The greener the better? Job creation effects of environmentally-friendly technological change

1. Introduction

The link between employment and innovation has been extensively investigated in the economic literature; however, the significance and direction of such relation are still among the most controversial topics in the economic and political debate. The past two decades have seen the emergence of novel forms of innovation due to the growing concerns regarding the environmental sustainability of the current production settings. Environment-related innovation, also called green or eco-innovation, has become a relevant phenomenon attracting the attention of scholars and policy makers for its potential to create new market opportunities, triggering a virtuous cycle of growth and employment. The reasons why eco-innovation should be beneficial for firms and for the economy are several. The key motivation lies in the potential exploitation of the opportunities associated to a new, growing market; the market for eco-friendly goods and services, in fact, represents nowadays around 2.5% of the European Union (EU) GDP and is expected to triple by 2030 (EU, 2011). Second, by inducing a more efficient and responsible use of resources, eco-innovation may improve the competitiveness of firms, and altogether their quality. Finally, since Europe is currently the market leader for environmental products, investing in environmental technologies and resource efficiency may help to create a long-lasting comparative advantage for European firms. Environmental protection is in fact a consolidate policy strategy in Europe since the very creation of the EU. ¹ . Within the framework of the Lisbon Strategy first, and Europe 2020 ² after, EU policies aim at obtaining a “smart, sustainable, inclusive growth” with greater coordination of national and European policies.

Nonetheless, at firm level, the net effect of environmental innovation is still unclear and it has yet to be determined whether it may be seen as an opportunity to penetrate new markets, or as a burden that may impair competitiveness and destroy jobs. Furthermore, it is debatable whether environmental innovation may potentially yield different effects at firm level with respect to generic innovation, justifying dedicated streams of literature and policy attention. The lack of conclusive evidence, the increasing concerns regarding the levels of unemployment, and the risk of declining firms’ competitiveness in Europe—particularly in Italy—and the strong policy attention devoted to the transition toward cleaner technologies call for a greater effort in investigating the relation between environment-related innovation and employment. Understanding their link at firm level is crucial in order to predict how the market will adjust to the increasing relevance of the green economy and whether the benefits may outpace the costs associated to the shift in the dominant technological paradigms. A better assessment of this aspect is as well important to implement, if necessary, an effective environmental innovation policy in the future. Finally, this kind of analysis may offer more generalizable insights on the challenges and opportunities associated to changes in technological trajectories in industrialized economies.

This contributes to the existing debate, through a careful investigation of whether technological change, broadly related to sustainability and environmental aspects, has led to positive changes in employment outcomes at firm level in Italy. Using a novel data set that matches firm-level data with patent records, we are able to distinguish between firms performing “green” and/or generic innovation and to assess the causal effect of environment-related technological change on employment growth, after controlling for firms’ attitude toward generic innovation. ³ This is, to the best of our knowledge, one of the few recent works providing fresh evidence on this relevant issue, trying to deal with the limitations that have traditionally affected previous studies on the link between innovation and employment and adding to the much more limited literature on the impact of eco-innovation. In this context our contribution is twofold. First, due to the nature of our data we are able to provide a consistent measure of environment-related innovation, overcoming the limitations of existing analyses that use survey data based on a more discretional definition. Secondly, we develop a reliable identification strategy, thanks to the availability of a longer and consistent time series and the exploitation of a novel instrumental variable approach, allowing the setup of a credible econometric setting and to deal efficiently with the issues associated to the investigation of the causal relationship between job creation and eco-innovation.

We find that environmental innovation positively and significantly affects job creation to a greater extent than generic innovation. This result holds with respect to several robustness checks ranging from measurement to endogeneity concerns and it is robust also when controlling for cost differences between green and generic innovation as measured with the methodology popularized by Harhoff and Thoma (2009) . This further suggests the existence of a positive net effect in terms of jobs creation that persists also when differences in the amount of innovative inputs between green and generic innovation are taken into account.

The article is structured as follows: section 2 offers a review of the literature on the relationship between employment and technological change, underlying theoretical rationale and available empirical evidence on their link and trying to provide some additional insights with respect to the specific case of environmental-related innovation. Section 3 describes the methodology and the main estimation challenges, while section 4 presents the data used for the analysis. Section 5 shows and discusses our results and section 6 concludes.

2. Literature review

The literature on the impact of eco-innovation on labor market outcomes is nested on a large number of contributions looking at the link between technological innovation and employment. Despite the impressive research efforts on the topic, there is however no wide consensus on the direction and magnitude of the abovementioned relation. The emergence of heterogeneous results has been often justified in the light of several dimensions.

A key aspect regards the typology of innovation under analysis with particular respect to the distinction between process and product innovation and to their different impact on employment (among others Pianta, 2005 ; Hall et al. , 2008 ; Harrison et al. , 2014 ). Process innovation has been generally associated to a labor-saving impact causing employment reduction, the so called displacement effect, while product innovation has been linked to employment-stimulating outcomes based of virtuous cycles on increasing sales and revenues, the so-called compensation effect.

