POLICY DESIGN, ECO-INNOVATION AND INDUSTRIAL DYNAMICS IN AN AGENT-BASED MODEL: AN ILLUSTRATION WITH THE REACH REGULATION

Documents de travail GREDEG GREDEG Working Papers Series

Nabila ArfaouiEric BrouillatMaïder Saint-Jean

GREDEG WP No. 2013-22http://www.gredeg.cnrs.fr/working-papers.html

Les opinions exprimées dans la série des Documents de travail GREDEG sont celles des auteurs et ne reflèlent pas nécessairement celles de l’institution. Les documents n’ont pas été soumis à un rapport formel et sont donc inclus dans cette série pour obtenir des commentaires et encourager la discussion. Les droits sur les documents appartiennent aux auteurs.

The views expressed in the GREDEG Working Paper Series are those of the author(s) and do not necessarily reflect those of the institution. The Working Papers have not undergone formal review and approval. Such papers are included in this series to elicit feedback and to encourage debate. Copyright belongs to the author(s).

Groupe de REcherche en Droit, Economie, GestionUMR CNRS 7321

1

POLICY DESIGN, ECO-INNOVATION AND INDUSTRIAL DYNAMICS IN AN AGENT-BASED MODEL: AN ILLUSTRATION OF THE REACH REGULATION

Nabila ARFAOUI,1 Eric BROUILLAT,2 and Maïder SAINT JEAN2*

1GREDEG, University of Nice Sophia Antipolis, 250 rue Albert Einstein, 06560 Valbonne, France

2GREThA, University of Bordeaux, Avenue Leon Duguit, 33608 Pessac, France nabila.arfaoui@gredeg.cnrs.fr, eric.brouillat@u-bordeaux4.fr, maider.saint-jean@u-bordeaux4.fr (*corresponding author, phone number: +33 556 848 640)

Abstract. This article proposes an agent-based model to study the impact of the European regulation REACH on industrial dynamics. This new regulation was adopted in 2006 and establishes a new philosophy of how to design environmental protection and health, especially through the authorization process and the extended producer responsibility. The main contribution of this article is to investigate how different combinations of flexible and stringent mechanisms create the incentives and constraints to shape market selection and innovation. The model outcomes stress that (1) stringency is the most determining feature of policy design (timing is also decisive but it appears to be of secondary importance); (2) technology substitution that brings radical technological change and significant pollution reduction is possible only if regulation is stringent enough but after many sacrifices, especially in terms of market concentration and number of failures; and (3) soft regulation does not lead to technology transition because of weak incentive and selection effects.

Keywords: policy design, technological substitution, agent-based model, REACH

2

1 INTRODUCTION In 2006, after a long legislative battle, the European Union (EU) adopted the REACH regulation (Registration, Evaluation and Authorization of Chemicals)1, one of the most ambitious regulations ever implemented in Europe. This regulation introduces a new legislative philosophy about how to handle chemicals, especially by adopting three essential principles: the “principle of reversal of the burden of proof” from authorities to industry by which manufacturers and importers of chemicals must register each substance2

According to the third principle, public authorization is required for the production and use of chemicals considered to be especially worrisome - so-called substances of very high concern (SVHC) - "with the aim of substituting them" (REACH, article 55). SVHC are to be gradually identified and prohibited for sale and use after a set date (the so-called sunset date) unless the company is granted special authorization.

and assess the health and environmental risks associated to avoid being excluded from the market (“No data, no market”); the extended responsibility to users, which is now closely associated with regulatory compliance, potentially widening the impact of REACH to all the industries; and the authorization and restriction of the most dangerous substances.

3

From the start, REACH was designed to balance environmental objectives with competitive aims, and has the scope to induce the development and adoption of eco-innovation

Thus, in the REACH regulation, the precautionary principle is complemented by a substitution principle.

