ABSTRACT
ABSTRACT
Almost all empirical research that has attempted to assess the validity of the Porter hypothesis (PH) has started from reduced-form models, for example, single-equation models for estimating the contribution of environmental regulation to productivity. This paper follows a structural approach that allows testing what is known in the literature as the ‘weak’ and the ‘strong’ version of the PH. Our ‘Green Innovation’ model includes three types of eco-investments to explain differences in the incidence of two types of eco-innovation, which are allowed to affect labor productivity. We allow for complementarity between the two types of eco-innovations. Using a comprehensive panel of Dutch manufacturing firm-level data we estimate the relative importance of environmental regulations on eco-investment and eco-innovations. The results of our analysis show a strong corroboration of the weak and a nuanced corroboration of the strong version of the PH.
1. Introduction
The relationship between technological change and environmental policy has received a lot of attention from scholars and policymakers during the last decades. This is partly because the environmental consequences of social and business activity are affected by the rate and direction of technological change, and partly because environmental policies may create new constraints as well as new incentives that may shape the path of future technological development (Jaffe, Newell, and Stavins 2003).
Environmental technological progress is a very broad phenomenon and every description of it cannot be more than very incomplete. Some examples concern (1) technologies that reduce pollution at the end-of-pipe, such as scrubbers for use on industrial smokestacks or catalytic converters for automobiles, (2) technologies that increase user value for consumer products (e.g. medicines) by introducing new production methods that use materials that are less harmful for the environment and (3) implementation of technologies that are targeted to changes in production processes that improve energy efficiency.
Environmental regulations (ER) in the form of carbon taxes, cap and trade or environmental standards have been put in place to solve two market failures: the lack of environmental innovations that have positive spillover effects on other firms just like any other kind of innovation and the negative spillover effects in the form of water and air pollution, noise, climate change and non-renewable energy depletion connected to industrial activity. Let alone, firms innovate too little especially in environmental technologies (see e.g. Jaffe, Newell, and Stavins 2002; Desrochers 2008; Cerin 2006).
Opponents to these ER argue that they increase costs and reduce the firms’ competitiveness. The Porter hypothesis (PH) asserts that polluting firms can benefit from environmental policies, arguing that well-designed and stringent environmental regulation can stimulate innovations, which in turn increase the productivity of firms or the product value for end users (Porter 1991; Porter and van der Linde 1995). The claim is that there is no trade-off between economic growth and environmental protection but a win–win situation instead. Environmental regulation would benefit both society and regulated firms by triggering dynamic efficiency, and these benefits may partially or fully offset the costs of complying with environmental restrictions.
As argued by Kriechel and Ziesemer (2009), the central issue behind the testing of the PH is the question whether regulation drives innovation. This calls for a structural modeling approach in investigating the contribution of ER on green investment and of green investment on innovation and productive efficiency.1 We will embark on this task by adopting a Green CDM (Crépon, Duguet, and Mairesse 1998) type of model, similar to the model used by Lanoie et al. (2011), that allows testing what Jaffe and Palmer (1997) called the ‘weak’ version and the ‘strong’ version of the PH, referring to the effect of ER on respectively environmental innovations and economic performance. A number of studies have found support for the weak version of the PH, little corroboration, however, exists of the strong version of the PH (see Wagner (2003); Popp, Newell, and Jaffe (2010); Ambec and Barla (2006); and Ambec et al. (2011) for extended reviews). This paper tries to shed new light on the PH by using a rich unbalanced Dutch firm panel data set constructed by matching firms from four different types of surveys.2
The plan of the paper is as follows. In Section 2 an overview of the literature is given. Section 3 discusses the model used in the empirical application. Section 4 presents the data. The results are presented and commented in Section 5. Section 6 concludes.
2. Literature review
There is a vast body of literature devoted to appraise the seminal contributions of Porter (1991) and Porter and van der Linde (1995). Originating primarily from empirical regularities found in the analysis of cross-country differences in the stringency of environmental regulation and economic performance, the PH has triggered a lot of research both theoretical and empirical in nature. It has been criticized for being merely based on anecdotal stories and for the lack of a sound theoretical basis (see Palmer, Oates, and Portney 1995; Cerin 2006).
Some research attempts to provide a theoretical underpinning of the PH. Mohr (2002) argues that it is a feasible outcome if one allows for the possibility of endogenous technical change. More recent theoretical contributions that link the environment to endogenous growth are given in Acemoglu et al. (2012) and Gans (2012). Ambec and Barla (2002) raise the question whether regulation is indeed needed for firms to adopt profit-increasing innovations, pointing to the first primitive of the PH, that is, that firms systematically ignore opportunities for profit-increasing innovations, and that ER can motivate firms to capture ‘low hanging fruit’ offered by environmental challenges to their business. The literature of behavioral economics offers several explanations for underinvesting in environmental innovations (see examples in Ambec and Barla 2006). Other studies have analyzed the interaction between competition and innovation. A recent example is the introduction of organic pharmaceutical products discussed in Constantatos and Hermann (2011). By avoiding the use of environmentally damaging fertilizers there is less environmental burden as well as more user value created because organic drugs are healthier. But a win–win situation between regulation and innovation is not self-evident because of conjectural variations, consumer inertia or potential first-mover disadvantages.
Another strand of research focuses on the second primitive of the PH, that is, the assertion that ER should be well-designed and stringent enough to be successful also from an economic point of view. An assessment of the instruments of environmental regulation and a judgment of their effectiveness can be found in Wagner (2003). The myriad of environmental instruments can best be understood by distinguishing between command and control and market-based regulations. The instruments that set emission limits and standards fall into the first class and are often labeled ‘end-of-pipe’ regulations. Environmental taxes and tradable emission permits are examples of the second class of instruments. Environmental effectiveness can be defined as the ability to achieve a predefined environmental target. The general view is that this definition is more appropriate for the first class of instruments. By contrast, the second class of instruments has a higher economic profile, because they are aimed at triggering static and dynamic efficiency and internalizing environmental externalities.
Looking at the empirical evidence provided in the literature it can be concluded that the picture is rather mixed (see also the meta-analysis of Cohen and Tubb 2015). The number of papers and articles that have put the PH to the empirical testing is overwhelming but they do not to lead to a general consensus. Much of this has to do with the different research strategies and the availability of data. Compared to empirical evidence at the macro or industry level, the number of papers that use firm-level data is rather scarce. Besides that, research is targeted at different measures of performance.