A second-order dimension refers to the scope of the analysis, which may concern either the firm or the aggregate level. Firm-level analyses have been generally characterized by a “positive bias” ( Vivarelli, 2014 , see also Chennels and van Reenen, 2002 for a survey). Most firm-level empirical studies in fact find a positive relationship between innovation and employment growth with a general consensus in the case of product innovation and less conclusive remarks for process innovation ( König et al. , 1995 ; Van Reenen, 1997 ; Garcia et al. , 2004 ; Hall et al. , 2008 ; Harrison et al. , 2014 ). This is partially explained by the limited possibility to fully account for compensating mechanisms operating at broader sectoral and spatial level such as potential detrimental effects linked to displacement.

With few exceptions such as Blechinger et al. (1998) reporting evidence of labor displacement induced by process innovation, and Van Reenen (1997) finding that the impact of process innovations is small and not significant, the majority of existing firm-level analyses supports the existence of a positive, though less immediate, impact on employment also in the case of traditional labor-saving process innovation. Among others, König et al. (1995) , Smolny and Schneeweis (1999) , Smolny (2002) , Lachenmaier and Rottmann (2011) report a positive and significant effect of process innovation on employment growth. Employment effects of process innovation are assumed to affect firms’ productivity lowering the amount of labor input and unit costs. However, in a dynamic perspective, lower prices might lead to higher demand and thus higher production, and consequently have a positive effect on employment. More straightforwardly in the case of product innovation, demand is expected to increase and employment is expected to grow ( Garcia et al. , 2004 , Harrison et al. , 2014 ).

Analyses at aggregate level have looked at the innovation–employment link under a broader perspective, and despite providing a less accurate measure for innovative activities carried out by specific economic actors, they are able to account for broader spatial and sectoral dynamics. Also, in this context however the balance between labor-saving and labor-stimulating effects determined by a (potential) virtuous cycle that generates additional production and employment is not straightforward ( Spiezia and Vivarelli, 2002 ; Gagliardi, 2014 ). Simonetti et al. (2000) and Tancioni and Simonetti (2002) found no univocal effect of technological change on employment, while Bogliacino and Vivarelli (2012) focusing on 25 European countries over the period 1996–2005 find that technological change is positively correlated to employment growth. More recently, aggregate studies at country or sectoral level have exploited information on skills heterogeneity to tackle the emergence of heterogeneous results. Relevant contributions (among others, Acemoglu, 2002 and Goldin and Katz, 2007 ) have documented the skill bias nature of technological change, arguing about its positive impact for the employment perspectives of high-skilled individuals and its negative correlation with employment outcomes for low-skilled people.

Beyond the large literature looking at the link between technological change and employment, recent studies have also focused on the emergence of specific technological paradigms. For instance, the information and communication technology (ICT) revolution has stimulated increasing research effort with the aim of disentangling the impact of these technologies on job market outcomes. While early works have documented the stylized facts associated to ICT ( Rosenberg, 1976 ; Dosi, 1982 ; Freeman et al. , 1982 ; David, 1985 ; Dosi et al. , 1988 ), more recent studies have investigated the extent to which these technologies are associated to distinctive trends in the job market. At firm level, ICT technologies have been associated to the so-called “skill biased organisational change” hypothesis ( Bresnahan et al. , 2002 ; Piva et al. , 2005 ; Giuri et al. , 2008 ). These studies suggest IT technologies are complementary to novel workplace organization, including more decentralized decision-making and more self-managing teams, which favor skilled workers. In this context, “IT-enabled organizational change is important components of the skill-biased technical change” ( Bresnahan et al. , 2002 ). At aggregate level, studies looking at the impact of ICT technologies have documented the progressive polarization of the workforce with middle-skilled occupations that have declined in respect to both higher- and low-skilled occupations ( Spitz-Oener 2006 ; Goos and Manning 2007 ; Goos et al. , 2009 ; Michaels et al. , 2014 ; Autor and Dorn 2013 ).

The rationale behind this evidence lies in the theoretical contribution of Autor et al. ( 2003 ), which offers a more nuanced interpretation of the impact of technology in general (and computers in particular) on the labor market outcomes. In this context, computers are supposed to replace labor in routine tasks (tasks that can be expressed in step-by-step procedures or rules) and to complement (i.e. increase the productivity of) nonroutine tasks.

Surprisingly, given the increasing relevance in terms of policy attention and financial resources, other types of emerging technologies such as environment-related innovation has received less attention. The OECD has recently suggested that investing in green activities has significant job creation potential ( OECD, 2011 ). Alternative studies have claimed that the employment benefit of green technologies has been overestimated, arguing that environmental policies, as initiatives aimed at providing incentives in environmental innovation, may have much less attractive labor market consequences (e.g. Michaels and Murphy, 2009 ; Morriss et al. , 2009 ; Hughes, 2011 ). Becker and Shadbegian (2008) examine the environmental product manufactures using a 1995 survey data for the United States, finding that they did not perform differently in terms of wage, employment, output, and exports than non-environmental product plants.