4

1 Regulation (EC) n°1907/2006 of the European Parliament and of the Council of 18 December 2006 concerning the Registration, Evaluation, Authorisation and Restriction of Chemicals (REACH).

as a side effect of the regulation itself. In the economic literature, many authors have emphasized a positive correlation between innovation and environmental regulation (Jaffe et al., 2003; Porter and van der Linde, 1995; Rennings, 2000). However, eco-innovations cannot

2 This regulation applies to substances used in quantities higher than one ton per year. 3 All requests for authorization must be accompanied by a safety report and an analysis of alternatives. 4 Eco-innovation (also called environmental innovation) can be defined as “the production, assimilation or exploitation of a product, production process, service or management or business methods that is novel to the organization (developing or adopting it) and which results, throughout its life cycle, in a reduction of environmental risk, pollution and other negative impacts of resources use (including energy use) compared to relevant alternatives” (MEI report, 2007, p.7).

3

be considered a systematic response to regulation. Policy design ends up being essential to inducing the development of eco-innovations (Ashford et al., 1985; Hahn, 1989; Jänicke and Lindemann, 2010; Johnstone, 2007). In this respect, a number of criteria, such as stringency, flexibility, timing and credibility, are important factors to consider. REACH seems to fit perfectly in this context and appears to be a privileged object of study to analyze how policy design can stimulate eco-innovations.

This article presents an agent-based model that investigates the key principles and mechanisms of REACH. We show how different combinations of flexible and stringent instruments designed for REACH regulation create incentives and constraints that shape market selection and innovation. In particular, the model is intended to assess to which extent authorization provisions on SVHC may lead to increased moves toward the substitution of those substances through the supply chain.

The article is organized as follows. Section 2 draws on the literature on environmental regulation and eco-innovation to bring into focus the main mechanisms of the REACH regulation that can stimulate innovation and substitution of chemical substances. Section 3 presents the model following the ODD (overview, design concepts, details) protocol (Grimm et al., 2006, 2010). Section 4 presents baseline simulations and examines the impact of regulation on market dynamics by considering various configurations in the policy design. Section 5 provides some concluding thoughts.

2 THE POTENTIAL EFFECTS OF REACH ON INNOVATION

2.1 Environmental regulation and innovation Theoretical and empirical analyses on the relationship between environmental regulation and innovation agree that eco-innovations are essentially “policy-driven” (Jänicke, 2012; Jänicke and Lindemann, 2010). According to Rennings (2000), standard determinants of technological change, that is, demand pull and technology push effects, play an important role in fostering eco-innovation, but they have to be backed by a third determinant called “regulatory push-pull effect.” In fact, unlike traditional innovation, eco-innovation is characterized by a double externality in the innovation and diffusion phases. The lack of incentive and ownership of innovations identified by Arrow (1962) is thus reinforced by this double externality

4

problem. For this reason, public intervention is essential and regulation is a key determinant of eco-innovation.5

We know from Porter and van der Linde (1995) that only well-designed regulations lead to innovation. In particular flexible regulatory policies give firms greater incentives to innovate and thus are better than prescriptive forms of regulation. In many instances, these innovations are likely to more than offset the cost of regulation. So policy design ends up by being essential to inducing the development of eco-innovations (Jänicke and Lindemann, 2010). Ashford et al. (1985) and Hahn (1989) have emphasized that regulators must be careful in the severity, flexibility, and timing of regulation. In the same vein, Jänicke (2012) argue that policy design in particular should be based on ambitious and reliable targets and provide a flexible policy mix supporting the innovation process from invention to diffusion.

The European Commission was very attentive to these criteria in the way REACH was designed. A combination of hard and soft laws was preferred so that REACH relies more on open-ended standards (Fuchs, 2011) that combine different criteria:

- Stringent6

- Reachable: Fuchs (2011) describes REACH as a pragmatic regulation that is both ambitious and realistic in its goals in order to represent a real incentive to undertake innovation.

: the consequences of an incorrect application of the REACH regulation are serious and immediate because they result in exclusion from the market (“No data, no market”).

