence suggest that several firms' structural characteristics are fundamental determinants of firms' growth. Schumpeter (1942) emphasised the positive influence of size on a firm's ability to grow and especially to innovate. Following this concept, a number of theoretical studies have claimed that larger companies have potential benefits because of, for example, economies of scale, lower risk and larger markets. In addition, they have greater opportunities for appropriation (Fernández 1996), which enables them, inter alia, to undertake comprehensive R&D projects and so thrive, owing to the innovations resulting from such activities. Nelson and Winter (1982) have shown that differences in firms' performances and growth are also affected by firms' age, thus starting an even more controversial debate. 12 In fact, neoclassical models consider growth as a means to reach a unique optimum size and a minimum efficient scale of production. In this respect, age does not have any 'role'. Instead, firm age only plays a role if growth is seen as an outcome of a 'learning-by-doing' process (Arrow 1962). In this regard, firm age is relevant, as younger firms tend to suffer more inefficiencies and consequently may be systematically disadvantaged in terms of growth. In an evolutionary perspective (Nelson and Winter 1982), age may affect growth in different directions, depending on the underlying mode of innovation: in a 'routinised regime', age may have a positive effect on growth, given that innovations tend to be generated by accumulated non-transferable knowledge, while in an 'entrepreneurial regime', 10 In this regard, see also Lotti et al. (2009). The authors defend Gibrat's law as a long-run regularity. 11 For instance, Fotopoulos and Giotopoulos (2010) rejected Gibrat's law for the total sample of Greek manufacturing firms, but they confirmed that, once size and age are taken into account, more heterogeneous results emerge. They also highlighted that the law had to be rejected for micro-, small and young firms, but could be accepted for medium-sized, large and old firms. The former statement implies that the growth patterns of the latter categories follow a random path and do not tend to persist in subsequent periods. Ciriaci et al. (2014) found instead a random pattern for small and medium-sized enterprises' employment growth, and negative autocorrelation of sales (first order) and innovative sales (first and second order) both for small and medium-sized enterprises and for large firms. Hall (1987) found (for publicly traded companies in the US manufacturing sector) that most of the change in employment at the firm level in any given year is permanent, that year-to-year growth rates are largely uncorrelated over time or with prior characteristics of the firm, and that there is almost no measurement error. Given this, Gibrat's law was weakly rejected for the smaller firms in the sample and accepted for the larger firms. Similarly, Hart and Oulton (1996), based on an analysis of 87,000 independent UK companies broken down by size group, confirmed a negative relationship between a firm's growth and its initial size for only the smallest firms (eight employees or fewer), while for larger firms no relationship between growth and size emerged. 12 See, for example, Navaretti et al. (2014) for a discussion and some empirical testing of the firm agegrowth link. age may be negatively correlated with growth, given that knowledge is not of a routine nature. In this respect, Geroski (2000) pointed to the unpredictable, stochastic and idiosyncratic nature of growth itself and concluded that this might be, inter alia, due to the very nature of innovation. As the outcome of innovations is highly unforeseeable, firm-level innovations could thus be the key determinant of the growth of companies.