Within this context, studies on the relationship between environmental innovation and employment represent a more recent and relatively less developed strand of research. This literature is based on the seminal work by Pfeiffer and Rennings (2001) and Rennings and Zwick (2002) . Pfeiffer and Rennings (2001) , in line with the conventional literature on the link between employment and technological change, argue that the effects of environment-related innovations on employment depend on the types of innovation activities performed. Product innovation has been found to generate positive direct effect on employment, while the effects of process innovation are more ambiguous. Employment effects have also been found to be unevenly distributed across skills, with strong negative effects of environmental innovations on low-skill intensive industries and potentially positive effects on other industries.

Rennings and Zwick (2002) , analyzing a sample of environmental-innovative firms for five EU countries in both manufacturing and service sectors, find that in most cases employment does not change as a consequence of eco-innovation. The evidence is stronger for manufacturing than services but results are generally at odds with the traditional skill biased hypothesis associated to technological change.

Rennings et al. (2004) show that environmental innovations in both products and services lead to positive outcomes in terms of employment (except for end-of-pipe innovation) and this finding has been recently confirmed by Horbach (2010) , documenting a positive impact on employment of environment-related innovations for a sample of German firms. More interestingly, he finds a higher impact of eco-innovation with respect to generic innovation on employment.

On the other hand, Cainelli et al. (2011) find a negative link between environmental innovation and growth in employment and turnover in the short term, analyzing the Community Innovation Survey (CIS) sample of Italian firms, while Horbach and Rennings (2013) , using data from the CIS 2008, document heterogeneous results distinguishing between different types of environmental technologies, such as process and product innovation, and material saving, energy savings, air emissions abatement, or recycling.

Licht and Peters (2013) survey the literature on the link between environmental innovation and employment and empirically test such link exploiting CISs data for 16 European countries, distinguishing between product and process innovation. They find a positive and significant effect on employment growth of product innovations, but no substantial difference between environmental and non-environmental innovation. According to their results, process innovation provides instead a little contribution in terms of employment growth.

Although insightful and besides the emergence of often conflicting results, all these studies have two main limitations. Firstly, it is not clear whether and through which channels environmental technologies affect employment differently than generic innovation. Also related to this issue, little insight is offered on the potentially different cost of carrying out environmental or non-environmental innovation.

Secondly, data used for the analyses are generally based on innovation surveys and as such are heavily influenced by the structure of the questionnaire. As remarked in Horbach and Rennings (2013) , the CIS questionnaire defines innovation as the development or the adoption of a “new or significantly improved product, process, organizational method or marketing methods that creates environmental benefits compared to alternatives” (p. 160). A measure of environmental innovation relying on this definition may be highly discretional and suffer from measurement problems and response bias. Furthermore, the time coverage of the CIS is limited and thus the analysis may not be able to capture medium-long term effects of eco-innovation.

Table 1 summarizes the relevant literature described in the current section.

2.1 Conceptual framework

In testing for the direct effect of environment-related innovation on employment, we expect green technologies to have a positive effect on firm-level employment. Our hypothesis is grounded in the well-established theory of technology life cycle phases ( Consoli and Vona, 2015 ). As noted by Agarwal and Gort (2002) , according to its life cycle, a market bears a different range of challenges and opportunities. Both theory and empirical evidence support the view that technological opportunities greatly change from infancy to the maturity of a market: in the early years, the scope for innovation is at its highest, but as a product or a market becomes more mature, technological opportunities decrease. Frankhauser et al. (2008) detects a similar trail for the green sector, adopting a time horizon approach to predict the evolutionary path of the green sector, mostly related to its ability to generate new jobs. In the short run, a direct employment effect generates job losses in sectors adversely affected by climate change policies and new jobs in sectors taking advantage from these initiatives. In the long run, the positive effect outweighs the negative impact since innovation and the development of new technologies create opportunities for investment and growth (dynamic effect of climate policy). However, learning by doing usually helps to increase labor productivity, so the impact on jobs of introducing a new technology may be attenuated over time. In this view, environmental innovation is intended as an expanding sector yielding a potentially dramatic change in the production patterns. As such, it can be used as a blueprint for any other forms of innovation that possess similar attributes, thus providing relevant insights that can be generalized to other types of technological advances.