7

- Flexible: it is shown through open-ended standards, flexible and revisable guidelines, and other forms of “soft law.” It was considered important that the system remain flexible in order to ensure its workability (Fuchs, 2011). Moreover, REACH promotes a mode of governance based on the idea of “self-responsibility.” This approach

5 In several empirical studies, regulation has been identified as an important determinant of eco-innovation: see, for example, Jaffe and Palmer (1997), Lanjouw and Mody (1996), and Popp (2005). For an overview, see the EEA technical report (2011). 6 According to Ashford et al. (1985), a regulation is stringent “either (1) because it requires a significant reduction in exposure to toxic substances, (2) because compliance using existing technology is costly, or (3) because compliance requires a significant technological change” (p. 426). 7 Pragmatism is also shown in other provisions, such as the multiple deadlines for phase-in substances, the collective setting of priorities under the authorization and restriction processes, the various exemptions incorporated in the regulation, and the limited risk assessment requirements for substances placed on the market in proportions of less than 10 tons.

5

involves giving more responsibilities to companies and more flexibility on how to achieve the goals (Fuchs, 2011). In total, these mechanisms can adapt to diversity, tolerate alternative approaches to problem solving, and make it easier to revise strategies and standards in light of evolving knowledge (Scott and Trubek, 2002).

2.2 Authorization process and extended responsibility principle REACH exhibits several mechanisms that can promote innovation in the chemical industry (Berkhout et al., 2003; Eurostat, 2009; Nordbeck and Faust, 2003). In our model, we focus mainly on two crucial innovation-friendly mechanisms: the authorization process and the extended responsibility principle.

The authorization procedure for SVHC is connected to the principle of substitution. The purpose of authorization is to ensure that the risks from SVHC are properly controlled and that these substances are progressively replaced by other substances or technologies, provided they are economically and technically viable. The authorization procedure is based on several steps: (1) identification of substances, (2) request for authorization before the sunset date, (3) granting or refusing authorization, and (4) reviewing authorization.

The granting or refusal of authorization is based primarily on the existence of economically and technically viable alternatives. If there are economically viable alternatives, companies will no longer be allowed to use substances after the sunset date; otherwise, authorizations are granted only if firms prove that they carry out serious analyses of alternatives. In that case, authorizations are granted until a review date by which the holder of the authorization will have to resubmit an application. So firms are encouraged to maintain technology watch on alternatives.

We see that the process of authorization is characterized by different features that combine stringency (target thresholds for techno-economic performances of alternatives), timing (the sunset date), and flexibility (review date), but also pragmatism (cost-benefit analysis) in order to support the innovation process from invention to diffusion. The Eurostat (2009) report about the impact of REACH on innovation in the chemical industry asserts that the process of authorization is an added value in terms of innovation and competitiveness. The report stresses that REACH has had a positive impact on research into new substances because the number of registrations of new

6

substances has increased in line with expectations before REACH was adopted.8

The second innovation-friendly mechanism in REACH is found in the extended responsibility to users, because they are now responsible for the compliance of their production factors to the requirements of the new regulation. According to Wolf and Delgado (2003), innovation in the chemical industry is influenced by many elements, including demand and supplier-client relationships. By extending the principle of responsibility, the aim of REACH is to get actors in the whole production chain to take into account the environmental impact of the activity and to change the demand of downstream users toward environmentally friendlier products.