Other authors have emphasised the role of international competition in fostering firms' efficiency and innovativeness and, consequently, that another structural characteristic influencing a firm's growth is its presence in external markets, namely its export orientation (see, for example, Castellani 2002, for a survey). In addition, the literature related to differences in growth behaviour between innovative 13 and non-innovative firms, in general, supports the hypothesis that innovating firms both are more profitable and grow faster than non-innovators (Geroski and Machin 1992;Freel 2000). In particular, innovative activities are widely found to be essential for explaining growth patterns at the firm level. For instance, Cefis and Marsili (2005) examined the effects of innovation on firm survival using data on manufacturing firms active in the Netherlands. They found that firms, in general, tend to benefit from an 'innovation premium' that extends their life expectancy. However, they also found that the significantly positive effect of innovation activities on the probability of firms' survival increases over time (Cefis and Marsili 2006). The authors concluded that, although small and young firms are the most exposed to the risk of exit, as also found in earlier studies, they are also the firms that are most likely to survive, especially in the longer term, thanks to innovation. 14 Colombelli et al. (2014) underlined the idea of an innovation premium by investigating highgrowth firms ('gazelles') and pointing out that increasing growth rates are associated with exploration (and exploitation) of new knowledge rather thanonly-exploitation of existing knowledge, created by others. 15 Notwithstanding the kind of innovations introduced into the market (i.e. incremental or radical), innovation increases profitability and drives organic growth (Geroski and Machin 1992). From a theoretical point of view, the main reason underpinning a growth differential between innovative and non-innovative companies is the process of R&D, which accompanies the creation and adoption of innovation and is likely to increase the firm's external absorptive capacity (Cohen 13 Please note that there are other economic factors determining the growth of an innovative company, such as the so-called intangible assets, which in turn are very much dependent on firm, technological, sector and socio-economic/market environment characteristics. Furthermore, complementarities among several types of investment at the firm level (such as R&D, human capital, information and communication technologies, physical capital, international collaboration) were identified as being very important, with the potential for higher returns if realised jointly instead of devoting resources to only one of these activities. 14 In this regard, see, for instance, Mohnen and Hall (2013), who reviewed the existing evidence regarding the effects of technological and non-technological innovations on the productivity of firms and the existence of possible complementarities between these different forms of innovation. 15 See, for example, Dosi and Nelson (2013) for a reflection of this concept from an evolutionary perspective (and in the context of the relevant literature on technological advancement). The role of entrepreneurship for firm entry and survival, for example, and in this regard also for growth patterns and innovation is discussed by, for instance, Vivarelli (2013a). and Levinthal 1990) as well as its internal knowledge base, increasing firms' flexibility and adaptability (Freel 2000). Other interesting managerial studies (Steiner and Olaf 1988;Baldwin and Johnson 1996) pointed to the idea that innovative firms grow more than non-innovative firms not only because they strongly focus on innovation and technological advance, but also because they are more concerned about human resources, markets and products, financing, and management skills and practices than non-innovative firms. In other words, they tend to value higher (and spend more on) each of these areas than non-innovative firms, and they tend to complement higher investments in R&D with higher investments in marketing and training, for instance.
In general, to the extent that innovation effort leads, through the introduction into the market of new products and/or processes, 16 to a (temporary) period of increased sales or profits (Geroski and Machin 1992;Freel 2000), the innovative firm is likely to be able to gain market shares at the expense of the non-innovative firm. These benefits are likely to persist insofar as the innovative firm is able to exert property rights or effectively employ other appropriability devices (e.g. learning curve effects, secrecy, first mover advantages) (Dosi 1988). Freel (2000) confirms it by pointing out that superior innovative firms' growth in employment, if it occurs, derives from increased sales and improved competitiveness (which are likely to be the direct consequences of the introduction of successful innovation). Several studies (Freel 2000;Coad and Rao 2008) have shown that the propensity for innovation is a firm's growth driver. In addition, a firm's propensity to innovate is considered a positive predictor of survival and of above-average post-entry performance of start-ups (Quatraro and Vivarelli 2014). 17 Accordingly, as regards the specific relationship between innovation 18 and job creation, there is a large body of literature at the micro-level that generally confirms a positive link between the two (see, for instance, Evangelista and Savona 2003;Mansury and Love 2008). 19 However, Greenan and Guellec (2000) found that the positive employment impact of product and process innovation at the firm level could disappear at the industry level. In fact, innovative firms may temporarily face no demand constraint (product innovation effect) and, given that they also operate more efficiently, they can expand output 16 It has been argued that process innovation, owing to its cost-cutting nature, may have a more ambiguous effect (Edquist et al. 2001). However, while clearly distinguishable at the level of the individual firm or industry, such differences tend to become more blurred at the level of the overall economy, because the product of one firm (or industry) may end up being used to produce goods or services in another. 17 For some controversial results on the relationship between start-up size and survival (i.e. implicitly growth rate and growth persistence), see, for example, Audretsch et al. (1999), who studied 1570 new Italian manufacturing firms and tracked their post-entry evolution for 6 years. The authors found that hazard rates increased markedly during the first 2 years and then tended to decrease, with a final survival rate after 6 years of activity equal to 59.1 %. 18 Note that some of the studies mentioned approximated firms' innovative activities by using their R&D expenditures. 19 See also Vivarelli (1995Vivarelli ( , 2013b, van Reenen (1997), Pianta (2005), Bogliacino and Pianta (2010), Coad and Rao (2011), Lachenmaier and Rottmann (2011), Bogliacino et al. (2012), Evangelista and Vezzani (2012) and Harrison et al. (2014). and jobs at the expense of competitors. 20 Conversely, as competitors in the same industry (sector) react, the overall industry (sector) potential for job creation could be later constrained by the industry demand. The latter phenomenon mitigates the innovative firm's initial higher competitiveness and growth in terms of business output and jobs creation (see, for example, Bogliacino et al. 2013). In another study, Bogliacino (2014) found that the generally positive impact of R&D and innovation on employment-empirically confirmed in this study by analysing top R&Dinvesting companies-varies according to how much the firm invests and also with its size in terms of sales. Hence, the positive job creation effect increases with the R&D intensity and the size of the firm.