Nonetheless, the green sector has some specificities that are unprecedented in other technological sectors and that provide the rationale for dedicated analysis on the effects of green innovation. First of all, differently from other sectors, green innovation is linked to the reduction of negative externalities, in terms of less environmental degradation, for the society as a whole. Although there are still some skeptical economists, the majority of them nowadays recognize the need for some form of intervention to redirect production toward cleaner forms. As Acemoglu et al. (2012) note, in fact, clean and dirty inputs are highly substitutable, meaning that without immediate and decisive intervention, firms would not change their production patterns. Furthermore, as any other forms of innovation, environmental innovation is characterized by the presence of high switching costs (a typical example is renewable technologies), which may impair the attractiveness of the sector for investors. As a consequence, many countries and the EU more than others, have opted for various degrees of policy intervention to reduce environmental degradation. This implies that environmental technologies have exploded mainly because of a clear, long-term policy agenda that has contributed to a reduction in the level of uncertainty about future return from innovation and helped the coordination among actors.

A second specificity refers to the importance of knowledge spillovers linked to environmental technologies. As shown by Dechezlepretre et al. (2013) , green technologies are characterized by substantially larger knowledge spillovers than substitute brown technologies and other recent radical technology fields such as robotics, biotechnology, nanotechnology, and 3D printing and in line with information technologies (ITs). While the presence of knowledge spillovers reduces the appropriability of the returns of technological innovations, it also suggests that environmental technologies enable the generation of a variety of complementary innovations, as they tend to be more general than other technology fields. As suggested by Haupt et al. (2007) , a disproportionately high number of citations is also a signal that the technology field is in its early stage of the technology lifecycle.

A third element that explains the increased interest of firms toward the green sector is the growing attention paid by consumers to the environmental sustainability of the products they purchase. Firms are increasingly willing to switch their product mixes and production processes or inputs, in order to exploit new market opportunities. While this latter dimension, namely the demand pull effect, is common to other technological advances developed in the past few decades, the policy-driven lever, which also responds to the variety of potential positive spillovers associated with investments in green technologies, remains specific of green innovation.

3. Methodology

The investigation of the impact of green technological change on employment growth using firm-level data brings along a number of methodological challenges, ranging from measurement issues to model specification and endogeneity concerns. Each of them will be carefully addressed in this section to support the reliability of our findings.

3.1 Measurement issues

Measuring technological change at firm level is not an easy task. Recent studies have mainly exploited the availability of micro-data coming from innovation surveys. Most notably, the work of Harrison et al. (2014) refers to the third CIS to recover information on employment and sales between 1998 and 2000 and whether the firm has introduced process and product innovations during the period.

In a similar vein Hall et al. (2008) , working on Italian manufacturing firms, used data coming from the Mediocredito-Capitalia surveys on sales per employee, growth rates of employment, and sales of old and new products for the period 1995–2003. Both databases allow to recover information on innovation activities carried out during the period under analysis (for both product and process innovation) and to relate them with changes in employment at firm level.

CIS data have been recently used also to look at the impact of environment-related innovation ( Cainelli et al. , 2011 ; Horbach and Rennings, 2013 ; Licht and Peters, 2013 ) exploiting the availability of a dedicated section of the survey.

Besides the traditional problems associated to survey data in particular with respect to the credibility of the innovation measure, the key limitation in the context of the investigation of the link between technological change and employment outcomes is the availability of short time series (generally 2–3 years). The impact of technological change on firms’ employment profiles is unlikely to be fully recoverable in a limited time span since the potential virtuous cycles of increasing sales, production, and employment need time to materialize.

To provide a more reliable investigation on the medium-long term impact of technological change in environment-related fields, we adopted an alternative data source that has been extensively used in the literature on technological change and employment, starting from the work of Van Reenen (1997) analyzing manufacturing firms in Britain. The empirical investigation has been based on a novel data set matching Italian firm-level information with records coming from the European Patent Office (EPO) and providing the possibility to attribute to each firm all inventions patented during the period 2001–2008. Patents, interpreted as “stock of blueprint technologies that can be actualized in the form of an innovation outcome when economic conditions are favourable” ( Van Reenen, 1997 : 263), allow to account for the technological knowledge gathered by each firm over time. In this context, the number of patented inventions during the period under analysis represents the recent stock of technological knowledge that each firm managed to accumulate. Furthermore and particularly relevant for this analysis, following the classification provided by the OECD (ENV-TECH), the sub-sample of environment-related patented inventions may be extrapolated from the full sample of patents, allowing to test for the existence of a specific effect on job creation coming from green technologies.

A number of preliminary considerations need to be highlighted with respect to the choice of patent data as proxy for technological change. First of all, the focus on patents implies that the paper investigates the link between employment and innovation with respect to firms who are the creator of new technologies rather than to the full sample of organizations that may use green technologies/practices in their production settings despite being not the owner of the invention. In principle, there are two channels that can influence employment in response to innovations: a direct effect on innovating firms, through the exploitation of market opportunities and higher investment, and an indirect effect on users of innovations, through process innovation and the demand for new skilled labor in order to adapt to the use of new technologies. As highlighted in the literature, these two effects may yield different implications on employment. In particular, while the direct effect is usually associated with an increase in employment at firm level, the indirect effect on employment is ambiguous, as process innovation may be both labor saving or call for new, high-skilled labor in order to be properly implemented. Measuring innovation with patents allows to predominantly capture the direct effect of (eco-)innovation while provides very little information about its potential indirect effect.