Typically, REACH must stimulate the development and adoption of alternatives to organic solvents. Solvents have been highly regulated and the use of organic solvents in Europe is ruled by the EC Directive 1999/13/CE on the limitation of emissions of volatile organic compounds due to the use of organic solvents in certain activities and installations (VOC solvents emissions directive). However, the impact of this directive concerning the reduction of volatile organic compound (VOC) emissions in the chemical and metallurgical industries appears to be ambiguous (Belis-Bergouignan et al., 2004). Although a strong and mixed-incentive framework prevails, some sector-specific factors impede technological adoption.9 Since the introduction of REACH, organic solvents are subject to an authorization procedure, which requires producers to develop and adopt alternatives. Biosolvents are good candidates10

8 In a recent report (2013), the European Commission notes that “increased obligations on SVHC through the Candidate listing and Authorisation provisions have led to increased moves toward the substitution of those substances through the supply chain” (p. 4).

to replace organic solvents because they are less toxic, have lower VOC emissions, and are biodegradable (IRSST, 2010). Because of the extended producer responsibility, downstream users are now induced to change their preferences and to transmit their needs to suppliers regarding product quality constraints that must be achieved with alternative solvents.

9 In surface treatment, for instance, the product performance criterion, which is a fundamental one in the supplier-user relationship, explains why solvent substitution cannot extend throughout the metallurgical industry. 10 See http://www.substitution-cmr.fr/. Other alternative solutions to organic solvents may exist, such as solvent-free reactions or environmentally benign solvents [such as supercritical fluids, in particular supercritical CO2 (e.g., decaffeination of coffee beans), ionic liquids (organic salts that are liquid at room temperature), and water]. In our model we will only consider potential substitution of organic solvents by biosolvents.

7

REACH can thus involve innovation in the product chain favored by a partnership and support from users in the experimentation stage of the new processes for the concerned applications. In the rest of the article, we will discuss the case of solvents used for surface treatments as a leading example.

3 THE MODEL

3.1 ABMs and the ODD protocol REACH is a complex regulation that combines several interrelated evolving mechanisms. It aims to “ensur[e] a high level of protection of human health and the environment while enhancing innovation and competitiveness” (REACH, preamble, 1). However, both empirical and theoretical literature about the impact of REACH on innovation is still very scarce because the implementation phase is ongoing and it is difficult to carry out impact assessment studies because of a lack of information11

Assessing the effect of REACH on industrial dynamics requires coping with a large number of decision makers facing a multitude of alternative products. This implies complex evolving interactions and feedbacks between heterogeneous suppliers and clients. According to Hanley et al. (2001), discrete choice modeling seems “to be ideally suited to inform the choice and design of multidimensional policies” (p. 452).

(Béal et al., 2013).

12

For these reasons, we investigate the effect of REACH on industrial dynamics by developing an agent-based model (ABM). ABM is a powerful tool to address complex evolving systems. It simulates explicit behaviors and complex interactions of autonomous and heterogeneous agents in order to generate and study the emergence of coherent system behaviors. The

However, traditional discrete choice models assume a static distribution of decision strategies. Even if recent discrete choice experiments support agent behavior change in response to external pressures, there is still a rather low level of detail regarding the feedback from firms' technological choices. In addition, discrete choice models require extensive data sets to produce significant results, which is a tricky task because of the lack of empirical studies on REACH.

11 Several prospective impact assessment studies were conducted in the pre-regulatory phase, that is, before REACH entered into force in 2007. The European Commission exhibits these studies at http://ec.europa.eu/enterprise/sectors/chemicals/documents/reach/archives/impact-assessment/index_en.htm. 12 See Hoyos (2010) for an extensive survey on discrete choice experiments in an environmental study.

8

objective is to investigate how system-level properties emerge from the adaptive behavior of individuals as well as how, in turn, the system affects individuals. ABM has become increasingly popular in innovation and industrial dynamics studies. It enables the modeling of technological diversity, the behavior of heterogeneous agents, and the change in selection environment that result from policy measures. Several ABM studies focus on transition challenges for cleaner technologies. They investigate the impact of regulation on the emergence and diffusion of environmentally friendly technologies, such as cleaner vehicles (Eppstein et al., 2011; Schwoon, 2006; van der Vooren and Brouillat, 2013), recyclable products (Brouillat and Oltra, 2012), water-saving technologies (Schwarz and Ernst, 2009), energy-saving heating systems (Sopha et al., 2011, 2013), micro-cogeneration technologies (Faber et al., 2010), and so on. However, there is no ABM study that investigates how instruments designed to address REACH regulation affect industrial dynamics. The aim of this article is to fill this gap by presenting an ABM that simulates the development and adoption of alternatives to organic solvents. In this modeling exercise, firm strategies, market structure, and customer choices co-evolve.