Cutting through different reviews of the empirical work testing the PH, it can be concluded that many studies investigate the impact of environmental regulation on productivity or productive efficiency in a reduced-form estimation approach and that the empirical evidence is mostly at the macro or industry level. In many cases this type of research leads to the conclusion that environmental regulation has a negative impact on productivity. This conclusion can be easily understood, because regulation forces firms to invest in the environment and doing so increases production costs. If these investments do not lead to the renewal of production processes then there is no reason for expecting substantial gains in resource efficiency. There is, however, a measurement issue. ‘End-of-pipe’ investments may reduce pollution but this reduction is not accounted for in output. The same capital and other inputs produce two types of output: bad and good output and it is hardly possible to value the contribution of (reducing) bad output. This raises serious problems when investigating the relation between environmental regulation and (productive) efficiency.3 An interesting solution to circumvent this problem is presented in Domazlicky and Weber (2004) and Weiss (2015). They use data on toxic releases as bad outputs together with traditional output measures such as real value added in a non-parametric analysis to identify technical change from efficiency change. Domazlicky and Weber (2004) conclude that regulation has a negative impact on total factor productivity (TFP). Weiss (2015) finds that Swedish ER do not affect conventional TFP growth but well green TFP growth in the Swedish pulp and paper industry.4 Huiban, Mastromarco, and Musolesi (2014) find that the elasticity of output with respect to pollution abatement capital stock is on average positive in the French food industry.
More interesting is the research that looks into the impact of environmental regulation on innovation. Notable examples of this type of research can be found in several papers based on the Mannheim Innovation Panel. The focus of research stretches from regulation driven innovation alone (Rennings and Rexhäuser 2010) to the impact of regulation driven innovation on competitiveness (Rennings and Rammer 2010), employment dynamics (Horbach and Rennings 2012) or profitability (Rexhäuser and Rammer 2014). Support for the PH is provided by concluding that environmental regulation does not harm competitiveness (Rennings and Rammer 2010) and that the contribution of regulation induced innovation to profitability is larger than the contribution of other (more) voluntary innovations (Rexhäuser and Rammer 2014). On data from the Swedish pulp and paper industry Weiss (2015) finds confirmation of the narrow version of the PH, according to which prescriptive regulations are less efficient than incentive-based regulations. Dussaux (2015) on French data does not find evidence in favor of either the strong or the narrow version of the PH but he finds that ER lead to a change in the direction of innovation towards more green innovation efforts.
Ideally, a thorough empirical testing of the PH would rely on data of particular types of regulations that would trigger innovations at the firm level. The most recent edition of Community Innovation Surveys (CIS) contains new questions on environmental innovation. However, for the Netherlands, this new module cannot be used to identify the role of different 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 anticipated regulations. By contrast, the German CIS allows a distinction between types of ER. After matching the firm responses with external data on the age of regulations, Rennings and Rexhäuser (2010) also investigated the long-term impact of different types of regulation on the adoption of environmental innovation. Surprisingly they find that command and control regulations of the end-of-pipe type can stimulate certain types of long-term innovation.
3. A Green CDM model
The empirical model used in this paper is a modified version of the so-called ‘CDM model’ (following the initials of Crépon, Duguet, and Mairesse 1998). A graphical presentation of our model is given in Figure 1.5 The upper part of the figure points to the investment decision stage. In the traditional CDM model this concerns the decision on how much to invest in R&D. In this paper we face another investment decision problem, that is, namely how much to invest in order to reduce the environmental burden of the firm’s operations. The second block of the model describes the green innovation function with eco-R&D and other eco-investments as an input and eco-innovations, possibly of different kinds, as outputs. The third block examines the link between innovation outputs and productivity as a measure of economic performance. We shall now describe each block in detail.
Published online:
15 July 2016![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/gein20/2017/gein20.v026.i01-02/10438599.2016.1202521/20170204/images/medium/gein_a_1202521_f0001_c.jpg"})
Figure 1. A ‘Green CDM model’.
3.1. The eco-innovation input decisions
We consider three types of eco-innovation inputs denoted by 
, where subscript i denotes the enterprise, subscript t the time period and superscript k the type of eco-innovation input. There are three types of them : (1) eco-R&D, that is, research and development expenditure aimed at green products or green technologies, (2) eco-investments in end-of-pipe facilities, those that reduce pollution but do not fundamentally change the production process and (3) eco-investments that are ‘process integrated’, that is, investments in new technologies that are less polluting. For each type of investment, we assume that firms decide first whether to invest or not and then, if they invest, they choose the investment intensity. Firms will invest in each of these types of investment if the latent value of investing (denoted by a
) exceeds some threshold value
, which for all practical purposes we can put equal to zero. We can think of
as the expected return from such an investment.
The decision to invest in input k, 
, can be expressed as follows:
(1)
The vector 
in (1) collects the variables that determine the decision to invest in eco-innovation input k. If firm i decides to invest in input k the amount that is going to be invested will be determined by the following equation:
(2)
The vector 
in (2) contains all the variables that determine the intensity of eco-innovation input type k. The unobservables that determine the decision to invest and the intensity of investment in eco-innovation input k, respectively
and
are assumed to be jointly normally distributed with mean 0 and variance-covariance matrix
. The system of equations (1) and (2) is a Tobit type II selection model (Amemiya 1984). It can be applied to each of the three innovation input decisions.
Although we are dealing in each case with a general investment problem, one can imagine that the three types of investment are rather distinct. This is particularly the case if we compare eco-investment performed to comply with ‘control type’ environmental regulation with R&D that is aimed at developing new goods. Not at least because of the difficulty of evaluating the output of ‘bad goods’ or of ‘process integrated’ eco-investment, which by definition aims at reducing bad output and increasing resource efficiency, it is impossible to use standard capital and investment theory to derive formal investment models. At least, we consider such an exercise beyond the scope of this research.
We let both the decision to invest and the intensity of eco-investment be a function of the size of the firm. Larger firms are expected to have greater means and incentives to invest in eco-investments. It is less certain that the intensity is also affected by the size of the firm, but we do not want to exclude a scale effect. Government can internalize the environmental externalities by imposing environmental levies or by taxing the price of energy. The environmental levies in total environmental exploitation costs are provided in the Environmental Costs of Firms (ECF) data set. Constructing an energy price, and especially a marginal tax for energy, is feasible but would substantially reduce the size of our sample. Therefore as an indirect measure of the incentive to save on energy because of an energy tax we use the energy cost share in total cost in both equations. We assume the relative size of the environmental levies to affect the decision to initiate eco-investments but not the intensity of these investments as opposed to the energy cost share, which we let affect both the decision to invest and the intensity of investment. In the intensity of investment equation we also control for the presence of eco subsidies, the relative price of investment and labor,6 as well as industry and time effects. Our central variable of interest, the importance of environmental regulation (existing or anticipated), is inserted as an explanatory variable in each equation. Whenever possible we use one-period lagged variables to minimize the possibility of endogeneity.