To sum up, while the contributions of size, age and innovation on firms' (sales and employment) growth have been extensively investigated, the extent to which there are systematic differences in the persistence (if any) of the jobs created by innovating versus non-innovating firms has not. The analysis of such differences is important not only because it can shed light on firms' performance in general, but also because analysing firms' growth contributes to accumulating knowledge on the growth pattern of the economy itself (van Stel et al. 2005; Carree and Thurik 2010).

Database
This research draws on an original balanced panel of 3304 Spanish firms over the period 2002-2009 (comprising a total of 26,432 observations), obtained by the authors matching eight waves of the annual Spanish Community Innovation Survey (Encuesta sobre Innovación en las Empresas). 21 As in this work we were exclusively interested in internal growth-as opposed to growth owing to mergers, acquisitions and disinvestments-firms that had undergone significant structural modifications were excluded. Therefore, in the initial stages of data processing, we excluded those firms that in any year had declared an increase in turnover of 10 % (or more) owing to a merger with another enterprise or part of it (this is the only information provided by the survey on this point). Similarly, any firms that declared a significant decrease in turnover (10 % or more) owing to sale or closure of part of the enterprise were also dropped. Consequently, the study captures only organic growth, and no fast growth induced by acquisition and/or fast decline owing to sale or closure of a firm. 22 In addition, following the methodology of Hall and Mairesse (1995), the dataset was 'cleaned' by removing all observations for which employment and/or sales were stated to be zero or missing. As a result, the total number of observations decreased from 26,432 (3304 firms) to 25,426 (3178 firms).
Almost 77 % of the sample firms are innovative firms; that is firms that over the period 2002-2009 introduced products/processes new to the market and/or new to the firm and declared that they invest in intramural R&D. Their average R&D intensity (in terms of turnover) is 1.8 %. It is worth mentioning that, among innovative firms (for instance, in 2002), about 2.5 % were YICs (i.e. companies younger than 6 years old, with fewer than 250 employees and with an R&D intensity (in terms of turnover) of at least 15 %). In general, 41.4 % of the sample firms were relatively large companies. The remaining 58.6 % were small and medium-sized enterprises (of which 43.2 % were medium-sized enterprises, 49 % were small firms and about 7.8 % were micro firms). 23 Firms were classified based on their sector of principal activity. Unfortunately, the data provided by the Spanish National Statistics Institute (Instituto Nacional de Estadística, INE) were grouped based on industry classification at a relatively high aggregation level (ten main sectors; see Table 4 in the Appendix). The sectoral breakdown is as follows: 45.7 % manufacturing, 12.4 % scientific and technical R&D, 11.1 % retail trade, 5.3 % construction, 4.8 % finance and insurance, 3.4 % transport, 2.8 % water supply and about 1 % mining and quarrying.
