Revisiting the Porter Hypothesis: An Empirical Analysis of Green Innovation for The Netherlands

Transcription

1 2013s-02 Revisiting the Porter Hypothesis: An Empirical Analysis of Green Innovation for The Netherlans George van Leeuwen, Pierre Mohnen Série Scientifique Scientific Series Montréal Janvier George van Leeuwen, Pierre Mohnen. Tous roits réservés. All rights reserve. Reprouction partielle permise avec citation u ocument source, incluant la notice. Short sections may be quote without explicit permission, if full creit, incluing notice, is given to the source.

2 CIRANO Le CIRANO est un organisme sans but lucratif constitué en vertu e la Loi es compagnies u Québec. Le financement e son infrastructure et e ses activités e recherche provient es cotisations e ses organisations-membres, une subvention infrastructure u Ministère u Développement économique et régional et e la Recherche, e même que es subventions et manats obtenus par ses équipes e recherche. CIRANO is a private non-profit organization incorporate uner the Québec Companies Act. Its infrastructure an research activities are fune through fees pai by member organizations, an infrastructure grant from the Ministère u Développement économique et régional et e la Recherche, an grants an research manates obtaine by its research teams. Les partenaires u CIRANO Partenaire majeur Ministère u Développement économique, e l Innovation et e l Exportation Partenaires corporatifs Autorité es marchés financiers Banque e éveloppement u Canaa Banque u Canaa Banque Laurentienne u Canaa Banque Nationale u Canaa Banque Scotia Bell Canaa BMO Groupe financier Caisse e épôt et placement u Québec Féération es caisses Desjarins u Québec Financière Sun Life, Québec Gaz Métro Hyro-Québec Inustrie Canaa Investissements PSP Ministère es Finances u Québec Power Corporation u Canaa Rio Tinto Alcan State Street Global Avisors Transat A.T. Ville e Montréal Partenaires universitaires École Polytechnique e Montréal HEC Montréal McGill University Université Concoria Université e Montréal Université e Sherbrooke Université u Québec Université u Québec à Montréal Université Laval Le CIRANO collabore avec e nombreux centres et chaires e recherche universitaires ont on peut consulter la liste sur son site web. Les cahiers e la série scientifique (CS) visent à renre accessibles es résultats e recherche effectuée au CIRANO afin e susciter échanges et commentaires. Ces cahiers sont écrits ans le style es publications scientifiques. Les iées et les opinions émises sont sous l unique responsabilité es auteurs et ne représentent pas nécessairement les positions u CIRANO ou e ses partenaires. This paper presents research carrie out at CIRANO an aims at encouraging iscussion an comment. The observations an viewpoints expresse are the sole responsibility of the authors. They o not necessarily represent positions of CIRANO or its partners. ISSN Partenaire financier

3 Revisiting the Porter Hypothesis: An Empirical Analysis of Green Innovation for the Netherlans * George van Leeuwen, Pierre Mohnen Résumé / Abstract Presque toutes les étues empiriques qui se sont penchées sur l hypothèse e Porter ont utilisé un moèle à forme réuite, en autres termes un moèle qui régresse la prouctivité sur la réglementation environnementale. Notre étue utilise un moèle à forme structurelle qui permet e tester les versions faible et forte e l hypothèse e Porter. Notre moèle innovation verte comporte trois types investissement environnementaux en plus e la R-D non-environnementale, qui ensemble expliquent l occurrence innovations, qui sont au nombre e trois : innovation e prouit, e procéé et éco-innovation. Nous testons la présence e synergie ans l introuction e ces trois types innovation et ans leurs effets sur la prouctivité totale es facteurs. À l aie e onnées e firmes en panel provenant e quatre enquêtes ifférentes, nous estimons l importance es prix e l énergie et es réglementations environnementales sur les investissements verts et les ifférents types innovation. Nos résultats corroborant la version faible mais pas la version forte e l hypothèse e Porter. Almost all empirical research that has attempte to assess the valiity of the Porter Hypothesis has starte from reuce-form moels, e.g. by using single-equation moels for estimating the contribution of environmental regulation (ER) to prouctivity. This paper aresses the Porter Hypothesis within a structural approach that allows us to test what is known in the literature as the weak an the strong version of the Porter hypothesis. Our Green Innovation moel inclues three types of eco investments an non-eco R&D to explain ifferences in the incience of innovation. Besies prouct an process innovations we recognize eco-innovation as a separate type of innovation output. We explicitly moel the potential synergies of introucing the three types of innovations simultaneously an their synergy in affecting total factor prouctivity (TFP) performance. Using a comprehensive panel of firm-level ata built from four surveys we aim to estimate the relative importance of energy price incentives as a market base type of ER an the irect effect of environmental regulation on eco investment an firms ecisions regaring the introuction of several types of innovations. The results of our analysis show a strong corroboration of the weak version of the Porter hypothesis but not of the strong version of the PH, in this case on TFP performance. Keywors: Porter Hypothesis, green innovation, environmental regulation, innovation complementarities, prouctivity. Coes JEL : H23, L5, O32, O38, Q55 * The first author expresses his gratitue for being able to carry out this research as a visiting fellow at UNU- MERIT. The views expresse in this paper are those of the authors an o not necessarily reflect any policy by Statistics Netherlans. We thank Jacques Mairesse an Georg Licht for helpful comments. Statistics Netherlans. UNU-MERIT an Maastricht University, p.mohnen@maastrichtuniversity.nl