The model is used as a learning tool and is not intended for accurate prediction. It aims to provide insights about the directional effect of instruments underlying the authorization procedure of REACH on firm innovation strategy and the associated shift to alternative substances.

In order to present the model we have built, we use the ODD protocol (Grimm et al., 2006, 2010), which provides a standard protocol for describing ABMs in order to make them easier to analyze, understand, and communicate. The protocol consists of structuring the information about an ABM in the same sequence: overview, design concepts, and details (Table 1).

Table 1. The three steps of the ODD protocol

Overview Purpose State variables and scales Process overview and scheduling

Design concepts Design concepts

Details Initialization Input Submodels

9

The logic behind the ODD sequence is first to provide context and general information, followed by more strategic considerations, and finally more technical details. Such a sequence enables the reader to easily absorb information in a progressive way.

3.2 Description of the model We follow the sequence given in Table 1.

3.2.1 Purpose The purpose of our model is to understand how different configurations in the policy design of REACH affect the dynamics of eco-innovation and shape market selection and innovation. In our model we take into account supplier-user interactions because they represent an essential element in the development of new technologies, particularly in the chemical industry. Additionally, some stylized facts that illustrate the competition between organic solvents and biosolvents in the surface treatment activity are considered. The objective is to examine to which extent different combinations of flexible and stringent instruments of REACH can lead to the development and diffusion of alternative solvents (biosolvents).

3.2.2 State variables and scales The model encompasses eight low-level entities: supplier, client, two types of product, and four product characteristics.

Two types of product-related technology may co-exist: technology 1 (T1) is “conventional” technology (e.g., organic solvents) and technology 2 (T2) is “green” technology (e.g., biosolvents). T1 is characterized by an identity number and T2 is characterized by an identity number and initial switching costs. At the start of the simulation run, only T1 exists and is developed by the suppliers.

Suppliers produce and sell products. They are characterized by these variables: identity number, identity of the technology portfolio, R&D share dedicated to T1, adoption threshold to opt for T2, and abandon threshold to leave T1. Suppliers who do not perform well and do not have enough money will exit the market; they are automatically replaced by new entrants. These new entrants have the additional stated variable absorptive capacity, which affects their ability to imitate an incumbent.

Clients buy and use one type of product (T1 or T2) in their production processes. They are characterized by these variables: identity number, identity

10

of the product they have bought, preferences, tolerance degree, reservation price, and minimum product quality.

Each product is described by four attributes similar to Lancaster (1971): technical performance, production costs, VOCs emissions, and biodegradability. Technical performance (X) is related to the solvent power of the technology. Production costs (Cost) depend on the raw materials that are used (petrol vs. biomass) but also on the production facility (traditional refinery vs. biorefinery). Emissions of VOCs represent those gases and vapors containing chemical elements emitted by the solvent. VOCs are emitted during the manufacture, storage, or use of the solvent. The volatility of these chemicals can have serious consequences on health and the environment. Biodegradability (Bio) represents the capacity of air emissions from solvents to degrade readily and to have a short atmospheric lifetime.

Each of these attributes is characterized by a potential of improvement that can be exploited by suppliers according to their R&D and innovation activities. Thus, technical performance is characterized by a maximum limit (Xmax) and production costs, VOC emissions, and biodegradability are characterized by a minimum limit (Costmin, VOCmin, and Biomin). These outer limits are assumed to be different depending on the technology: T1 or T2. In particular, the potential of improvement regarding environmental characteristics is higher for green technology T2 than for conventional technology T1: Covmin T2 < Covmin T1 and Biomin T2 < Biomin T1. We also take into account the technological differences between T1 and T2 in the initial values. Because the green technology T2 is emergent compared to the well-established T1, we assume that T2 has an initial disadvantage in terms of techno-economic characteristics such that initial production costs are higher and initial technical performance is lower than T1 (see Table A.1 in Appendix A).