3.2. Eco-innovation outputs
The middle part of Figure 1 indicates that eco-investment leads eco-innovation output. We consider two types of eco-innovation output: pollution-reducing and resource-saving eco-innovations. The former captures environmental innovations targeted at lower utilization of energy or materials and can be seen as a form of end-of-pipe eco-innovation, aiming just at reducing pollution but not at changing the production technology altogether. The latter captures environmental innovations targeted at lower CO2 emissions, lower use of polluting materials, less pollution in the production process and improvements in recycling, and can be considered as a deeper innovation that aims at curbing pollution by improving the production technology itself. We observe dichotomous variables indicating whether a certain innovation has been adopted or not. The two types of innovation are likely to be interrelated in the sense that the return to one innovation could depend on the adoption of the other one for reasons of complementarity or substitutability between them. It is well documented in the econometric literature (see e.g. Heckman 1978; Tamer 2003; Lewbel 2007) that the estimation of a bivariate probit with endogenous dummy variables raises severe problems of identification. There can be no solution (in which case the system is said to be incoherent) or multiple solutions (in which case it is said to be incomplete). The empirical literature offers several solutions to this problem. In general, these solutions boil down to imposing zero restrictions on the coefficients of some of the binary endogenous explanatory variables or by relying on recursive or triangular systems, in which one of the choices is assumed to be leading (see for a discussion of completeness and coherency section 2 of Tamer (2003)). One way to avoid incoherency and incompleteness is to start from a McFadden (1974) solution by considering a multinomial choice problem based on a random utility model. This framework was proposed more recently by Lewbel (2007) and applied by Miravete and Pernías (2006) and Kretschmer, Miravete, and Pernías (2012).
Let the total utility (in this case profit) be
(3)
The dichotomous variables for the two types of innovation are given by 
There are in total four possible combinations of innovation choices yielding respectively the following profit outcomes:
(4a)
(4b)
(4c)
(4d)
The ‘complementarity parameters’ 
and
are placed in parenthesis because only their sum can be identified.7 If
(<0), the two innovations are complements (substitutes). The model is complete because (latent) profitability is specified for all possible strategies and coherent because every strategy should have a latent profit that exceeds the profits of all other strategies. As pointed out by Lewbel (2007), the difference with respect to the traditional multinomial choice framework is that we do not have a separate specification for
such as
. Instead, we use (4d) derived from the model for the total latent profit function. In the Appendix we give a more detailed account of the construction of the likelihood function and its estimation for a somewhat more general model, namely a trivariate simultaneous probit, of which the bivariate is just a special case. The
(l = 1,2), are random errors that are supposed to be jointly normally distributed.
The profitability of pursuing a particular eco-innovation output 
depends on the adoption of the other eco-innovation output through the ‘complementarity parameter’. It also depends on the magnitude of the three types of eco-innovation inputs – it is, indeed, interesting to find out which investments affect which types of innovation. We also control for the size of the firm, a dummy for innovation cooperation, a set of industry and time dummies and especially the environmental regulation dummy, which captures whether firms respond to either existing or anticipated ER. Since the three eco-investment intensities are endogenous, we replace them by the conditional predictions of their latent variables estimated in the first step. In essence, the coefficient of environmental regulation provides a test of the weak version of the PH in the terminology of Jaffe and Palmer (1997).
3.3. Productivity
Finally, we investigate the strong version of the PH by relating ER to labor productivity (LP) via the effects of eco-innovation output on the TFP component of LP. Suppose we have a Cobb–Douglas production function. We can then write the expression for LP as follows:
(5) where
is gross output,
is labor,
is the capital stock,
stands for materials,
is a vector of other control variables and
is a random error term that captures the unobservables. Thus we regress the log of LP on the log of the capital-labor ratio, the log of the materials-labor ratio and the log of employment, the coefficient of which captures deviations from constant returns to scale. We make TFP depend on the presence of the two eco-innovation outputs, pollution-reducing and resource-saving environmental innovations. We control for industry and year dummies. Since the innovation output measures are endogenous, we instrument them by the predicted latent innovation variables from stage 2 (see e.g. Wooldridge 2002). As a robustness check we also account for the endogeneity of the traditional inputs by instrumenting them by their lagged values.
4. Data
We have constructed a comprehensive data set by linking manufacturing firm-level data for 2000–2008 from three different surveys:8
The Survey on ECF: This yearly survey covers the period 2000–2008 and beyond. This is one of the most important data sources for this research project. The survey collects (amongst others) data on environmental current exploitation costs, environmental subsidies, expenses on environmental R&D and two types of non-R&D environmental investments: ‘end-of-pipe’ eco-investments and those related to the renewing of production processes (so-called ‘process-integrated eco-investment’). Because this survey only collects data for manufacturing, our empirical analysis will be restricted to this branch of the economy.
The CIS survey that during our period of interest was conducted every two years, covering the periods 2002–2004, 2004–2006 and 2006–2008. This survey contains data on the existence or anticipation of environmental regulation and about a number of environmental innovation targets: the reduced use of materials and of energy per unit of output (which we shall denote as resource-saving eco-innovations) and the reduced CO2 footprint, the replacement of materials with less polluting and hazardous substitutes, the reduced soil, water, noise and air pollution, and the recycled waste, water and materials (all four of which we shall regroup under the heading of pollution-reducing eco-innovations).9 Respondents were also asked whether they cooperated in innovation.
The Production Statistics Survey (PS): This yearly survey contains firm-level data on gross output, turnover, value added, intermediate inputs, replacement investment and the total energy costs. If the depreciation rate is constant, replacement investment can serve as a proxy for capital stock.
Table 1 summarizes the coverage of different surveys for manufacturing before data linking and before deleting ‘item non-response’ and/or implausible values (such as a recorded negative value added). The ECF survey has the highest coverage and the CIS survey the lowest. To include as many observations as possible we chose not to start with the data that are available after matching all three available surveys. Instead, we tried to maximize the number of observations by taking all the yearly data from the ECF and PS surveys and repeating the data from CIS for two successive years. For the modeling of the three types of eco-investment we used essentially the ECF and PS panel. The only variable that came from CIS was the environmental regulation. For that variable we put the values for the years 2003, 2005 and 2007 equal to those for 2004, 2006 and 2008 respectively, which is not unreasonable since the CIS surveys of 2004, 2006 and 2008 cover respectively the periods 2002–2004, 2004–2006 and 2006–2008. The predictions from the eco-innovation investment equations are then used for modeling the eco-innovation outputs. At this stage we absolutely need the CIS data because it is the only source that collects data on the perception of ER. The dummy variables that are taken from CIS, that is, the two eco-innovation choices, ER and innovation cooperation, are again imputed for the years 2003, 2005 and 2007 as explained above. All other variables are directly taken from the yearly surveys.