In summary, the resulting sample, compared with the data used in similar studies, has three main advantages. First, the dataset includes innovative firms from the manufacturing and the service sectors 24 and, apart from large-scale companies, it also includes small and micro firms (both frequently neglected). Second, thanks to the merging of annual survey waves, we have yearly data that are compared with other empirical studies [e.g. relying on Community Innovation Survey (CIS) data], a relatively high frequency allowing considerations of year-to-year sales and employment changes. Third, the data include additional company characteristics, such as company age. This may help to overcome certain weaknesses that usually appear when working with anonymised data.
Unfortunately, because of its nature of being a balanced panel, working with this dataset also had some drawbacks, as it does not contain firms entering or exiting the market at any time during the period of interest. In fact, by considering only firms that were already in the market at the beginning of the observed period and survived until the end, a certain bias was introduced, as the negative growth rates of those firms that left the market and the high growth rates of 'new-born' firms were both left out. However, over the analysed period, the Spanish economy was characterised by a relatively low exit rate, 25 which limits the corresponding bias to a certain extent. 26 Moreover, according to Lopez-Garcia et al. (2009), in general, the firstyear employment growth of Spanish start-ups tends to be higher than that of subsequent years. In the empirical analysis, we controlled for the age of each firm (innovative and non-innovative), which may have partially corrected the outlined bias. In addition, controlling for firms' age may have further compensated for the balanced nature of the panel because, for any given size of firm, the probability of exit is a decreasing function of a firm's age, as older firms are likely to have more precise estimates about their innate efficiency, thereby reducing the likelihood of failure (Farinas and Moreno 2000). Table 1 illustrates the evolution of innovative companies in Spain over the period 2002-2009 and shows that, in 2009, they represented almost 20.5 % of Spanish firms. Not surprisingly, this percentage is significantly lower that the percentage of innovative firms in our sample, obtained by matching several waves of the Spanish CIS. Interestingly, both the innovation intensity 27 of Spanish innovative firms and the percentage of turnover owing to new or significantly improved products (the socalled innovative sales) are increasing over time (from 1.8 to 2.2 % and from 8.6 to 14.9 %, respectively). However, it is also interesting to note that, although we observe a positive trend, a closer look at the data shows that the positive trend is limited to the period 2002-2007, whereas a downturn is observed in 2009.

Descriptive statistics
We have examined in the univariate distribution of firms' employment growth rates (yearly changes in the number of employees) during the period 2003-2009, which have been cleaned of size 28 dependence, serial correlation and heteroskedasticity following Bottazzi et al. (2011) and Coad and Rao (2008). Irrespective of the year considered, we found that the distribution of firms' employment growth rates is tent shaped. Stated simply, this suggests that there is, compared with a normally distributed variable, a higher probability of finding growth rates near the mean and also a higher probability of rather extreme values. Empirically, this result suggests that a minority of firms experience very rapid growth or very rapid decline, while the average firm does not grow at all. These findings point to the Laplace 25 See, for example, Ruano (2000), Nunez (2004), Lopez-Garcia and Puente (2007a, b) and Lopez-Garcia et al. (2009). 26 To verify if this argument also holds in our case, we ran a robustness check limiting the time period analysed to the growth period (2002)(2003)(2004)(2005)(2006)(2007), namely excluding the two crisis years (2008)(2009). Results were confirmed. 27 The innovation intensity is calculated as the percentage of the expenditure on innovative activities on turnover. Source: INE. 28 In the following text, firm size corresponds to the number of employees. distribution and appear virtually the same as the empirical growth rate distribution commonly observed for CIS data (see, for example, Hölzl and Friesenbichler 2010).
In addition, the descriptive analysis confirms that the density distributions of employment growth rates for innovative and non-innovative firms differ significantly. This difference implies diverse growth dynamics. In fact, the observed yearly growth rate distributions for non-innovative firms have longer tails (i.e. the number of extreme events-rapid growth or rapid decline-is higher than in the case of innovative firms 29 ), whereas the peak of the Laplace distribution is significantly higher for innovative firms. 30 The anlaysis of the employment growth rate distribution for innovative and non-innovative firms outlines similar differences all of the years considered in the analysis: the tail for non-innovative firms is slightly 'fatter' than the tail for innovative firms. This result has a theoretical and an empirical implication. First, it confirms that the Laplace distribution is less appropriate for approximating non-innovative firms' growth rate distribution, as it appears unsymmetrical, while it seems to be a fairly good approximation of the employment growth rate distribution for innovative firms. Second, it suggests that, for non-innovative firms, rapid decline is more likely to occur than rapid growth. All in all, these findings underline that growth rate distributions are quite stable over time and probably always display fat tails, where outperformers and underperformers are concentrated.