3.2.3 Process overview and scheduling In the model, one time step represents one period of purchase, and simulations are run for 250 periods. Within each time step, six modules are processed in the following order: purchase, budget, entry/exit, technology portfolio, R&D watch and innovation, and rebuy.

Purchase (Fig. 1) depends on the highest score obtained by a product on the characteristic that is preferred by each client, provided that economic and/or technical constraints are also satisfied (reserve price and reserve

11

technical quality). Once a product is selected by a client, the corresponding supplier registers a sale.

Budget of each supplier (Fig. 1) takes into account the R&D expenses and the profits derived from the sales.

Within the exit/entry module (Fig. 1), each supplier with a negative budget exits and is replaced by a new firm so that a constant number of suppliers is observed over the whole time period (van der Vooren and Brouillat, 2013). Each new entrant will be able to copy with more or less success (specific absorptive capacity) an installed firm with high visibility (high market shares). By allowing automatic entry after each exit we force the system to experience permanent competition over time. Doing so we avoid the possibility of ending up with oligopolistic situations in which strategic behaviors and interdependent decisions are important to consider but are out of the scope of our research.

Fig. 1. The execution order of the REACH model processes: purchase, budget, and exit/entry modules

Technology portfolio (Fig. 2) enables a supplier to adopt T2 or not on the one hand and to keep or abandon T1 on the other hand so that in the end the supplier’s portfolio can be constituted by T1 and/or T2.

N

Y

Purchase Sales Profit Total budget

Budget < 0?

Technology portfolio

Exit

Entry

12

Fig. 2. The execution order of the REACH model processes: Technology portfolio module

R&D watch and innovation (Fig. 3) enables suppliers to improve the characteristics of their product. R&D watch concerns only suppliers who have

N

N

N

Y

Y

Budget > Switching costs?

No adoption of T2 Adoption of T2

R&D watch on T2

Is it worth still developing T1?

TurnoverT2/ Total turnover

> abandon threshold

Leave T1 Keep T1

Technology portfolio =

T2

Technology portfolio =

T1 + T2

Technology portfolio = T1

Innovation

Technology portfolio

Y

N Y

Is T2 adopted?

Adoption index of T2 < adoption threshold?

13

not yet adopted T2. They are searching for substitutes and thus accumulate knowledge about T2. Innovation activities may then involve improvements on T1 and/or T2, depending on the technology portfolio of each supplier.

Fig. 3. The execution order of the REACH model processes: R&D watch and innovation module

Rebuy (Fig. 4) enables each client to compare the performance achieved by its current supplier on its preferred characteristic with the best industry performance. If the current supplier does not underperform, the client keeps the same supplier; otherwise, the client switches to a new supplier and selects another one with the purchase module.

N

Y

Y

N

R&D watch on T2

Success?

Lower switching costs for T2

Higher knowledge on T2

Innovation Success?

Innovative outcome: ∆X; ∆Cost; ∆VOC;

∆Bio

Rebuy

New product characteristics

New maximum industry performance

14

Fig. 4. The execution order of the REACH model processes: rebuy module

3.2.4 Design concepts Our model draws on basic principles developed by the evolutionary theory of technological change (Chiaromonte and Dosi, 1993; Malerba et al., 1999; among others). Thus, a strong emphasis is put on dynamics, changing structures, and disequilibrium processes with an evolutionary perspective.