Table 1. Sample coverage (number of manufacturing firms).
Table 2 presents some descriptive statistics for the variables used in the models. The means of the variables do not change very much before and after merging datasets. It can be seen that about half as much is paid on eco-R&D as on the other two types of (non R&D) eco-investments. The end-of-pipe eco-investment is roughly twice as big as the process-integrated eco-investment. About 30% of the firms responded to ER, either existing or anticipated. Approximately 30% of the firms in the sample underlying the productivity equation had introduced resource-saving eco-innovations and 3% more had introduced pollution-reducing eco-innovations.10
Table 2. Descriptive statistics.
5. Discussion of the results
We shall now present and discuss the estimation of each part of the model: the investment equations, the innovation output decisions and the contribution of innovation to the productivity performance. The focus of this paper is on the contribution of ER to innovation. We postulate that environmental investment can be brought into the picture for obtaining a more in-depth analysis of the PH letting ER affect the different stages of innovation and indirectly productivity.
We pool the data for the years 2003–2008. We control for industry effects and year fixed effects (except for the investment selection equations). Some of the variables are lagged by one year to partly circumvent an endogeneity problem.
5.1. Eco-innovation inputs
The selection and the outcome equations of the investment decisions were estimated simultaneously by maximum likelihood using the tobit type II model. The results for the probit part of the estimates (see Table 3) clearly indicate that selectivity is present in the data: the correlation coefficients between the error terms in the selection and the outcome equations are statistically significant. Larger firms have a higher propensity to invest in eco-R&D but lower propensity to invest in the other two types of eco-investment. The intensity of eco-investments decreases with size. The energy cost share, environmental levies and environmental regulation all push firms to invest in eco-R&D and other eco-investments. Firms tend to invest more in eco-innovation when the relative price of investment decreases, when the energy cost goes up, when they get innovation subsidies or when they are forced to innovate because of environmental regulation. Almost all the firms do some eco-R&D and therefore they are less responsive to ER for increasing their eco-R&D. The other two types of eco-investments react more strongly to ER with semi-elasticities of 0.4 and 0.32.
Table 3. Eco-innovation input equations (tobit type II).
Thus, already at the investment stage of innovation, there is a role for ER and market-based incentives in the form of taxes, price and cost considerations in explaining differences in eco-investment intensities.
5.2. Eco-innovation outputs
The main focus of this paper is on the innovation decisions of firms (i.e. the innovation output stage of our ‘Green CDM’ model). Table 4 presents the results for the model that uses two types of innovations: pollution-reducing and resource-saving innovations. The former can be assimilated with end-of-pipe eco-innovations, the latter with process-integrated eco-innovations. We report two types of estimates, those of a bivariate probit model in which the effects of common unobservable variables are captured by correlations between the error terms of the three equations, and a simultaneous bivariate probit model with endogenous dummies, our model (3) inspired by Lewbel (2007), in which we allow for complementarity between the two innovation adoption decisions. In principle, the former model is nested in the latter, but the programs used to estimate the two bivariate probit models are different in their computations of the boundaries of the integrations and therefore not strictly comparable in terms of achieved value of the maximum likelihood function. But, since the synergy coefficient, the only difference between the two models, is significant, we clearly prefer the model with simultaneity. We shall therefore base the discussion on the estimates from that model.
Table 4. Eco-innovation outputs (bivariate probit model, with and without simultaneity).
Eco-R&D has a positive effect on both types of innovations. A 1% increase in eco-R&D increases the probability of adopting a pollution-reducing innovation by 0.36 percentage points and the probability of adopting a resource-saving innovation by 0.24 percentage points. End-of-pipe eco-investments have a weak effect on eco-innovation: a marginal effect of 0.12, significant only at 10%, for pollution-reducing innovations and an insignificant effect on resource-saving innovations. A one percentage increase in process-integrated eco-investments decreases by 0.37 percentage points the likelihood of obtaining a pollution-reducing innovation but increases by 0.10 percentage point the probability of obtaining a process-integrated innovation. These direct effects get magnified by the synergy between the two types of innovations as the presence of one type of innovation increases by 47 percentage points the occurrence of the other type of innovation. Consequently the total effects of any exogenous changes in the determinants of innovation are roughly twice as large as the direct effects.
Our results also show that innovation cooperation increases the incidence of both types of innovation but that size only ‘matters’ for pollution-reducing innovation. Most interestingly, our results show that ER influence the incidence of both types of innovation. The presence of ER increases by 45 percentage points the occurrence of pollution-reducing eco-innovation [0.45 = exp(0.37) − 1], and by 25 percentage point the occurrence of resource-saving innovation. In other words, in addition to the indirect effect of ER on innovation investment, there is also an important direct effect of ER on the incidence of each type of eco-innovation. We consider this last result as a strong corroboration of the weak version of the PH.
5.3. Productivity
This section looks at the productivity impact of different types of eco-innovations, that is, it examines indirectly the strong version of the PH. The OLS estimates presented in the first two columns of Table 5 show that there are hardly any economies or diseconomies of scale: the coefficient of the logarithm of employment is not different from 0. The direct effect of eco-innovation is negative for pollution-reducing innovations and positive for resource-saving innovations.
Table 5. Labor productivity (in log).
A drawback of the OLS estimation is that the innovation dummies are not exogenous. To circumvent these shortcomings we have re-estimated the productivity equation using the GMM instrumental variables (IV) method. The GMM estimation presented in columns 3 and 4 uses the predicted propensities derived from the innovation output model (3) as instruments for the two eco-innovation outputs. In the last two columns of Table 5 we also let the traditional factors of production be endogenous and instrument them by their one-period lagged values. The results of the GMM estimation clearly show that the endogeneity of the innovation output dummies is an important issue. Although the coefficients have quite different magnitudes, in essence they convey the same story. Pollution-reducing assimilated to end-of-pipe eco-innovations are negatively correlated to productivity whereas resource-saving assimilated to production-integrated eco-innovations are positively correlated to TFP. Whenever we introduced environmental regulation directly in the productivity equation, it never turned out significant. In practice both types of eco-innovation occur jointly: the frequencies of occurrence of resp. pollution-reducing and resource-saving eco-innovations are 63% for (0,0), 6% for (0,1), 10% for (1,0) and 21% for (1,1). This high frequency of co-occurrence confirms the complementarity reported in the previous section. It also explains why the environmental innovation is insignificant when not split into its two components: most of the time the two innovations occur jointly, the chances of single occurrence are almost equal and the marginal effects are also almost equal but of opposite sign.11
Unlike most of the literature (see Ambec et al. 2011; Lanoie et al. 2011), we do come up with a corroboration of the PH but only for process-integrated eco-innovations. As Horbach and Rennings (2012) have found with German data regarding the employment effects of environmental innovations, only process-integrated eco-innovations have a positive effect on economic performance.12
6. Conclusion
This paper presents a new attempt to investigate the validity of the PH using a more structural modeling approach than mostly used up to now in the mainstream of empirical research on this topic. We apply a ‘Green type of CDM’ innovation model to a very comprehensive data set built after matching three Dutch firm survey data. We distinguish three types of eco-investments: eco-R&D, end-of-pipe eco-investments and process-integrated eco-investments and two types of eco-innovation outputs: pollution-reducing and resource-saving. We model the determinants of eco-investments, eco-innovations and productivity, with eco-investments affecting innovation outputs and the latter affecting productivity. We are particularly interested in testing the so-called PH that predicted a positive impact of ER not only on eco-investments and eco-innovations but also on economic performances like TFP.