Secondly, although being a widely used output measure, patents are likely to be skewed toward innovation in large firms and technologically intensive sectors. This may provide a significantly different perspective of analysis with respect to data coming from innovation surveys (especially the CIS), relying to a relevant proportion of small and medium enterprises and built in order to provide a balanced sample in terms of sector of activity. Furthermore patent data are notably more representative of product rather than process innovation, preventing from the possibility to address the two dimensions independently. With respect to this latter aspect it is important to bear in mind that our expectation on the sign of the relation between technological change, measured by means of patent data, and employment growth is strongly driven by previous findings. There is a general consensus on the existence of a positive link between product innovation and changes in employment ( Peters, 2004 ; Hall et al. , 2008 ), while no clear evidence has been provided on the effect of process innovation. Given the nature of our proxy for technological change we expect a positive contribution to employment growth. Nonetheless the existence of a specific impact associated to environmental technologies, that is the focal object of our analysis, is less straightforward to assess, as well as still understudied within the existing literature.

Finally, it is widely acknowledged that patents differ substantially in their economic and technological value (i.e. quality), with few patents that are very valuable and many patents with little or no value ( Schankerman and Pakes, 1986 ). The failure to account for the heterogeneity in the value across patents may give rise to substantial measurement errors and biased estimates. For this reason, we try to control also for these differences by employing a series of indicators of patent quality that will be described in section 4.

3.2 Model specification

The key interest of this article lies in the investigation of the potential job creation effect of green technological change. This implies accounting for this dimension while controlling for both firm-level characteristics, which may increase the likelihood of innovation, as well as firms’ capability to develop other kinds of innovative activities that cannot be classified as environmentally friendly.

Despite its simplicity, the above specification allows to test the hypothesis regarding the specific impact that green technologies may have in terms of job creation. In the evaluation of the reliability of our findings, it is important to highlight the possibility to control for detailed measures of firms’ financial performance and additional information that are not common in alternative studies exploiting data from innovation surveys.

3.3 Endogeneity issues

The main concern within our estimation framework is the potential endogeneity of technological change. The characteristics of our data and the variability in the temporal window for which different firms are present in our database (due to both lacking information, especially for 2001, and firms’ exit), prevent from the possibility to estimate the equation of interest in differences (i.e. controlling for time invariant firm fixed effects). The exploitation of the balanced dimension of our panel would in fact come at a great cost in terms of number of observations. Furthermore due to the characteristics of our measure of technological change we believe a pure difference in the number of patents between 2001 and 2008 would be a misleading proxy for firm technological trajectories, leading to a poor exploitation of the information available with respect to the stock of knowledge accumulated over time.

In the evaluation of the estimation strategy adopted, it has to be borne in mind that the decision to invest in innovation-enhancing technologies, bringing to the emergence of technological change as measured by the number of patents by firm, is generally taken in advance based on firm’s specific productivity level and economic performance. While the role of initial firm-level conditions, determining the incentives to carry out technological investments, is factored out by controls included in the specification, it is still possible that unobserved productivity shocks over the period taken into account may shift firms’ incentives to perform innovation-enhancing activities ( Chennels and Van Reenen, 1999 ). This is a particularly relevant issue if we assume that investment decisions and the subsequent patenting output take place within the same time window or if we allow for the possibility of any anticipation effects of future technological shocks at firm level.

Two considerations need to be taken into account with respect to the above concerns. First, it is reasonable to assume that investment decisions associated to inventions patented over the period 2001–2008 have been taken based on firms’ conditions pre-2001. ⁶ Secondly, it is also plausible that anticipation effects of future firm technological shocks are unlikely to affect substantially the decision to carry out technological investments ( Harrison et al. , 2014 ). Notwithstanding these clarifications, our estimation is still at risk of both simultaneity and reverse causality bias. If the investment decision and the realization of the innovation output take place during the time span 2001–2008 (especially for firms observed for a longer period), we may be unable to disentangle the sign of the causality (i.e. firms may shift toward different technologies in response to changes in the nature and typology of available workers). Furthermore we cannot exclude a priori the possibility that firms are stimulated to engage in innovation by the anticipation of future technological shocks, implying that they may decide to change their employment profile (e.g. hiring R&D personnel working on the development of such innovations) due to, for example, expected future increases in labor productivity.