4 1. Introuction The relationship between technological change an environmental policy has receive a lot of attention from scholars an policymakers uring the last ecaes. This is partly because the environmental consequences of social an business activity are affecte by the rate an irection of technological change, an also because environmental policy interventions may create new constraints an incentives that may shape the path of future technological evelopment (Jaffe et al., 2003). Environmental technological progress is a very broa phenomenon an every escription of it cannot be more than very incomplete. Some examples concern 1) technologies that reuce pollution at the en-of-pipe, such as scrubbers for use on inustrial smokestacks or catalytic converters for automobiles 2) technologies that increase user value for consumer proucts (e.g. meicines) after introucing new prouction methos, which, at the same time, ecrease the environmental buren of their prouction by using materials that are less harmful for the environment an 3) implementation of technologies that are targete to changes in prouction processes to improve energy efficiency. Policy responses to environmental problems often start from the assertion that the link between overall technological change an (e.g.) climate base environmental policies is merely macro oriente. However, for unerstaning the interaction between environmental policy an technology it also makes sense to go own to the micro level. After all, environmental regulation an public funing of R&D are the first impetus to have more green technologies evelope by iniviual firms. Similar to other types of innovations, the benefits of environmental technological innovations may accrue to society at large rather than to the aopter of these new technologies alone. This market failure relate to innovation in general is pivotal to the numerous iscussions surrouning the so-calle Porter-Hypothesis (PH). It is often argue (see e.g. Wagner, 2003) that one cannot fin a 10-Dollar bill on the groun, because if it was there, someboy else woul alreay have picke it up. This metaphor neglects three things 1) that market forces alone o not provie enough incentives for firms to be engage in green innovation an 2) that green innovation is not very ifferent compare to innovation in general an 3) that policy responses such as environmental regulation have a role to play to bring economic opportunity in line with the environment (see e.g. Jaffe et al., 2002, Desrochers, 2008 an Cerin, 2012). The central issue is the question whether regulation rives innovation. The PH asserts that polluting firms can benefit from environmental policies, arguing that well esigne an stringent environmental regulation (ER) can stimulate innovations, which in turn increase the prouctivity of firms or the prouct value for en users (Porter, 1991; Porter an van er Line, 1995). The message of this hypothesis is that there seems to be no trae-off between economic growth an environmental protection but a win-win situation instea. Environmental regulation woul benefit both society an regulate firms by triggering 1

5 ynamic efficiency of firms an these benefits may partially or fully offset the costs of complying with environmental restrictions. The empirical evience available supporting the Porter Hypothesis seems to be rather scanty an in most cases the PH is rejecte by the ata (see e.g. Wagner (2003), Popp et al. (2010), Ambec an Barla (2006), an Ambec et al. (2011) for an extene review). For the Netherlans the evience seems to be very scarce. This paper tries to she a new light on the PH by using a rich unbalance panel constructe by matching Dutch firm level ata from four surveys an by moeling the complementarities of ecoinnovation with traitional moes of innovation. As argue by Kriegel an Ziesemer (2009), the main problem regaring the empirical testing of the PH, in essence, boils own to having a better unerstaning of the (eco) innovation aoption ecisions of firms. This assertion asks for a structural moeling approach in investing the contribution of energy prices an environmental regulations on green investment an of green investment on innovation an prouctive efficiency. We will embark on this task by aopting a Green CDM (Crépon-Duguet- Mairesse) type of moel for the Netherlans, similar to the Lanoie, Laurent-Lucchetti, Johnstone an Ambec (2011) moel, that allows testing what Jaffe an Palmer (1997) have calle the weak version an the strong version of the PH, referring to the effect of environmental regulations on respectively environmental innovations an economic performance, in this case total factor prouctivity (TFP). Contrary to Cainelli et al. (2010) our moel inclues several types of eco investments an regars eco innovation as a special moe of innovation output. Eco, environmental an green innovation will be interchangeably use, inicating each time an innovation with a lower environmental impact. Likewise eco, environmental an green investment all point to investments aime at reucing the environmental buren of prouction (for more iscussion on the efinition, see Kemp (2011)). Our starting point of investigation of the PH is the impact of energy prices on ifferent types of eco investment. The fact that carbon taxes are a substantial part of gross energy prices makes them a potentially useful instrument for environmental policy aime at improving the energy relate static as well as ynamic efficiency of firms. In particular, this may be the case if such price incentives invoke environmental investment combine with the renewing of the prouction process (so-calle process integrate eco investment). Our empirical moel starts from the estimation of four innovation input equations: two for R&D (eco R&D an other R&D) an two for other types of ecoinvestments, en-of-pipe an process integrate respectively. Subsequently, we use the preictions from these equations for moeling the incience of ifferent types of innovation. At the en, we will estimate a labor prouctivity equation an test for complementarity or substitutability of ifferent innovation strategies in affecting the total factor prouctivity (TFP). 2