Our empirical results strongly corroborate the weak version of the PH. There is a significantly positive contribution of existing or anticipated ER on eco-innovations, directly and indirectly via their positive effect on eco-investments that in turn boost the propensity of introducing environmental innovations. This driving effect is strengthened by the complementarity between the two types of eco-innovations. Moreover the use of environmental levies seems to be an important element in the decision making of firms to invest in eco-R&D and eco-investments and play indirectly a major role in the introduction of eco-innovations.
Whereas many studies cannot corroborate the strong version of the PH we are able to detect one way in which ER can boost TFP. Resource-saving eco-innovations, which can be assimilated to process-integrated eco-innovations, increase TFP, whereas pollution-reducing, that is, end-of-pipe, eco-innovations tend to reduce TFP. The marginal effects of both are almost equal, and in the Netherlands at least they occur mostly jointly and have about the same propensity of occurring individually. Therefore eco-innovation globally does not turn out significantly. ER do not affect TFP directly. Eco-regulations, if properly aimed at process-integrated eco-innovations, can have a positive effect on TFP. And maybe if we allowed for more time between the inception of regulations, the occurrence of innovations and their effect on productivity, we would get an even stronger result.
| PS | CIS | ECF | |
|---|---|---|---|
| 2002 | 5751 | 8782 | |
| 2004 | 4966 | 2538 | 7867 |
| 2006 | 4300 | 2133 | 7296 |
| 2008 | 3808 | 2164 | 7230 |
PS, Production Statistics Survey; CIS, Community Innovation Survey; ECF, Survey on Environmental Costs of Firms.
| Eco-investment equations | Eco-innovation equations | Productivity equation | |||||||
|---|---|---|---|---|---|---|---|---|---|
| N | Mean | Std | N | Mean | Std | N | Mean | Std | |
| Employment in ftea | 5989 | 132.6 | 292.3 | 6039 | 132.4 | 291.3 | 6581 | 143.1 | 293.8 |
| Eco subsidies received (dummy) | 5989 | 0.339 | 0.473 | 6039 | 0.340 | 0.474 | |||
| Eco-R&D per fte (1000 Euro) in t | 5987 | 0.106 | 0.256 | 6039 | 0.063b | 0.043 | 6581 | 0.113 | 0.320 |
| End-of-pipe eco-investment per fte (1000 Euro) in t | 5987 | 0.137 | 1.487 | 6039 | 0.014b | 0.019 | 6581 | 0.168 | 2.409 |
| Process-integrated eco-investment per fte (1000 Euro) in t | 5987 | 0.070 | 0.571 | 6039 | 0.020b | 0.026 | 6581 | 0.085 | 0.980 |
| Energy cost shares in total production costsa | 5989 | 0.017 | 0.024 | ||||||
| Environmental levies/total environmental exploitation costsa | 5989 | 0.081 | 0.121 | ||||||
| Existing and anticipated ER (dummy) | 5989 | 0.285 | 0.451 | 6039 | 0.284 | 0.451 | 6581 | 0.300 | 0.458 |
| Pollution-reducing eco-innovations (y31, dummy) | 6039 | 0.321 | 0.467 | 6581 | 0.334 | 0.472 | |||
| Resource-saving eco-innovations (y32, dummy) | 6039 | 0.287 | 0.453 | 6581 | 0.302 | 0.459 | |||
| Innovation cooperation (dummy) | 6039 | 0.268 | 0.443 | ||||||
| Gross output per fte (1000 Euro) | 6581 | 229.2 | 311.3 | ||||||
| Capital inputs per fte (1000 Euro) | 6581 | 7.4 | 10.0 | ||||||
| Material inputs per fte (1000 Euro) | 6581 | 167.3 | 294.0 | ||||||
| Value added per fte (1000 Euro) | 6581 | 61.8 | 39.3 | ||||||
aIn the eco-innovation investment models and the eco-innovation output model we used one-year lagged variables.
bConditional predictions of eco-innovation investments per fte from tobit type II model for 2003–2008.
| Eco-R&D | Eco-investments | |||||
|---|---|---|---|---|---|---|
| End-of-pipe | Process integrated | |||||
| N total | 5987 | 5987 | 5987 | |||
| N censored | 28 | 3666 | 3829 | |||
| Selection | Coefficient | SE | Coefficient | SE | Coefficient | SE |
| log(employment in fte) in t − 1 | 0.25*** | 0.07 | −0.06*** | 0.02 | −0.06*** | 0.02 |
| log(energy cost share) in t − 1+ | 0.21*** | 0.06 | 0.06*** | 0.02 | 0.05*** | 0.02 |
| log(environmental levies share) in t − 1# | 0.06 | 0.07 | 0.06*** | 0.01 | 0.07*** | 0.01 |
| Environmental regulation dummy | 0.36* | 0.21 | 0.18*** | 0.04 | 0.20*** | 0.04 |
| Constant | 2.72*** | 0.51 | 0.39*** | 0.11 | 0.32*** | 0.11 |
| log(eco-investments per fte) | Coefficient | SE | Coefficient | SE | Coefficient | SE |
| log(employment in fte) in t−1 | −0.14*** | 0.02 | −0.08* | 0.04 | 0.03 | 0.05 |
| Environmental regulation dummy | 0.10*** | 0.03 | 0.40*** | 0.09 | 0.32*** | 0.09 |
| log(p_investment/p_labor) in t − 1 | −0.49*** | 0.06 | −0.65*** | 0.13 | −0.34*** | 0.12 |
| log(energy cost share) in t − 1 | 0.10*** | 0.02 | 0.41*** | 0.05 | 0.36*** | 0.05 |
| Eco subsidies received dummy | 0.81*** | 0.05 | 1.04*** | 0.11 | 1.25*** | 0.14 |
| Industry dummies | Yes | Yes | Yes | |||
| Time dummies | Yes | Yes | Yes | |||
| rho | 0.45*** | 0.06 | 0.80*** | 0.03 | 0.51*** | 0.16 |
| Log likelihood | −8619.4 | −8228.8 | −7908.9 | |||
Note: +In total costs; #In total environmental exploitation costs; *Significant at 10%; ***Significant at 1%.