The existing literature has tried to address the abovementioned concerns associated to the endogeneity of technological change relying mainly on Instrumental Variables (IV) techniques and exploiting a range of possibilities. Harrison et al. (2014) used information on the increased range of goods and services reported in the CIS questionnaire. Their identification strategy builds on the structure of the CIS questionnaire disentangling the reasons for the introduction of innovation. Due to the presence of two related questions referring to “increased market share” and “improved quality in goods and services” as alternative motivations to engage in technological innovation, the authors suggest that the “increased range of goods and services” variable must be interpreted as a “measure of the extent to which firm’s innovation is associated with an increase in demand for reasons other than changes in product prices and quality” ( Harrison et al. , 2014 ). As a result, they expect this instrument to be uncorrelated with both changes in the price of new products compared to old products and with productivity shocks. Despite the appealing rationale this identification strategy is questionable. Data exploited to construct the relevant instrument come from the same CIS wave reporting information on both innovation activities and motivations behind the innovation activities carried out over the period 1998–2000, as well as data on changes in employment during the same period. The risk of substantial simultaneity bias cannot be fully ruled out.

Hall et al. (2008) exploited data on R&D expenditures in the last year of the 3-year survey period, the same measure lagged 1 year (in the middle year of the survey period), the R&D employment intensity in the last year of the survey period, and a dummy variable for whether the firm assigned high or medium importance to developing a new product as the goal of its investment. Among this set of instruments, those taking advantage from information on R&D expenditures and employment intensity in the last year of the survey try to deal with the potential simultaneity bias referring to the end of their time window, but the lag is likely to be too limited to rule out any doubts regarding the existence of a significant time trend driving firms’ investment decision.

In order to address the endogeneity concern for our measures of technological change, for both environmental and non-environmental technologies we adopt a novel identification approach taking advantage from the strategy popularized by Ellison et al. (2010) and Haskel et al. (2007) , instrumenting the geographical concentration of economic activities in the United States with that in the UK and FDI inflows in the UK with those in the United States, respectively. Exploiting data on EPO patent applications count (for both non-environmental and environmental patents) for the period 1996–2004 ⁷ filed by companies in Western Europe ⁸ in the same sector (four-digit NACE rev. 2), for the same size class (more or less than 250 employees in median value) and the same age class (more or less than 10 years), ⁹ we instrument our proxies of technological change with comparable measures for a similar sample of European firms. The instrument relies on the idea that international technological trends (among technologically coherent countries) affect, for homogenous categories of firms (in terms of size, sector of activity, and age), the probability to engage in technological innovation and its intensity independently on shifts in firms’ specific incentives. Coherently with previous contributions applying a similar strategy the exclusion restrictions for our IV implies that innovations developed by Western European firms in the same four-digit NACE sector do not directly impact on the evolution of employment in Italian firms. In other words, we assume that these technological innovations are not sufficiently global in scope to influence domestic firms directly. The focus on technology producers only and the legal rights associated with the patenting procedure reinforce the credibility of the abovementioned assumptions.

The instrument is expected to be significantly and positively correlated with the regressor of interest, but uncorrelated with unobserved firm’s productivity shocks.

4. Data

Our sample consists of 4507 Italian manufacturing firms. We selected these firms from a panel of 49,590 manufacturing firms in the AIDA (Bureau van Dijk) database based on the criterion that they should have applied for at least one patent at the EPO between 1977 and 2008. The link between firms in AIDA and applicants at the EPO is described in Lotti and Marin (2013) . For each firm, we know the whole record of patent applications at the EPO for the period 1977–2008.

The focus on patenting companies (either in the considered period or before the period) allows relying on a homogeneous population of potentially innovative firms for which patenting is (or has been) a relevant tool to protect their inventions/innovations. This criterion may lead to a selection bias ¹⁰ but it is also likely to substantially reduce the unobserved heterogeneity in patent propensity across firms. Given the object of the investigation (i.e. the potential specific effect on job creation attributable to environment-related technological change with respect to general innovation) the latter aspect is considered far more relevant than the former for the reliability of our estimation strategy.

We retrieved balance sheet and income statement information together with employees’ headcount for each firm in our sample for the period 2001–2008. Real turnover (in euro) has been deflated by means of sector-specific deflators for gross output (NACE rev. 2, two-digit, reference year 2000). ROI is the ratio between the Earnings Before Interest and Taxes and total assets, both in nominal terms. We also use the variation of cost for employees (labor compensation) as an alternative way of measuring employment growth (in real terms, deflated with sector-specific deflators of value added). We obtained information on location (province—NUTS3), sector (NACE rev. 2, four-digit), and age of the firm from the AIDA database. In our baseline specification, we aggregate firms by macro-region (four NUTS1 regions) and two-digit NACE sector. We excluded outliers based on having the value of the outcome variable three standard deviations greater than the third quartile or smaller than the first quartile (severe outliers). ¹¹