According to the evolutionary approach, bounded rationality characterizes economic agents that have limited cognitive capacities to collect and treat information. Suppliers seek increased market share from innovation whereas users seek to select the best product according to their preference and requirement criteria. Individuals cannot predict the future conditions they will experience; they are myopic and their decisions follow some routines and a satisficing principle rather than a maximizing one. In our model, suppliers make their decisions regarding their technology portfolio by considering specific thresholds that reflect bounded rationality. A similar procedure is applied by clients when deciding to keep or leave their supplier. In addition, complex, demanding, and imperfect information lead rationally bounded clients to select only one product characteristic among four to make their purchase decision.

The decision rules are adaptive, which means the agents adapt according to their performance and their past experience. In our model, suppliers adapt their strategy of R&D investment based on sales achieved in the past, and

N Y

Rebuy Supplier’s performance ≤

Best industry performance*tolerance?

Keep the same supplier

Switch to a new supplier

Selection of a new supplier on the

market

Purchase

15

customers adapt their requirement levels according to supplier performance. Adaptation is thus modeled through the change in threshold levels used in the decisions of agents.

Given that decision rules are agent-specific, heterogeneity among individuals is a core aspect of such an evolutionary theory. Interactions between heterogeneous agents generate permanent diversity. Industry dynamics emerge from the behavior of the heterogeneous individuals; on the one hand, each supplier decides on the technology to develop or adopt (T1 and/or T2) and then the product characteristics to improve; on the other hand, each client decides on the product to purchase and whether her supplier should be kept or replaced by another one. Each entry of a new firm replacing a failing supplier corresponds to the creation of a new individual endowed with attributes based on the industry average performance.

Innovation is an endogenous and uncertain process. Indeed, firms cannot know perfectly the results of their R&D activity. That is why we model a stochastic process of innovation. Other stochastic processes are included in our model: most behavioral parameters are randomly drawn, the accumulation of knowledge that results from technology watch on T2 is stochastic, and the selection of a supplier by a client is based on a purchase probability (reflecting errors or imperfect information).

Regarding innovation, a distinction is implicitly made between incremental and radical innovation. Incremental innovation allows small changes whereas radical innovation leads to a technological jump with significant costs and experience effects. In our model, the adoption of T2 brings radical changes that are materialized by high switching costs.

3.2.5 Initialization At the start of a simulation run, there are 10 suppliers and 200 users. Some initial values of the state variables are randomly chosen in an admissible range. Others are scale parameters that have been set to calibrate the model plausibly. Regarding product characteristics (VOC emissions, biodegradability, costs, and technological performance), initial values are based on data to account for the difference in order of magnitude between organic solvents and biosolvents.13

13 See http://action4p.net/solvants,

16

3.2.6 Input data The model does not use input data to represent time-varying processes.

3.2.7 Submodels Here, we specify the equations and the assumptions underlying them to better understand the modules listed in the process overview and scheduling (see 3.2.3).

Purchase: the purchase process is composed of the three following steps: 1. Each client (j=1,…,200) chooses one product characteristic randomly.

The probability of a characteristic being chosen is proportional to the client-specific preferences a, b, c, d.

2. The client scans all the products (k=1,2) marketed by each supplier (i=1,…,10) and gives them a score proportional to the selected characteristic in step 1. We consider the following score functions (U) when the selected product characteristic is respectively technological performance (X), price (P), VOC emissions (VOC), or biodegradability (Bio):

𝑈𝑘,𝑖,𝑡𝑗 = �𝑋𝑘,𝑖,𝑡 − 𝐴� × �𝑀𝑠𝑖,𝑡 + 𝑢(0,0.1)�𝑒 (1)

𝑈𝑘,𝑖,𝑡𝑗 = �𝐵 − 𝑃𝑘,𝑖,𝑡� × �𝑀𝑠𝑖,𝑡 + 𝑢(0,0.1)�𝑒 (1’)

𝑈𝑘,𝑖,𝑡𝑗 = �𝐶 − 𝑉𝑂𝐶𝑘,𝑖,𝑡� × �𝑀𝑠𝑖,𝑡 + 𝑢(0,0.1)�𝑒 (1’’)