6 A novelty of our paper is to try an assess the existence of complementarities (as efine in Milgrom an Roberts, 1990 an 1995) between prouct, process an eco innovations. We istinguish complementarities in the incience of innovation an in their effects on prouctivity performance. To estimate the structural moel we have implemente a proceure propose by Lewbel (2007) for solving the coherency an incompleteness problem when estimating a system of equations with ummy enogenous variables. The plan of the paper is as follows. In section 2 an overview of the literature is given. Section 3 iscusses the moel use in the empirical application an section 4 elaborates on the econometric issues of this research. Thereafter the ata are presente in section 5, followe by a iscussion of the main results in section 6. Section 7 conclues. 2. Literature review One can harly fin any branch of economics that has been concerne with the PH as much as environmental economics. There is a vast boy of literature evote to appraise or peruse the seminal contributions of Porter (1991) an Porter an van er Line (1995). Originating primarily from empirical regularities foun in the analysis of cross-country ifferences in the stringency of environmental regulation an economic performance, the hypothesis has triggere a lot of research both theoretical an empirical in nature. The Porter hypothesis has been criticize, for being merely base on anecotal stories (which in some cases seem to be even erroneous) an for the lack of a soun theoretical basis (see e.g. Palmer et al., 1995, an Cerin, 2006). More recent research attempts to fill the gap between empirics an theory to provie a theoretical unerpinning of the PH. Mohr (2002) argues that it is a feasible outcome if one allows for the possibility of enogenous technical change. More recent theoretical contributions that link the environment to enogenous growth are given in Acemoglu et al. (2012) an Gans (2012). Ambec an Barla (2002) raise the question whether regulation is inee neee for firms to aopt profit-increasing innovations, an pointing amongst others to the 10-Dollar metaphor mentione above. This last criticism is targete to the primitive of the PH stating that firms systematically ignore opportunities for profit increasing innovations an that environmental regulation can motivate firms to capture low hanging fruit offere by environmental challenges to their businesses. Another source is the literature of behavior economics. This literature offers several explanations for unerinvesting in environmental innovation (see e.g. Ambec an Barla, 2006, for examples). Similarly, several theoretical attempts have trie to frame regulation in moels often use for analyzing the interaction between competition an innovation. A recent 3

7 example is given in Constantatos an Hermann (2011). The example concerns the introuction of organic proucts in Pharmaceutical Manufacturing. By avoiing the use of environmentally amaging fertilizers there is less environmental buren as well as more user value create because organic rugs are healthier. In this case a win-win situation between regulation an innovation is not self-evient because there is much scope for conjectural variation, especially if such innovations take place in markets that are characterize by fierce competition an because the environmental quality of the new prouct will be highly correlate with other prouct attributes. The latter argument explains why inertia at the consumer sie of the market may enlarge the risk of investing ue to potential first-mover isavantages. The example illustrates that there are many similarities with the traitional view on the relationship between innovation an competition as a source of unerinvesting in innovation. Moreover, the example also shows that the scope for environmental policies is very broa in principle, but that it is questionable how environmental regulation can offer a solution to the problems involve. This brings us to another stran of research that focuses on the secon primitive of the PH, i.e. the assertion that environmental regulations shoul be well esigne an stringent enough to be successful also from an economic point of view. An assessment of the instruments of environmental regulation an a jugment of their effectiveness can be foun in Wagner (2003). The myria of environmental instruments can be better unerstoo when using a classification or typology. A first elineation is between comman an control type regulation an market base regulation. The instruments that set emission limits an stanars fall into the first class an are often labele en of pipe regulations. Environmental taxes an charges an traable emission permits or certificates are examples of the secon class of instruments. Environmental effectiveness can be efine as the ability to achieve a preefine environmental target. The general view is that this efinition is more appropriate for the first class of instruments. By contrast, the secon class of instruments has a higher economic profile, because they are aime at triggering static an ynamic efficiency an internalizing environmental externalities in an between markets. In particular these instruments play a role in the empirical testing of the premises of the PH. Looking at the empirical evience provie in the literature it can be conclue that the picture is rather mixe. The number of papers an articles that have put the PH to the empirical testing is overwhelming but they o not to lea to a general consensus. Much of this has to o with the ifferent research strategies an the availability of ata. Compare to empirical evience at the macro or inustry level, the number of papers that use firm level ata is rather scarce. Besies that, research is targete at ifferent measures of performance. Cutting through ifferent reviews of empirical work it can be conclue that much research is aime at investigating the impact of environmental regulation on 4

8 prouctivity or prouctive efficiency in a reuce form estimation approach. In many cases this type of research leas to the conclusion that environmental regulation has a negative impact on prouctivity. This conclusion can be easily unerstoo, because regulation forces firms to invest in the environment an this increases prouction costs of firms to comply with the environmental restrictions. There are several issues at stake here. If these investments o not lea to renewing of prouction processes then there is no reason for expecting substantial gains in resource efficiency. A secon issue is relate to measurement. En-of-pipe investments may reuce pollution but this reuction is not accounte for in output. The same capital an other inputs prouce two types of output: ba an goo output an it is harly possible to value the contribution of (reucing) ba output. This raises serious problems when investigating the relation between environmental regulation an (prouctive) efficiency. 1 An interesting solution to circumvent this problem is presente in Domazlicky an Weber (2004). They use volume ata on toxic releases an traitional output measures such as real value ae in a non-parametric analysis to ientify technical change from efficiency change. However, these results also lea to the conclusion that the impact of regulation on total factor prouctivity (TFP) is negative. 2 More interesting for the PH is the research that looks into the impact of environmental regulation on innovation. This type of research seems to be a necessary if not sufficient conition for the PH. Again the evience is scanty an weak in general. Notable examples of this type of research can be foun in several papers an articles of ZEW. In most cases the research uses the ata collecte on innovation an regulation in the Mannheim Innovation Panel that is constructe from several eitions of the German Community Innovation Survey (CIS). The focus of research varies between regulation riven innovation alone (Rennings an Rexhäuser, 2010) to the impact of regulation riven innovation on competitiveness (Rennings an Rammer, 2010), employment ynamics (Horbach an Rennings, 2012) or profitability (Rexhäuser an Rammer, 2011). Support for the PH is provie by concluing that environmental regulation oes not harm competitiveness (Rennings an Rammer, 2010) an that the contribution of regulation inuce innovation to profitability is larger than the contribution of other (more) voluntary innovations (Rexhäuser an Rammer, 2011). Ieally, a thorough empirical testing of the PH requires ata on which types of regulations trigger which innovations at the firm level. The most recent eition of CIS contains new questions on environmental innovation. However, for the Netherlans, this new moule cannot be use to ientify the role of ifferent instruments of environmental regulation properly. The only variable available on regulation in this research concerns firm responses to environmental regulation in general, either existing or anticipat- 1 This problem is well recognize by statisticians an environmental accounting is an important avenue for National Accounts. See Muller et al. (2011) for a recent contribution to this problem. 2 The metho use is the irectional output istance approach evelope by Chung et al. (2007) for constructing the Malmquist-Luneberger inex to ecompose (changes in) TFP. 5