From a methodological point of view, these descriptive statistics imply that regression estimates based on the assumption of normally distributed standard errors may perform poorly, mainly owing to the presence of (significant numbers of) 'outliers' (Coad 2007). In addition, in line with the general aim of this study, we should focus on firms that grow largely more (or largely less) than the average (that are experiencing very low growth rates), as understanding the reasons behind these growth patterns appear to be of the utmost policy relevance. An econometric analysis focusing on the 'average' firm, in this particular respect, would arguably be of fairly limited interest. That is why the following analysis is based on the use of quantile regression techniques, which are robust to outliers and allow the autocorrelation structure across the entire distribution of employment growth rates to be investigated.

The quantile regression approach
The quantile regression model is a semi-parametric technique first introduced by Koenker and Bassett (1978) and has been used since its introduction by many authors (e.g. Rao 2006, 2008). The technique has several useful features that make its application particularly appropriate when dealing with very heterogeneous observations whose heterogeneity, however, is of interest. In essence, when analysing firms' growth rates, outliers are carriers of fundamental information that should be preserved rather than eliminated. As such, the quantile regression model can be used to exploit and characterise the entire conditional distribution of a dependent variable (i.e. a firm's employment growth rate) given a set of regressors and control variables (Buchinsky 1998). This methodological approach allows taking into account the fact that different solutions at distinct quantiles may reflect differences in the response of the dependent variable to changes in the regressors at various points in the conditional distribution of the dependent variable itself (Buchinsky 1998). In other words, by applying this technique we account for the fact that the autocorrelation between a firm's growth rates is not the same for all firms regardless of their size (or sector; see Coad and Rao 2008, and ''Box A1'' in the Appendix for details). 31 The regression model to be estimated for this study is specified as follows 32 : Growth i;t ¼ a 0 þ a 1 log size i;tÀ1 À Á þ a 2 INNO Ã Growth i;tÀ1 þ a 3 INNO Ã Growth i;tÀ2 þ a 4 non-INNO Ã Growth i;tÀ1 þ a 5 non-INNO Ã Growth i;tÀ2 þ a 6 INNO þ a 7 year þ a 8 age i þ a 9 sector i þ a 10 YICs þ e i;t where size i is the logarithm of the number of employees at t -1; 'year' is a vector of yearly dummies accounting for macro-economic phenomena common to all firms (such as inflation, market cycles, etc.); age i is the age of the firm (expressed in years and inserted to control for the degree of establishment of both innovative and non-innovative firms); sector i is a vector of industry dummies; YICs is a dummy identifying the so-called 'young innovative companies' (i.e. a firm-specific attribute 33 ); and e i,t is the vector of 31 In addition, in the case of quantile regressions, the error terms do not need to satisfy the restrictive assumption according to which they must be identically distributed at all points of the conditional distribution. 32 See the ''Appendix'' for further methodological details. 33 Using data over the time period 2002-2009 implies that YIC status changes over time. In fact, if the YIC status was a time attribute and the YICs were oversized from an employment point of view at the beginning, they might have much more potential in terms of sales than employment growth. It is also worth noting that, given the YIC definition used, this dummy and the age control inserted for both innovative and non-innovative firms do not capture the same firm's characteristics.
residuals. 34 Finally, INNO is a dummy identifying innovative firms and INNO*Growth i,t-1 , INNO*Growth i,t-2 , non-INNO*Growth i,t-1 and non-INNO*Growth i,t-2 are four interaction terms introduced to disentangle differences in the first-and second-order autocorrelation of growth rates of innovative and non-innovative firms.