| Without simultaneity | With simultaneity | |||||||
|---|---|---|---|---|---|---|---|---|
| Pollution-reducing eco-innovations | Resource-saving eco-innovations | Pollution-reducing eco-innovations | Resource-saving eco-innovations | |||||
| MEa | SE | MEa | SE | MEa | SE | MEa | SE | |
| N observations (2002–2008) | 6039 | 6039 | ||||||
| log(fte) t − 1 | 0.01 | 0.01 | 0.02 | 0.01 | 0.12*** | 0.01 | 0.00 | 0.00 |
| Predicted log(eco-R&D per fte) | −0.04 | 0.05 | 0.16*** | 0.04 | 0.36*** | 0.06 | 0.24*** | 0.04 |
| Predicted log(end-of-pipe eco-investment per fte) | 0.13** | 0.05 | 0.06 | 0.05 | 0.12* | 0.06 | 0.04 | 0.04 |
| Predicted log(process-integrated eco-investment per fte) | −0.06 | 0.06 | 0.06 | 0.06 | −0.37*** | 0.07 | 0.10** | 0.05 |
| Environmental regulation (ER) | 0.57*** | 0.02 | 0.42*** | 0.01 | 0.37*** | 0.02 | 0.22*** | 0.01 |
| Innovation cooperation | 0.15*** | 0.01 | 0.13*** | 0.01 | 0.50*** | 0.02 | 0.24*** | 0.01 |
| Industry dummies | Yes | Yes | Yes | Yes | ||||
| Year dummies | Yes | Yes | Yes | Yes | ||||
| Synergy between both types of innovation output | – | – | 0.47*** | 0.03 | ||||
| Rho | 0.68***b | 0.01***b | ||||||
| Log likelihood | −4591.41 | −6898.96 | ||||||
aThe marginal effects are evaluated at the means of the covariates.
bThe rho parameter is calculated using ‘generalized residuals’ (see Gourieroux et al. 1987).
*Significant at 10%; **Significant at 5%; ***Significant at 1%.
| OLS | GMMa | GMMb | ||||
|---|---|---|---|---|---|---|
| Coeff. | SE | Coeff. | SE | Coeff. | SE | |
| log(capital/labor) | 0.05*** | 0.00 | 0.05*** | 0.01 | 0.06*** | 0.01 |
| Log(intermediate inputs/labor) | 0.76*** | 0.00 | 0.77*** | 0.01 | 0.77*** | 0.01 |
| log(labor) | −0.00 | 0.00 | −0.01 | 0.01 | −0.01* | 0.00 |
| Pollution-reducing eco-innovations | −0.09** | 0.04 | −0.48*** | 0.15 | −0.52*** | 0.17 |
| Resource-saving eco-innovations | 0.00 | 0.00 | 0.51*** | 0.17 | 0.56*** | 0.19 |
| _cons | 1.48*** | 0.01 | 1.47*** | 0.03 | 1.47*** | 0.03 |
| Year dummies | Yes | Yes | Yes | |||
| Industry dummies | Yes | Yes | Yes | |||
| R2 | 0.96 | |||||
| N | 6581 | 4606 | 4516 | |||
aThe two types of eco-innovations are instrumented by all exogenous variables and the predicted occurrence of the eco- innovation adoption obtained from the estimates reported in Table 4.
bThe two types of eco-innovations and the three inputs are instrumented by all exogenous variables, the lagged inputs and the predicted occurrence of the eco-innovations obtained from the estimates reported in Table 4.
Significant at 1% (***), 5% (**), 10% (*). No Hansen test is reported since there is no overidentification.
Acknowledgement
This paper was presented at various seminars, conferences and workshops. We thank all participants for their critical remarks. We are especially grateful to Andreas Stephan for his careful reading of the manuscript and to Mark Cohen for his critical remarks.
Disclosure statement
No potential conflict of interest was reported by the authors
Notes
1. Green investment refers to investments aimed at reducing the environmental burden of production (see Kemp 2011).
2. The terms eco, environmental and green innovation will be used interchangeably, indicating each time an innovation with a lower detrimental impact on the environment.
3. This problem is well recognized by statisticians and environmental accounting is an important avenue for National Accounts. See Muller, Mendelsohn, and Nordhaus (2011) for a recent contribution to this problem.
4. The method used is the ‘directional output distance approach’ developed by Chung, Färe, and Grosskopf (1997) for constructing the Malmquist–Luneberger index to decompose (changes in) TFP.
5. The figure is an adapted version of the one presented in Crépon, Duguet, and Mairesse (1998).
6. The investment deflator is specific to each type of investment but common for all firms in a given sector. The price of labour, however, is firm specific and is obtained by dividing the cost of labour by the number of full-time employees.
7. Notice that if the
’s are equal to zero, we are in the presence of a bivariate Probit model.
8. We could also have used the yearly Energy use Survey, which collects volume data on energy consumption of different types. These could have been used to construct marginal energy prices at the firm-level after linking them with the data on energy costs collected in the Production Surveys. Unfortunately, because of the poor coverage when combined with CIS we would have faced a considerable loss of data.
9. Since we concentrate on environmental regulations and firm performance, we do not exploit the data on environmental benefits from the after sales use of a good or service by the end user.
10. Notice that we have constructed the predicted eco-investments per fte conditional on doing an eco-investment also for firms for which some observations on environmental levies were missing. Therefore we have slightly more observations for the eco-innovation output equations than for the eco-investment equations.
11. Our preferred specification is exactly identified, and therefore we cannot perform the test of overidentification. But if we add the lagged values of the three predicted eco-investments, the Hansen test does not reject the test of overidentifying restrictions attesting to the quality of our instruments. We do not report the latter specification because the eco-innovation output effects are then no longer significant.
12. In a previous version of the paper we tried to examine the complementarity/substitutability between eco-innovations and product or process innovations. Unfortunately we cannot separate the eco and non-eco product and process innovations, and therefore product and process innovations partly contain eco-innovations. Contrary to Marin (2012) who used a model similar to ours but patent counts instead of innovation occurrences as a measure of innovation output on Italian data, we did not find that environmental innovations crowd out technological innovations.