9 e regulations. By contrast, the German CIS allows a istinction between types of environmental regulations. After matching the firm responses with external ata on the age of regulations, Rennings an Rexhäuser (2010) also investigate the long-term impact of ifferent types of regulation on the aoption of environmental innovation. To keep things tractable, they mae a istinction between three types of environmental regulation (ER): a) en-of-pipe regulation, b) circular flow economy regulation an c) climate change base regulations. An important conclusion is that the long-term impact of regulation only triggers innovation that is strongly relate to control type ER. By contrast, the support for PH in terms of increasing static or ynamic efficiency is limite. As convincingly argue by Rennings an Rexhäuser (2010), the contribution of control type ER to ynamic efficiency is expecte to be limite, as, once installe, en-of pipe environmental investment cannot contribute to the innovation process anymore. One can a to this that other types of ecological investment may yiel higher returns to investment. Combining eco investment with renewing of prouction processes seems a better example of innovation relate eco investment. We take this consieration into account in the next section. 3. A Green CDM type empirical moel The empirical moel use in this paper is a moifie version of the so-calle CDM moel. Toay, this moel is the work horse for research on innovation when using firm-level ata (see Crépon, Duguet an Mairesse, 1998). A graphical presentation of our moel is given in Figure 1. 3 The upper part of the figure points to the investment ecision stage. In the traitional CDM moel this concerns the ecision on how much to invest in R&D. In this paper we face another investment ecision problem, i.e. namely whether or not to invest in orer to reuce the environmental buren of the firm s operations. In section 3.1 we will iscuss the specification of the input stage of the moel. The secon block of the moel escribes innovation a separate prouction process with R&D an eco-investment as an input an knowlege creation, in the form of new proucts, new prouction processes an eco-innovation as the outcomes. The thir block examines the link between innovation outputs an prouctivity as a measure of economic performance. We shall now escribe each block in etail. 3 The figure is aapte version of the one presente in Crépon, Duguet an Mairesse (1998). 6

10 3.1 The investment ecisions We consier four investment ecisions: three ecisions concerning types of eco investment an one concerning investment in non-eco R&D. Eco investment (X 1 ) is comprise of 1) eco R&D, 2) eco investment in en-of-pipe facilities, an 3) eco investment that is process integrate, i.e. eco investment combine with the renewal of the main prouction processes of the firm. The investment inputs into non-eco innovation (X 2 ) are efine as gross expenitures in R&D less expenitures in eco R&D. Irrespective of which type of investment is concerne, we assume that firms first ecie whether to invest or not an then choose the investment intensity. Firms will invest in each of these types of investment if the latent value of investing excees some threshol value c. For example, for X 1, this can be expresse as follows: * 1 c DX 1 if DX i1 i Zi 1 1 i1 1 * 1 c DX 0 if DX i1 i Zi 1 1 i1 1., an (1) The vector Z i1 in (1) collects the variables that are assume to etermine the selection of firms that invest in X 1, i.e. the occurrence of having performe an reporte eco invest- 7

11 ment of the type consiere. In a similar way we can moel the ecision concerning non-eco R&D investment (DX i2 ). Moel (1) is extene with a set of equations that moels the level of investment intensity for each of the three types of eco investment an for non-eco R&D. For example, for some type of eco investment, this yiels * X i1 X i1 X i 1 1 e i 1 if DX 1, an i1 X 0 if DX 0. i1 i1 (2) Moels (1) an (2) are applie to each of the four types of innovation investment. Assuming a bivariate normal istribution for an e (in case of a specific type of eco 1 1 investment), an similarly for an e (in case of non-eco R&D investment), the sys- 2 2 tem (1) - (2) is a Tobit II type selection moel (Amemiya,1984). 3.2 Implementing the investment moels Although we are ealing in each case with a general investment problem, one can imagine that the four types of investment are rather istinct. The traitional view is that expenitures on non-eco R&D have a higher economic profile, an that such expenitures are quite ifferent from eco investment (incluing eco R&D) when juge from the point of the strategies of the firms. This is in particular the case if we compare eco investment performe to comply with control type environmental regulation with R&D that is aime at eveloping new goos. Not at least because of the ifficulty of evaluating the output of ba goos or of process integrate eco investment, which, by efinition aims at reucing ba output an increasing resource efficiency, it is impossible to use stanar capital an investment theory to erive formal investment moels. At least, we consier such an exercise beyon the scope of this research. To a lesser extent the same problems also carries over to process integrate eco investment, which, by efinition aims at reucing ba output while increasing resource efficiency. Likewise for investment in non-eco R&D, we miss the necessary price ata to estimate an investment equation that woul be embee in capital an investment theory. For both eco-an non-eco investments we have information on subsiies receive that can be use for explaining ifferences in investment intensities between firms. The R&D subsiies are not the only incentives for eco R&D (an investment). Internalizing environmental externalities can also be achieve via energy prices. As mentione before, energy prices are an interesting instrument of market base ER. A nice feature of the ata is that we can construct firm-specific marginal energy prices to explain ifferences in each type of eco investment. This brings us to the 8