Although the correlation between all the regressors included in the equation turned out to be low, we checked for multi-collinearity, as three or more independent variables can be multi-collinear, without having high pair-wise correlations. However, the variance inflation factor is low (2.07), a result that is symptomatic of the lack of multi-collinearity. Finally, it is worth mentioning that, in quantile regressions, standard errors are bootstrapped, 35 which makes it harder to reach significance, but, at the same time, less likely that spurious results will be accepted as significant.

Empirical results
In this section, we present and discuss the results obtained from computing five quantile regressions (Q.10, Q.25, Q.50, Q.75 and Q.90), 36 and allow two lags in serial correlation (see Tables 2, 3). The reported coefficients can be interpreted as the partial derivative of the conditional quantile of the dependent variable with regard to the specific explanatory variable. The estimates and the corresponding statistical significance, along with the standard errors, are reported in Tables 2  and 3.

Results and discussion
The results obtained for the median firm (column 'Q.50' of Table 2) suggest that employment growth is not systematically determined by firm size, but that it is negatively affected by firm age (although with a very low elasticity), suggesting that younger firms grow more than the average. However, although the final result is in line with the corresponding literature (see, for instance, Coad and Tamvada 2012), it is not very telling that, in the median quantile, the majority of firms are not growing at all. Nevertheless, at first glance, these results would lead us to partly support Gibrat's law (1931), which implies the absence of any structure in the (employment) growth process. 34 We did not insert the interaction term 'year*sector' because of the loss of degrees of freedom that this would have implied. In addition, it is worth noting that, at the initial stage of the estimation, we also inserted a dummy to identify those firms belonging to a group, but it turned out not to be significant and was excluded from the final model specification. 35 Bootstrapping is used instead of the calculation of asymptotic standard errors, as this would require the estimation of densities (Angrist and Pischke 2009). 36 We also ran a robustness check limiting the time period analysed to the growth period (2002)(2003)(2004)(2005)(2006)(2007), namely excluding the two crisis years (2008 and 2009). The corresponding table is not reported (but available upon request) for the sake of brevity. However, the results obtained analysing the whole period were confirmed.  However, if we consider those firms that are not at (or around) the median, the empirical evidence reveals a much more diversified picture. In particular, nonrandom growth patterns and strong asymmetries in downsizing and upsizing processes emerged (see, in particular, the extreme quantiles, i.e. columns Q.10 and Q.90 of Table 2). 37 In fact, both innovative and non-innovative declining firms do not show any autocorrelation patterns in their growth rates. However, in the case of high-growth firms (column Q.90 of Table 2), a positive autocorrelation for innovative firms and a negative autocorrelation for non-innovative firms were found, and the F test conducted to test for the significance of these differences confirmed that finding. These last two results point towards the idea that being active in innovation tends to sustain a firm's (high) growth pattern, as a positive highgrowth episode of an innovative firm is likely to be followed by another positive high-growth episode in the following year. On the other hand, for non-innovative firms, negative autocorrelation of employment growth was found, that is these companies appear to be unable to repeat a positive high-growth episode achieved in 1 year in the following year and, therefore, do not experience stable growth patterns (i.e. a sustainable growth process with persistence of the created jobs). Therefore, innovative firms appear to possess the capabilities needed for sustained and high employment growth in an internationally competitive selection environment.
In addition, being innovative not only helps to sustain firms' employment growth patterns, but also, ceteris paribus, increases the level of employment growth itself across all quantiles. That is to say, innovative firms create more jobs than the average, a result that confirms our theoretical expectations. Therefore, innovation is a means to enhance Spanish employment growth and to achieve above-average performances. 38 Overall, these findings may be interpreted in the light of previous empirical evidence on the differences in adjustment cost for employment between innovative and non-innovative firms. According to Meghir et al. (1996), adjustment costs are significantly lower for firms with larger stocks of past innovations. In this sense, innovative firms may be considered more 'flexible' than non-innovators and may innovate not only to increase production, but also to insure themselves against shocks.