13. To simplify the exposition, we have ignored the i and t subscripts and replaced the choice superscripts by subscripts.
14. We use superscripts to denote the sum of
and
. Thus,
References
- Acemoglu, D., P. Aghion, L. Bursztyn, and D. Hemous. 2012. “The Environment and Directed Technical Change.” American Economic Review 102 (1): 131–166. doi: 10.1257/aer.102.1.131 [Crossref], [PubMed], [Web of Science ®], [Google Scholar]
- Ambec, S., and P. Barla. 2002. “A Theoretical Foundation of the Porter Hypothesis.” Economic Letters 75: 355–360. doi: 10.1016/S0165-1765(02)00005-8 [Crossref], [Web of Science ®], [Google Scholar]
- Ambec, S., and P. Barla. 2006. “Can Environmental Regulation be Good for Business? An Assessment of the Porter Hypothesis.” Energy Studies Review 14 (2), 42–62. doi: 10.15173/esr.v14i2.493 [Crossref], [Google Scholar]
- Ambec, S., M. A. Cohen, S. Elgie, and P. Lanoie. 2011. “The Porter Hypothesis at 20.” Discussion Paper 11-01, Resources for the Future. [Google Scholar]
- Amemiya, T. 1984. “Tobit Models: a Survey.” Journal of Econometrics 23: 3–62. doi: 10.1016/0304-4076(84)90074-5 [Crossref], [Web of Science ®], [Google Scholar]
- Cappellari, L., and S. P. Jenkins. 2003. “Multivariate Probit Regression Using Simulated Maximum Likelihood.” The Stata Journal 3 (3): 278–294. [Crossref], [Web of Science ®], [Google Scholar]
- Cappellari, L., and S. P. Jenkins. 2006. “Calculation of Multivariate Normal Probabilities by Simulation, with Applications to Maximum Simulated Likelihood Estimation.” The Stata Journal 6 (2): 156–189. [Crossref], [Web of Science ®], [Google Scholar]
- Cerin, P. 2006. “Bringing Economic Opportunity into Line with Environmental Influence: A Discussion on the Coase Theorem and the Porter and van der Linde Hypothesis.” Ecological Economics 56: 209–225. doi: 10.1016/j.ecolecon.2005.01.016 [Crossref], [Web of Science ®], [Google Scholar]
- Chung, Y. H., R. Färe, and S. Grosskopf. 1997. “Productivity and Undesirable Outputs: A Directional Distance Function Approach.” Journal of Environmental Management 51: 229–240. doi: 10.1006/jema.1997.0146 [Crossref], [Web of Science ®], [Google Scholar]
- Cohen, M., and A. Tubb. 2015. “The Impact of Environmental Regulation on Firm and Country Competitiveness: A Meta-analysis of the Porter Hypothesis.” Accessed SSRN. http://ssrn.com/abstract=2692919. [Google Scholar]
- Constantatos, C., and M. Hermann. 2011. “Market Inertia and the Introduction of Green Products: Can Strategic Effects Justify the Porter Hypothesis?” Environmental and Resource Economics 50: 267–284. doi: 10.1007/s10640-011-9471-0 [Crossref], [Web of Science ®], [Google Scholar]
- Crépon, B., E. Duguet, and J. Mairesse. 1998. “Research, Innovation and Productivity: An Econometric Analysis at the Firm Level.” Economics of Innovation and New Technology 7 (2): 115–158. doi: 10.1080/10438599800000031 [Taylor & Francis Online], [Google Scholar]
- Desrochers, P. 2008. “Did the Invisible Hand Need a Regulatory Glove to Develop a Green Thumb? Some Historical Perspective on Market Incentives, Win-Win Innovations and the Porter Hypothesis.” Environmental and Resource Economics 41: 519–539. doi: 10.1007/s10640-008-9208-x [Crossref], [Web of Science ®], [Google Scholar]
- Domazlicky, B. R., and W. L. Weber. 2004. “Does Environmental Protection Lead to Slower Productivity Growth in the Chemical Industry?” Environmental and Resource Economics 28: 301–324. doi: 10.1023/B:EARE.0000031056.93333.3a [Crossref], [Web of Science ®], [Google Scholar]
- Dussaux, D. 2015. “Testing the Porter Hypotheses: Evidence from French Manufacturing Firms.” Paper presented at the economics of innovation, diffusion, growth and the environment conference, London, September 2015. [Google Scholar]
- Gans, J. S. 2012. “Innovation and Climate Change Policy.” American Economic Journal 4 (4): 125–145. [Web of Science ®], [Google Scholar]
- Gourieroux, C., A. Monfort, E. Renault, and A. Trognon. 1987. “Generalized Residuals.” Journal of Econometrics 34: 5–32. doi: 10.1016/0304-4076(87)90065-0 [Crossref], [Web of Science ®], [Google Scholar]
- Heckman, J. 1978. “Endogenous Variables in a Simultaneous Equation System.” Econometrica 46 (4): 931–959. doi: 10.2307/1909757 [Crossref], [Web of Science ®], [Google Scholar]
- Horbach, J., and K. Rennings. 2012. “Environmental Innovation and Employment Dynamics in Different Technology Fields – An Analysis Based on the German Community Innovation Survey 2009.” ZEW Discussion Paper No. 12–006. [Google Scholar]
- Huiban, J. P., C. Mastromarco, and A. Musolesi. 2014. “The Impact of Pollution Abatement Investments on Technology: Porter Hypothesis Revisited.” mimeo. [Google Scholar]
- Jaffe, A. B., R. G. Newell, and R. N. Stavins. 2002. “Environmental Policy and Technological Change.” Environmental and Resource Economics 22: 41–69. doi: 10.1023/A:1015519401088 [Crossref], [Web of Science ®], [Google Scholar]
- Jaffe, A. B., R. G. Newell, and R. N. Stavins. 2003. “Technological Change and the Environment.” Chapter 11. In Handbook of Environmental Economics, Vol. 1. edited by K. G. Mäler and J. R. Vincent, 461–516. Amsterdam: Elsevier Science Publishers. [Crossref], [Google Scholar]
- Jaffe, A., and K. Palmer. 1997. “Environmental Regulation and Innovation: A Panel Data Study.” Review of Economics and Statistics 79 (4): 610–619. doi: 10.1162/003465397557196 [Crossref], [Web of Science ®], [Google Scholar]
- Kemp, R. 2011. “Eco-Innovation: Definition, Measurement and Open Research Issues.” Economia Politica 3: 397–420. [Google Scholar]
- Kretschmer, T., E. J. Miravete, and J. C. Pernías. 2012. “Competitive Pressure and the Adoption of Innovations.” American Economic Review 102 (4): 1540–1570. doi: 10.1257/aer.102.4.1540 [Crossref], [Web of Science ®], [Google Scholar]
- Kriechel, B., and T. Ziesemer. 2009. “The Environmental Porter Hypothesis: Theory Evidence and a Model of Timing of Adoption.” Economics of Innovation and New Technology 18 (3): 41–69. doi: 10.1080/10438590801943235 [Taylor & Francis Online], [Google Scholar]