12 specification of the vectors Z an X in (1) an (2) for non-eco R&D innovation investment an for the three types of eco investment respectively Non-eco R&D investment For the selectivity an intensity equations non-eco R&D investment (Z 2 an X 2 respectively), we introuce the same explanatory variables as in Poler et al. (2010). Z 2 inclues the begin-of-perio size (Z 21 ), a ummy inicating whether the firm belongs to an enterprise group (Z 22 ), a ummy to inicate whether firms are involve in innovation cooperation (Z 23 ) or are epenent on foreign markets (Z 24 ), a ummy variable that inicates whether firms respone to ER or not (Z 25 ), a set of inustry ummies (Z 26 ) an a two ummy variables inicating the importance of eman-pull an cost-push objectives for innovation (Z 27 - Z 28 ): Z 2 = {Z 21, Z 22, Z 23, Z 24, Z 25, Z 26, Z 27, Z 28 }. The probability to invest in non-environmental R&D is suppose to increase with size an to be higher for firms that belong to a group, that cooperate for innovation, that operate on foreign markets, that are subject to environmental regulations an that are riven by cost-push or eman-pull consierations. Those same variables are assume to affect the intensity of non-eco R&D, i.e. are collecte in the vector X 2, with the exception of the innovation cooperation ummy (Z 23 ) an with the aition of subsiies receive from local authorities, government boies an the EU (X 21 - X 23 ) an a set of time ummies (X 24 ). Hence we have X 2 = {Z 21, Z 22, Z 24, Z 25, Z 26, X 21, X 22, X 23, X 24 } Eco investment In principle, some of the variables use for moeling other innovation investment coul also be use for the eco investment equations (Z 1 an X 1 ). However, this woul lea to a consierable loss of ata at the outset because of the small overlap of the ata of the four surveys that woul have to be matche (see section 5). To minimize the lack of ata coverage between surveys, we focus on the variables that are collecte in two surveys: the Prouction Statistics survey (PS) an the survey on environmental costs of firms (ECF). The use of ECF is imperative here. It is this survey that collects ata on the three types of eco investment. Combining these with ata on energy costs an volumes collecte in the PS-survey allows us to construct (gross) marginal energy prices (pe gt ) 9

13 for the firms for which ata on eco investment are available. 4 In this way, an after using a measure for the prices of eco investment collecte in the ECF survey (pi t ), we obtain a measure for the relative price of eco investment (relative to energy prices) X ln( pi t ) ln( pe 11 gt ). Furthermore, an to etermine the importance of ER for eco investment, we inclue an environmental regulation (ER) ummy (X 12 ). The moel is complete further by using the logarithm of beginning-of-perio energy cost shares (X 13 ), the logarithm of beginning-of-perio size (X 14 ), a ummy variable that inicates whether firms receive eco subsiies (X 15 ), an a set of inustry an year ummies (X 16 ): X 1 = {X 11,, X 16 }. Finally, we take into account that selectivity may be epenent on the responses to ER (X 12 ), on the (pre-existing) energy intensity (X 13 ), on firm size (X 14 ), on the importance of (pre-existing) environmental levies (X 17 ) an on a set of inustry ummies (X 18 ): Z 1 = { X 12, X 13, X 14, X 17, X 18 }. 3.3 Innovation output The mile part of figure 1 inicates that innovation investment leas to innovation output. We consier three types of innovation output: prouct -, process - an eco innovation. We observe ichotomous variables inicating whether a certain innovation has been aopte or not. The three types of innovation are likely to be interrelate in the sense that the return to a certain type of innovation coul epen on the aoption of the other innovations for reasons of complementarity or substitutability between them. It is well ocumente in the econometric literature (see e.g. Heckman, 1978, Tamer, 2003, Lewbel, 2007) that the estimation of a trivariate probit with enogenous ummy variables raises severe problems of ientification. There can be no solution (in which case the system is sai to be incoherent) or multiple solutions (in which case it is sai to be incomplete). The empirical literature offers several solutions to this problem. In general, these solutions boil own to imposing zero restrictions on the coefficients of some of the binary enogenous explanatory variables or by relying on recursive or triangular sys- 4 For a much smaller sample (obtaine after matching also the Energy Use Survey (ES)) we can ecompose gross marginal energy prices as follows: pe gt pe nt te nt pe nt ( 1 stent), where pe nt is the marginal energy price net of taxes, te nt the marginal energy tax an ste nt the ratio of marginal energy tax over marginal energy prices net of taxes. Constructing this variable is only possible for the intersection of ES an PS ata. Matching this intersection with the CIS an ECF ata leas to a consierable loss of ata. 10