Other interesting differences among quantiles emerged when the roles of size and age were accounted for. For declining firms, both size and age entered the equation with a positive algebraic sign, whereas, for growing and fast-growing firms, both variables were significant and affected firms' growth performance negatively. In line with previous empirical evidence (see, for instance, Coad 2009;Ciriaci et al. 2014), 39 among firms declining in terms of employment figures, a higher company 37 It is worth noting that these asymmetries did not emerge in the case of the 'median firm' (column Q.50 of Table 2), confirming the need to properly account, from an econometric point of view, for firms' heterogeneity. 38 The latter result is in line with Calvo (2006) who, while analysing a panel of Spanish manufacturing firms, has shown that any activity leading to both process and product innovations is a strong positive factor in a firm's survival and employment growth. 39 These results are also in line with those of Fizaine (1968), Evans (1987a, b), Geroski and Gugler (2004) and Yasuda (2005), who showed that age has, as predicted by Jovanovic (1982), a negative impact on firms' employment and/or sales growth. Eurasian Bus Rev (2016) 6:189-213 203 age tends to limit the losses in employment growth. That is to say, among declining firms, older companies tend to decline less (e.g. laying off employees at a slower rate), whereas, among growing firms, the smallest and youngest tend to experience a faster employment growth track than older (and larger) firms. Therefore, the proportional rate of company employment growth decreases the older a firm is, pointing towards the idea that younger firms may need to adjust the levels of employment (Jovanovic 1982;Hopenhayn 1992;Ericson and Pakes 1995) more frequently than older firms because of their limited knowledge of the 'right operational size'. This result may also suggest that smaller and younger firms grow at a faster pace than larger and older firms because of the need to reach a minimum efficient scale of production and because they face lower adjustments costs than their larger counterparts (Meghir et al. 1996). Finally, when looking at the dummy for YICs, a clear distinction between declining/low-growth firms and high-growth firms emerged. Being a YIC was found to have a positive effect in periods of decline (for declining firms). Stated simply, being young and small and investing intensively in R&D apparently helps such firms to decline more slowly than the average. On the other hand, if high-growth episodes are considered, being small and/or young seems to be a 'sufficient precondition' for potentially experiencing high-growth episodes: the control inserted for YICs is not significant for high-growth firms. 40 This result contradicts Czarnitzki and Delanote (2013), who suggested that (Flemish) YICs grow more than other types of firms. 41 However, it should be noted that our result could be partly biased by the balanced nature of our panel. In fact, the high growth rates of 'newborn' firms were left out. 42 In the literature, it has also been argued that substantial R&D expenditures do not necessarily ensure high growth for a firm (Scherer 1999). In other words, stimulating R&D and innovation does not necessarily lead to observable growth in the short term, particularly not in terms of employment. In fact, technological advancement is often even labour saving (Harrison et al. 2014;Bogliacino et al. 2013;Dachs and Peters 2014). Therefore, a company might grow considerably in terms of sales without any employment growth.
Accordingly, it might be reasonable to expect, for YICs, a higher sales growth than employment growth owing to the successful introduction of new products into the market (Czarnitzki and Delanote 2013). To test if this is the case, we have run a separate quantile regression for sales growth (see Table 3). In contrast to the results obtained for employment growth, the control inserted for YICs is now significant (although at 10 %) only for high-growth firms (and not any more for those firms at 40 As previously discussed, age and size enter the equation with negative significant coefficients and the aforementioned dummy for innovative firms enters the equation with a positive sign. 41 However, the study by Czarnitzki and Delanote (2013) has a drawback that limits the possibility to generalise their results: it uses Flemish firm-level data over the period 2001-2008 obtained by combining three different datasets and, out of 3537 Flemish firms, only 1.16 % were observed over all of the years of the analysis. 42 Concurrently, however, these findings are in line with previous empirical evidence on Spanish firms, which, in the case of YICs, suggests low persistent and quite erratic innovation behaviour (García-Quevedo et al. 2014). or around the median) and enters the equation for the extreme positive quantile with a positive sign. This result suggests that young, small and R&D-intensive companies are likely to experience higher/faster sales growth 43 (although, ceteris paribus, they do not show high employment growth) than the rest of the firms in the sample. In addition, among declining firms, no growth autocorrelation is observed for innovative firms (which is also confirmed for low-and high-growth firms), while, for non-innovative firms, a negative first-order autocorrelation is found, which is non-significant if growing and high-growth firms are considered. In other words, being a non-innovative firm increases the pace of decline (accelerates downturn), but it does not lead to a positive growth pattern. Concurrently, we lose the significance on the positive autocorrelation coefficient for innovative highgrowth firms.