Environmental patents have been identified by means of the definition of environmentally sound technologies prepared by the OECD ¹² based on a list of relevant IPC and ECLA ¹³ classes. They include: general environmental management (pollution abatement, waste management, soil remediation, environmental monitoring), energy generation from renewable and non-fossil sources, combustion technologies with mitigation potential, technologies specific to climate change mitigation (e.g. CO2 capture and storage), technologies with potential or indirect contribution to emissions mitigation, emission abatement and fuel efficiency in transportation, and energy efficiency in buildings and lighting. The full list of technology fields and IPC/ECLA classes employed to identify environmental patents are reported in  Table A2 of Appendix 2. ¹⁴

Finally, we account for heterogeneity in the value and relevance of patents by employing a series of indicators of patent quality in addition to the raw count of patents. We select six different measures of patent quality from the ones reviewed and defined by Squicciarini et al. (2013) . ¹⁵ Details on the measures of patent quality are reported in Appendix 1. Table 2 reports some descriptive statistics for our variables of interest, while Table 3 shows the distribution of observations and patent applications by sector and initial size. Table 4 reports the average values of our dependent variable by initial size and sector for different categories of patenting outcome during the considered period. Firms with at least one patent application in the period tend to grow, on average, substantially faster (or to shrink more slowly) than those without patents. This evidence is common for all size classes and most sectors, the only exception being sector CD (coke and refined petroleum products). Looking at firms with at least one environmental patent ( Env patent ), we observe an above-average long run growth rate of employment for all size classes, although this evidence is inconsistent for some sectors. The difference in performance for firms with heterogenous patenting behavior is clearly visible in Figure 1 , in which we plot the estimated kernel density of our dependent variable. The distribution of the long run growth of employment for firms with at least one patent in the period is slightly shifted to the right relative to the distribution of firms that did not apply for patent in the same period. Moreover, the distribution of long run growth for firms with at least one environmental patent is further shifted to the right, denoting an above-average growth in employment for firms active in the creation of green technologies. This descriptive evidence suggests a strong positive relationship between general patenting and job creation as well as a substantial premium for firms that are active in the field of environmental technologies.

5. Results

5.1 OLS estimates

Table 5 reports the results of our OLS baseline estimates. In column 1 we include our set of control variables only. Long run employment growth is positively related to firm’s initial profitability (ROI); profitability is expected to stimulate new investments and, consequently, firm growth. The negative (raw) relationship between initial size and growth is a common finding in empirical analyses as well as the negative relationship between firm’s age and firm’s growth. Results are also consistent with respect to the control for patenting history since firms with a longer patenting history tend to grow slower.

In column 2 we add the total count of patents in the considered period. Sign, magnitude, and statistical significance of our controls remain unchanged but we find a strong positive effect of patenting outcome on long run employment growth. Each additional patent results, on average, in an increase of employment of about 0.72%. The positive sign is consistent with most of the existing recent contributions investigating the link between product innovation and employment. As discussed in the previous section, even though our measure of innovation output (patent count) includes both product and process innovations, product innovations tend to be over-represented relative to process innovations ( Arundel and Kabla, 1998 ).

In column 3, we split our measure of overall innovation into “green” innovations ( Env patents (count) ) and other innovations ( Non-env patents (count) ). We find a big and statistically significant effect of “green” innovation on employment growth. Applying for one additional “green” patent results in an average increase of long run employment of about 2.7%, which should be compared to the increase driven by a non-environmental patent of about 0.58%. Despite still positive and significant, the magnitude of the regressor for non-environmental innovation is significantly lower.

All in all, results based on OLS estimates show that environment-related technological change is associated to employment growth in Italian firms. This implies that investments in green technologies are likely to generate a (gross) return in terms of employment growth that is substantially bigger (more than four times) than the return of non-environmental technologies. In interpreting this result it is important to bear in mind two caveats.

First, our baseline results do not account for differences in patents’ quality. If environmental patents differ systematically from others in terms of quality, this may drive substantially our results. Table A4 in the Appendix 2 reports average measures of patent quality for environmental and non-environmental patents. We observe that, with the only exception of patent family size, environmental patents tend to be of higher “quality” than non-environmental patents, especially when considering forward citations (in line with Dechezlepretre et al. , 2013 ) and scope. This difference may partially explain the greater job creation effect of environmental patents providing additional evidence on the rationale behind our empirical results.