14 tems in which one of the choices is assume to be leaing (see for a iscussion of completeness an coherency section 2 of Tamer (2003)). One way to avoi incoherency an incompleteness is to start from a McFaen (1973) solution by consiering a multinomial choice problem base on a ranom utility moel. This framework has been propose more recently by Lewbel (2007) an aapte by Miravete an Pernías (2006) an Kretschmer, Miravete an Pernías (2012). Let the total utility (in this case profit) be V V y, y, ) ( 1 2 y3 1 ( 1 x1 12y2 13y3 1) y (3) ( 2 x2 21y1 23y3 2) y2 ( 3x3 31y1 32y2 3) y3. The ichotomous variables for the three types of innovation are given by ( i 1,2,3 ). There are in total eight possible combinations of innovation choices yieling respectively the following profit outcomes: y i V( 0,0,0) 0 V( 0,0,1) x V( 0,1,0) x V( 0,1,1) V( 1,0,0) x 2 x2 3x3 ( 23 32) V( 1,0,1) 1 x1 3x3 ( 13 31) V( 1,1,0) 1 x1 2x2 ( 12 21) (3a) (3b) (3c) (3) (3e) (3f) (3g) V( 1,1,1) x x x ( 12 21) ( 13 31) ( 23 32) (3h) The complementarity parameters ij an ji are place in parenthesis because only the sums can be ientifie. 5 If 0 ( 0), the corresponing pair of inno- ij ji 5 Notice that if the ij s are equal to zero, we are in the presence of a trivariate Probit moel. Methos for estimating such moels are reaily available (see Capellari an Jenkins (2003, 2006) an Train (2003)). 11

15 vations are complements (substitutes). The moel is complete because (latent) profitability is specifie for all possible strategies an coherent because every strategy shoul have a latent profit that excees the profits of all other strategies. As pointe out by Lewbel (2007), the ifference with respect to the traitional multinomial choice framework is that we o not have a separate specification for V(1,1,1) such as X 3 3 but instea we use (3h) erive from the moel for the total latent profit function. To our knowlege this moel has not been put to the empirical testing for more than two strategies because of computational ifficulties. We refer to Annex B for a more etaile account of the empirical implementation. The i ' s are ranom errors that are jointly normally istribute. The profitability of pursuing a particular innovation yi epens on the aoption of the other innovation moes, through the complementarity parameters, an on a certain number of variables summarize in x i. It also epens on the non-eco R&D an the three types of eco-investments. It is, inee, interesting to fin out which investments affect which types of innovation an o enter as explanatory variables in the innovation equations. For this reason we use the preictions of the latent variables of the investment equations estimate in the first step: ˆ * ˆ *,, ˆ * X X, the preictions for the ˆ * X13 three types of eco investment, an X, the preiction for non-eco R&D. Furthermore, x i contains C is a vector of control variables (e.g. inustry an time ummies) an especially Dreg, the environmental regulation ummy, which captures whether firms respon to either existing or anticipate ER. Thus, besies the possible influence of ER on (eco) investment, we also account for a more irect effect of ER on innovation output. In essence, the coefficient of Dreg provies a test of the weak version of the PH in the terminology of Jaffe an Palmer (1997). 3.4 Prouction function Finally, we investigate the strong version of the PH by relating ER to labor prouctivity (LP) via the effects of innovation on the TFP component of LP. Again we shall allow the three types of innovations to have synergy effects. But whereas in the previous sub-section the synergy manifeste itself in terms of a latent profit function, this time we let the synergy manifest itself in terms of TFP. To this en we estimate an augmente labor-prouctivity (LP) moel: LP t [ ijki( y1 t i, y2t j, y3 t k)] X 3t 3,( i, j, k{0,1 }), ijk (4) with the first term on the RHS of (4) a short-cut for the set of seven innovation combination ummies for the TFP contribution of innovation an besies the reference 12

16 category I(0,0,0). X 3t is a set of control variables (incluing capital-labor intensities, labor inputs, inustry ummies an incluing the constant term) an 3 is a ranom error term that is assume to be normally istribute. Since the innovation output measures are latent an enogenous, we also have to use instrumental variable estimation methos to obtain an unbiase assessment of their ex-post contribution to TFP (see e.g. Woolrige, 2002). It is possible to test formally the complementarity an substitutability of innovation moes for prouctivity. This can be achieve by applying the super (sub) moularity test propose by Mohnen an Röller (2005). This test is base on super- an sub moularity of the LP equation in terms of the innovation combination ummies. If LP is super moular with respect to a combination of innovation moes, this is evience of complementarity between innovation moes. In the case of sub moularity, the moes are substitutes. For N = 3 we have 8 possible combinations of innovation moes, similar as in Poler et al. (2010). Let I j enote a possible combination of innovation moes, where j = 1,,8. Note that if I j = 1 I k (k# j ) = 0.We will use the shorthan f(ij) to enote the value of the LP equation when Ij = 1. 6 Super moularity is then efine as f ( I j ) f ( Ik ) f ( I j Ik ) f ( I j Ik ) j, k An likewise, sub moularity is efine as f ( I j ) f ( Ik ) f ( I j Ik ) f ( I j Ik ) j, k where is the component wise maximum of Ij an Ik, an the component wise minimum. We o not nee all these inequalities. To test the complementarity between two innovation moes, we only nee to make pairwise comparisons keeping the thir moe constant. In aition, some inequalities are trivial. For example, for Ij = (0,0,0) an Ik = (1,1,0) we have f(0,0,0) + f(1,1,0) < f(1,1,0) + f(0,0,0). Only the combinations where the minimum an maximum operators lea to ifferent combinations than the left-han sies are non-trivial. Thus, combination Ij shoul have at least one element that is smaller than the corresponing element in Ik, an at least one element shoul be bigger (i.e. at least one innovation moe shoul occur in Ij but not in Ik an vice versa). For testing the complementarity between, for example, prouct (y 1 ) an process innovation (y 2 ) in the three-imensional case, we there- 6 The contribution of other variables than innovation ummies cancels out an thus can be exclue from the exposition 13