Conclusions
This study has analysed the organic employment growth patterns of an original panel of Spanish firms over the period 2002-2009. It investigated whether or not, ceteris paribus, employment growth and its persistence (if any) differs between innovative and non-innovative firms. Notwithstanding the balanced nature of the dataset used, which prevented us from considering the growth behaviour of new firms and that of those exiting the market, the empirical findings presented in the study confirm the important role that innovative companies can play in helping to accelerate job creation in the economy and, in the long run, to ensure endurable levels of employment. Empirical evidence points to the existence of large asymmetries in downsizing and upsizing processes, and confirmed that the econometric approach that was chosen for this study is particularly appropriate in situations in which outliers' growth behaviour (high-growth or declining firms) is of interest in itself.
This result is less rhetorical than it seems. If it is straightforward to expect that innovative firms grow more than non-innovative firms in terms of sales during a certain transitory period after the introduction into the market of a new product, there is little reason to anticipate a direct link between innovation and employment growth.
In addition, our results suggest that, ceteris paribus, innovative firms not only are growing more than their non-innovative counterparts, but also are more likely to experience high employment growth episodes than their non-innovative counterparts. Interestingly, among high-growth firms, being small and young and investing heavily in R&D (i.e. being a YIC) does not necessarily imply higher than average employment growth rates, 44 whereas it does in the case of sales growth. This is a 43 As pointed out by Ciriaci et al. (2014), larger firms are less likely to experience decline: the bigger a firm is the lower the rate at which sales growth is declining. Smaller firms, in turn, are more likely to experience positive and high sales growth. A positive effect of size on lower quantiles (declining firms) combined with a negative effect on upper quantiles (high-growth firms) indicates that larger firms experience a lower variance in growth rates, that is they are less likely to experience either fast decline or fast growth in the following years. 44 However, during downturn periods, YICs decline slower than the average. Eurasian Bus Rev (2016) 6:189-213 205 result that, in line with the previous argument, can be interpreted by considering the fact that technological advancement is often labour saving and that, in the short run, young firms that have very recently entered a market and have invested relatively strongly in R&D (the returns of which may take time to show) may decide to 'wait' before hiring additional employees. Therefore, an advantage is more likely to be gained in terms of sales growth than in terms of employment growth, as confirmed by our empirical evidence. In addition, smaller or younger firms that have not invested heavily in R&D grow more than the average, which suggests that they might be more flexible and quicker to react to economic opportunities and/or that, given their sub-optimal size and the higher risk of early failure (owing to diseconomies of scale), they must grow to survive and reach a minimum efficient scale. Concurrently, larger and older firms-which may have already reached their minimum efficient scale of production and therefore do not need to grow fast or at all-have a bigger 'buffer' in times of decline, probably because they typically have a more diversified portfolio and the ability to benefit from economies of scale. Furthermore, and most importantly, the empirical evidence showed that highgrowth innovative firms are able to reiterate a high employment growth episode over time; that is they are able to create more and persistent jobs than noninnovative high-growth firms, which instead are largely unable to keep this pace of job creation. Innovative firms seem, therefore, to be able to react quickly to positive shocks and sustain their high growth performance, probably because of their higher flexibility and lower adjustment costs.
These outcomes are a strong foundation on which policies seeking to stimulate R&D and innovation activities and those aiming to enhance job creation in Europe can be based. This highlights the need for a stronger integration of these two policy fields, the 'individual' effects of which (as a result of targeting them separately) might indeed be mutually reinforcing. Our findings support policy interventions and instruments that target the growth of innovative companies in Europe and those that are aiming to increase the number of companies that undertake innovative activities to remain competitive.