Second, our estimation does not control for the cost of different innovations. Indeed, it could be the case that the cost for obtaining green patents is different from the cost for obtaining other patents. If cost differentials are significant, the net effect on employment can be overestimated in particular with respect to the impact of non-environmental innovation. Unfortunately, it is difficult to recover reliable information on the cost of the innovative process; however, we try to shed some light on this dimension by comparing the number of inventors associated with our sample of green and non-green patents. As suggested by Harhoff and Thoma (2009) , the number of inventors needed to obtain a patent is a good proxy for R&D investment, due to the relevance of wages for researchers in overall R&D expenditure. Table A5 in the Appendix 2 shows some descriptive statistics on inventors count for our sample of patents. On average, each patent requires 1.88 inventors, while one environmental patent requires on average 2.1 inventors. The distribution is quite skewed, with more than half of non-environmental patents requiring just one inventor. This evidence suggests a greater “cost” for obtaining an environmental patent with respect to non-environmental innovations. In evaluating this claim it is however important to highlight that the number of inventors per patent is likely to be specific to each technology field. This implies that part of the difference in the number of inventors may be explained by characteristics other than the simple environmental versus non-environmental dichotomy. In Table A6 in the Appendix 2 we investigate the extent to which environmental patents require, on average, more inventors than other patents when controlling for year dummies and technology fields covered by the patent. ¹⁶ Evidence confirms that, on average, environmental patents require more inventors than other patents even after controlling for technology fields. The difference ranges between 0.088 (but not statistically significant) and 0.24 inventors, which corresponds, in percentage terms, to a range going from about 4.7% to 12.8% more inventors than non-environmental patents. When comparing estimated cost differentials and estimated return differentials, however, the gap in net returns between environmental and non-environmental patents remains remarkable, suggesting that, despite requiring a greater innovative effort, eco-innovations are still likely to yield a significant return in terms of employment effects. This evidence reinforces our baseline claim on the additional job creation effect of environmental with respect to other forms of innovation.

5.2 Robustness checks

The reliability of our results is further tested through a number of robustness checks. Our baseline results are confirmed in their sign and significance when focusing on the extensive margin only (column 4 of Table 5 , binary indicator of whether the firm applies for at least one “green” patent), in which the effect remains positive despite the slight reduction in the level of significance. In column 5 of the same table, we include the stock of patents ¹⁷ prior to the initial year, for both non-environmental patents and environmental patents. Past patenting performance affects employment dynamics similarly to current patenting, with both patent stocks having a positive effect on employment growth and the effect being greater for environmental than for non-environmental patents. Indeed the two measures recall different dimensions. Whereas the flow of patents over the period of analysis proxies recent investments in innovation, past stocks provide an indication for the accumulation of knowledge over time. The difference between the two measures, which are traditionally highly correlated since novel innovations are more likely to emerge from cumulative patterns, may be more relevant in the case of environmental technologies. Innovative efforts in this context increased significantly in recent years, and a number of firms have grown in size or entered the market, thanks to the new opportunities associated to the “green economy.” Past stocks may substantially underestimate this dimension, implying a weaker explanatory power for green innovation than for generic innovation. This hypothesis finds some suggestive evidence in terms of R squared that shrinks sensibly when using stock measures (column 5, Table 5 —R squared of 0.127) instead of flow measures (column 3, Table 5 —R squared of 0.136).

Table 6 performs some additional robustness checks. Results tend to be robust to the omissions of control variables (columns 1 and 2) even though the estimated return of non-environmental patents in terms of job creation turns out to be substantially underestimated relative to our baseline results. No substantial difference in the effect of our variables of interest is found when adding more detailed dummy variables (four-digit NACE rev. 2 and NUTS3 in column 3), when assuming a nonlinear relationship between initial size and employment growth (column 4), when using initial size expressed in terms of employees (column 5), and when using the growth rate of total compensation to employees as an alternative measure of employment growth (column 6).

Finally, we compare the return of environmental patents in terms of job creation (or destruction) with the one of other specific fields which have experienced a rapid development in the past decades, such as ICTs. ¹⁸ The specification reported in column 7 highlights that the link between ICT patents and job creation is statistically significant only at the 10% level but the magnitude (1.7% increase in employment for each additional ICT patent) is only slightly smaller than for environmental patents and still well above the job creation potential of other “general” patents. ¹⁹ Therefore the impact of green technology remains positive and significant also when compared with other emerging fields.

Finally, Table 7 reports the results of our preferred specification for different samples based on alternative ways of identifying outliers. We use the whole potential sample of firms (column 1), a sample which excludes both “severe” and “mild” outliers ²⁰ in terms of outcome variable (column 2), samples excluding the top and bottom 1% and 5% of the distribution of employment growth (columns 3 and 4) and a sample excluding influential observations based on Cook’s distance ²¹ (column 5). Finally, column 6 reports the results for the whole sample obtained with robust regression, in which observations are weighted by a measure negatively related to their influence on the results. The estimated coefficients for our variables of interest and main controls remain stable across all different samples as well as in the robust regression, with somewhat weaker results in some cases, suggesting that our results are not driven by the composition of the sample.

5.3 Instrumental variables

Despite the promising stability of our results with respect to a number of robustness checks, a more careful analysis of the potential role of endogeneity concerns is still needed. As acknowledged in section 3, there are a number of considerations that may question the causality between technological change and employment, ranging from simultaneity to reverse causality biases. To deal with them we adopted the identification strategy based on an Instrumental Variable (IV) approach as discussed in section 3.3 ( Table 8 ).