17 fore have Ij = (0,1,D) an Ik = (1,0,D), with D = {0,1}, an the inequality restrictions are: f ( 0,1,0) f (1,0,0) f (1,1,0) f (0,0,0) f ( 0,1,1) f (1,0,1) f (1,1,1) f (0,0,1) Similar inequality conitions can be erive for the (conitional) pairwise comparisons of the TFP regression coefficients pertaining to other innovation combinations. Furthermore, the inequalities for sub moularity are easily obtaine by replacing <= with >=. Koe an Palm (1986) erive a Wal test-statistic for testing these inequalities for regression coefficients. For N = 3 { 000, 001, 010, 011, 100, 101, 110, 111} is the vector of coefficients on the ummies for innovation moe combinations in the augmente T F P mo e l. We will apply the Koe-Palm (Koe an Palm, 1986) test using the IV estimates ( IV ) of (4). 0. 0, 4. Data We have constructe a comprehensive ataset by linking firm-level ata (manufacturing only) for These ata were source from four surveys: 1) The survey on environmental costs of firms (ECF). The survey covers the years an beyon. This is one of the most important ata sources in this research project. The survey collects (amongst others) ata on environmental current exploitation costs, two types of environmental investment, environmental subsiies an expenses on environmental R&D. Environmental investment other than eco R&D can be broken own into en-of-pipe investment an investment relate to the renewing of prouction processes (so-calle process integrate eco investment ). Because of the fact that this survey only collects ata for manufacturing, our empirical analysis will be restricte to this branch of the economy. 2) The energy use survey (ES), which covers the same perio as the ECF Survey. This survey collects volume ata on energy consumption of ifferent types of energy use an these can be use to construct marginal energy prices at the firm-level after linking with the ata on energy costs collecte in the Prouction Surveys. As natural gas an electricity are the most important energy sources for almost all firms, our measures for energy prices will be erive from firm-level ata for these two energy sources. The ata for the two types of energy can be lumpe into one measure by using weights that reflect the energy content in TJ of each constituent source of energy. Another interesting contribution of this ata source is that it enables the calculation of the carbon-tax compo- 14

18 nent of (gross) marginal energy prices for gas an electricity by using ata on the energy price tariff structures (tariff schemes) an the tax structure of energy prices for these two types of energy use. For a limite number of firms we can thus make a istinction between gross marginal energy tariffs an the carbon tax component of gross marginal energy tariffs for the two types of energy. But this exercise is only possible for the firms sample in the ES-survey. Because of the poor coverage of ES with CIS we face a consierable loss of ata when trying to account for the carbon-tax component of energy prices as well as for ER. We have estimate some eco-investment equations with carbon taxes inclue, but the core moels presente in this paper are base on marginal energy prices without paying attention to carbon taxes. 3) The Community Innovation Surveys for , an This survey is use to obtain ata on the various types of innovation aopte, the R&D inputs into (technological) innovation an other variables, such as e.g. the epenence on foreign markets, innovation subsiies receive from ifferent boies an innovation cooperation. The final eition of CIS can be use to investigate the synergies of simultaneously aopting environmental innovations targete at prouction cost reuctions an environmental innovations targete at ecreasing the environmental buren of final consumption by creating new user value. Because all firms respone to the question on environmental regulation (existing or anticipate) this variable can be seen as an important eterminant for explaining synergy effects among innovations. 4) The Prouction Statistics Survey (PS). This survey contains firm-level ata on gross output, turnover, value ae, intermeiate inputs an the total energy costs of firms. After matching with inustry-level eflators, this source can be use to construct ifferent output measures such as value-ae an gross-output prouctivity, energy cost shares an profitability. Table 1a: Sample coverage (manufacturing only) PS CIS ECF ES ECF&PS ES&ECF ECF&PS &CIS Table 1a summarizes the coverage of ifferent surveys for manufacturing before an after ata linking an before eleting item non-response an/or implausible values (such as a recore negative value ae). The ECF survey has the highest coverage with PS. The ES survey can be consiere as the best source for volume ata on energy use an the istinction between marginal energy prices net of carbon-taxes an the car- 15