- Lanoie, P., J. Laurent-Luccheti, N. Johnstone, and S. Ambec. 2011. “Environmental Policy, Innovation and Performance: New Insights on the Porter Hypothesis.” Journal of Economics and Management Strategy 20 (3): 803–842. doi: 10.1111/j.1530-9134.2011.00301.x [Crossref], [Web of Science ®], [Google Scholar]
- Lewbel, A. 2007. “Coherency and Completeness of Structural Models Containing a Dummy Endogenous Variable.” International Economic Review 48 (4): 1379–1392. doi: 10.1111/j.1468-2354.2007.00466.x [Crossref], [Web of Science ®], [Google Scholar]
- Marin, G. 2012. “Do Eco-innovations Harm Productivity Growth through Crowding Out? Results on an Extended CDM Model for Italy.” IMT Lucca EIC working paper series 03. [Google Scholar]
- McFadden, D. L. 1974. “Conditional Logit Analysis of Qualitative Choice Behavior.” In Frontiers in Econometrics, edited by P. Zarembka, 105–142. New York: Academic Press. [Google Scholar]
- Miravete, E., and J. Pernías. 2006. “Innovation Complementarity and Scale of Production.” Journal of Industrial Economics 54: 1–29. doi: 10.1111/j.1467-6451.2006.00273.x [Crossref], [Web of Science ®], [Google Scholar]
- Mohr, R. D. 2002. “Technical Change, External Economies, and the Porter Hypothesis.” Journal of Environmental Economics and Management 43: 158–168. doi: 10.1006/jeem.2000.1166 [Crossref], [Web of Science ®], [Google Scholar]
- Muller, N. Z., R. Mendelsohn, and W. Nordhaus. 2011. “Environmental Accounting for Pollution in the United States Economy.” American Economic Review 101: 1649–1675. doi: 10.1257/aer.101.5.1649 [Crossref], [Web of Science ®], [Google Scholar]
- Palmer, K., W. E. Oates, and P. R. Portney. 1995. “Tightening Environmental Standards: The Benefit-cost or No-cost Paradigm?” Journal of Economic Perspectives 9 (4): 119–132. doi: 10.1257/jep.9.4.119 [Crossref], [Web of Science ®], [Google Scholar]
- Popp, D., R. G. Newell, and A. B. Jaffe. 2010. “Energy, the Environment, and Technological Change.” Chapter 21. In Handbooks of the Economics of Innovation, Vol. 2. Amsterdam: Elsevier Science Publishers. [Crossref], [Google Scholar]
- Porter, M. 1991. “America’s Green Strategy.” Scientific American 264: 168. doi: 10.1038/scientificamerican0491-168 [Crossref], [Web of Science ®], [Google Scholar]
- Porter, M., and C. van der Linde. 1995. “Toward a New Conception of the Environment-Competitiveness Relationship.” Journal of Economic Perspectives 9 (4): 97–118. doi: 10.1257/jep.9.4.97 [Crossref], [Web of Science ®], [Google Scholar]
- Rennings, K., and C. Rammer. 2010. “The Impact of Regulation-driven Environmental Innovation on Innovation Success and Firm Performance.” ZEW Discussion Paper No. 10–065. [Google Scholar]
- Rennings, K., and S. Rexhäuser. 2010. “Long-term Impacts of Environmental Policy and Eco-innovative Activities of Firms.” ZEW Discussion Paper No. 10–074. [Google Scholar]
- Rexhäuser, S., and C. Rammer. 2014. “Environmental Innovations and Firm-profitability: Unmasking the Porter Hypothesis.” Journal of Environmental and Resource Economics 57 (1): 145–167. doi: 10.1007/s10640-013-9671-x [Crossref], [Web of Science ®], [Google Scholar]
- Tamer, E. 2003. “Incomplete Simultaneous Discrete Response Model with Multiple Equilibria.” Review of Economic Studies 70 (1): 147–165. doi: 10.1111/1467-937X.00240 [Crossref], [Web of Science ®], [Google Scholar]
- Train, K. 2003. Discrete Choice Methods with Simulations. Cambridge: Cambridge University Press. [Crossref], [Google Scholar]
- Wagner, M. 2003. “The Porter Hypothesis Revisited: A Literature Review of Theoretical Models and Empirical Tests.” Research Memorandum Center for Sustainability Management (SM), University of Lüneburg. Lüneburg, Germany. [Google Scholar]
- Weiss, J. F. 2015. Essays on Externalities, Regulation, Institutions, and Firm Performance. Jönköping International Business School, Jönköping University, Dissertation No, 102. [Google Scholar]
- Wooldridge, J. M. 2002. Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press. [Google Scholar]
Appendix: The likelihood function for the simultaneous multivariate probit model
In this appendix we describe the derivation of the likelihood function for the simultaneous multivariate probit model, suggested by Miravete and Pernías (2006) and Lewbel (2007), used to estimate the system of equations (4) in our model.
Suppose there are N = 3 types of innovation. The profit function is given by13
Then there are eight possible combinations of the three types of innovation. Thus, for every adopted combination, seven comparisons are at stake [(2N − 1)]. To keep things tractable we will focus on strategy (0,0,0), that is, no innovation at all (thus all comparisons are against zero profits). The computational complexities are due to the derivation of the error support over all possible combinations (strategy choices). The choice of strategy (0,0,0) yields the following set of inequalities:14
In the above inequalities we make a distinction between the deterministic part (indicated by 
) and the stochastic part of the right-hand side (RHS). Notice that, for N = 3, we have one inequality involving
, two involving
and four involving
. Any coherency problem is lifted if we take the minimum of the upper bounds of the inequalities on the right-hand sides.
So we replace the inequalities for 
by
and similarly for the inequalities involving
:
The (joint) probability for the case of no innovation at all is given by:
Similar expressions can be derived for the other combinations of strategies. The expressions involve conditioning upon unobservable variables to enable GHK simulation for evaluating the integration bounds in the likelihood function. For example for choice P(0,0,0), the corresponding part of the likelihood function is given by
where f(.) stands for the density function of the normal distribution.
We estimate this model using maximum simulated likelihood. Methods for estimating such models are readily available (see Cappellari and Jenkins (2003, 2006) and Train (2003)). The number of draws we use in the simulation is 50.