19 bon-tax component of gross marginal energy prices. But its match with ECF an PS (which collects ata on energy costs an volumes for gas an electricity) is rather poor. 7 A similar poor match can also be foun when linking ES to CIS (not shown in the table). For this reason we choose not to start with the ata that are available after matching all four available surveys. Instea we use two separate blocks of ata: for the moeling of the three types of eco investment (incluing eco R&D) we use the ECF&PS panel an after calculating (gross) marginal energy prices using the ata available on the volumes of energy use for gas an electricity in the PS survey an the corresponing energy price tariff schemes publishe by SN. For non-eco R&D innovation investment we use the ECF&CIS-panel for istinguishing between eco R&D an non-eco R&D. 8 Thereafter, the preictions from the innovation investment moels are use for moeling the ecisions to innovate. At this stage the CIS ata are imperative, not at least as this is the only source that collects ata on the importance of ER (either existing or anticipate). Annex A presents a list of the variables that are available for assessing the PH in this stuy. A subset of these variables is use in the empirical application. Table 1b summarizes some escriptive statistics for the variables use in the moels. We restrict the iscussion to some interesting results. It is noticeable that when the CIS ata are merge with the PS an ECF ata some variables, such as firm size, eco-r&d per employee, eco-investment an eco subsiies receive, isplay a higher average than before the merging. This is ue to the fact that the CIS survey uses relatively larger firms. 9 The means of the variables that originate from the CIS survey o not change very much after merging with other surveys. It can be seen that eco-r&d is consierably lower than other (non-eco) R&D investment. The share of eco-r&d in total R&D investment expeniture amounts to 30 %. Furthermore, the share of process integrate eco investment in total eco investment (eco R&D exclue) is about 44 %. These percentages remain of the same orer of magnitue when calculate for the full panel obtaine after linking the PS, ECF an CIS surveys. Finally, it can be seen that about 31 % of the firms in this panel respone to ER, either existing or anticipate. A more etaile account of the istribution of some key variables is given in table 1c. In general the istributions are very skew, with small values for the bulk of firms an relatively few firms with substantial eco R&D or other types of eco investment. However, it can also be seen that eco investment is relatively more process integrate if eco-investment is more substantial. Finally, skewness is relatively much smaller for the prouctivity measures use in this stuy. 7 These surveys are carrie out every year. 8 As CIS collects ata on total R&D, only this match enables a istinction between eco - an non-eco R&D. 9 The main objective of the ES survey is to prouce aggregate energy statistics an the istribution of energy use is very skewe to the right. 16

20 Table 1b: Descriptive statistics PS&ECF * CIS ** PS&ECF&CIS ** N Mean st N mean st N mean st eco R&D per fte (1000 Euro) non-eco R&D per fte (1000 Euro) eco investment per fte (1000 Euro) employment in fte's log marginal energy price per TJ share energy tax (after linking with ES) eco subsiies receive (ummy) energy cost share t belonging to enterprise group (ummy) engage in innovation cooperation (ummy) epenent on foreign markets (ummy) subsiies receive from local authorities (ummy) subsiies receive from government boies (ummy) subsiies receive from EU institutions (ummy) existing an anticipate ER (ummy) share of environmental R&D in total R&D share of process integrate eco investment in total eco investment value ae per fte (1000 Euro) log (TFP) prouct innovation aopte (ummy) process innovation aopte (ummy) eco innovation aopte (ummy) * Averages for **Averages for 2004, 2006,

21 Table 1c: Distributions for selecte variables using ES&CIS&PS sample N mean P5 P10 P25 P50 P75 P95 eco R&D per fte (1000 Euro) non-eco R&D per fte (1000 Euro) eco investment per fte (1000 Euro) energy cost share share of environmental R&D in total R&D share of process integrate eco investment in total eco investment employment in fte's ,0 value ae per fte (1000 Euro) log (TFP)

22 5. Discussion of the results We shall present an iscuss in turn the estimation of each part of the moel: the investment equations, the innovation output ecisions an the contribution of innovation to prouctivity performance. The focus of this paper is on the contribution of ER to innovating an the estimation of synergies between environmental innovations an other types of innovations. We postulate that environmental investment can be brought into the picture for obtaining a more in-epth analysis of the Porter Hypothesis an to account for the response of firms to energy price incentives. We also investigate whether ER has a role to play in the ifferent stages of innovation an inirectly on prouctivity. We pool the ata for the years 2004, 2006 an Some of the variables, like the innovation choices, refer to a three-year perio ening respectively in the years just mentione. We control for inustry effects an year fixe effects (except for the investment selection equations). Some of the variables are lagge by one or two years to partly circumvent a simultaneity problem. 5.1 Investment The selection an the outcome equations of the investment ecisions were estimate simultaneously by maximum likelihoo using the tobit type II moel. The results for the probit part of the estimates (see Table 2a) clearly inicate that selectivity is present in the ata. At least for non-eco R&D an en-of-pipe eco investments the correlation coefficients between the error terms in the selection an the outcome equations are statistically significant. For non-eco R&D (column 4) we have controlle for some of the variables that are usually foun in the literature for explaining R&D selection: group belonging, epenency on foreign markets, eman pull an cost push consierations. As often reporte in the literature, size is a significant eterminant of the probability to invest in R&D as well as eman pull an the epenence on foreign markets. It is fair to say that there is perhaps little sense to correct for selection in eco-r&d investment (column 1) as only 24 out of 5528 observations have no eco-r&d investment. Nevertheless size an the importance of environmental levies push firms to invest in eco R&D. It is noteworthy that environmental regulations lea firms to invest even in non-eco R&D. Other (than R&D) eco investments are more frequent in small firms. They seem to be riven by the importance of energy in total cost, the buren of environmental levies in total exploitation cost an the existence of environmental regulations. 20