𝑈𝑘,𝑖,𝑡𝑗 = �𝐷 − 𝐵𝑖𝑜𝑘,𝑖,𝑡� × �𝑀𝑠𝑖,𝑡 + 𝑢(0,0.1)�𝑒 (1’’’)

A, B, C, and D are technical parameters used only to avoid negative terms in the score calculation. u(0,0.1) is drawn from a uniform distribution with values between 0 and 0.1 to avoid U = 0 when the market share (Ms) of the product is null. The parameter e can be interpreted as a bandwagon effect (Leibenstein, 1950) reflecting imitation behaviors.14

3. The client randomly selects one product. The probability of a product being chosen is proportional to its score U. Each client is also supposed

http://www.ineris.fr/rsde/fiches/fiche_dichloromethane_2005.pdf, http://www.irsst.qc.ca/media/documents/PubIRSST/B-079.pdf 14 There is information asymmetry regarding supplier performance. Consequently, clients refer to the behavior of other customers buying similar goods (Cowan et al., 1997). The clients use the market share of the firm (Ms) because it reflects the relative reputation of the supplier; the market share as an indicator provides information on the quality of the product observed by customers who have already adopted it.

17

to be limited by economic and technical constraints. So we assume a reserve price and a minimum technical quality requirement for each client. If the selected product does not satisfy one of these constraints, it is discarded and the client selects another product following the same procedure. If there is no product that satisfies these constraints, the client does not buy and own any product during the period.

The price P is deduced from the production cost by applying a mark-up rate (µ)15

𝑃𝑘,𝑖,𝑡 = (1 + 𝜇) × 𝐶𝑜𝑠𝑡𝑘,𝑖,𝑡 (2)

:

Budget: the budget B is determined by the residual budget from the

previous period, the profits (π), and the R&D expenses (RD): For typical suppliers:

𝐵𝑖,𝑡 = 𝐵𝑖,𝑡−1 + 𝜋𝑖,𝑡−1 − 𝑅𝐷𝑖,𝑡−1 (3)

For new T2 adopters:

𝐵𝑖,𝑡 = 𝐵𝑖,𝑡−1 + 𝜋𝑖,𝑡−1 − 𝑅𝐷𝑖,𝑡−1 − 𝑆𝐶𝑖,𝑡−1 (3’)

where SC are the switching costs resulting from the adoption of T2. The profits are determined as follows:16

𝜋𝑖,𝑡 = �𝜇 × 𝐶𝑜𝑠𝑡𝑖,𝑡 × 𝑄𝑖,𝑡� − 𝐹𝐶 (4)

where Q is the total number of products sold by the firm and FC are the fixed costs.17

Entry/exit processes: firms with a negative budget B go bankrupt and disappear from the market. When one firm exits the market, we assume that a new firm enters so that the number of firms in the industry is kept constant.

Entry occurs with a new firm imitating an existing one. This choice is based on probabilities proportional to the installed firms’ market shares. The new firm copies the technology portfolio and the product characteristics of the imitated firm. We assume that the new firm has an absorptive capacity that

15 For simplicity, µ is supposed to be constant and identical for every firm. 16 After simplification of π = P × Q - TC, where TC are the total costs (variable costs and fixed costs). 17 Fixed costs are supposed to be identical for all the firms for simplicity reasons.

18

enables it to copy the attributes of the imitated firm in a range of [0.8;1.2]. This enables the new entrant to underperform or inversely to overperform in comparison with the imitated firm.18

Technology portfolio: every period, firms examine the possibility of changing their technology portfolio. They compare an adoption index with a firm-specific threshold. The adoption index is as follows:

𝐴𝑑𝐼𝑛𝑑𝑒𝑥𝑖,𝑡𝑇2 = 𝐾𝑖,𝑡−1 × (𝑀𝑠𝑡−1𝑇2 ) (5)

where MsT2 represents the total market share of T2. K stands for the cumulated knowledge stock on T2 derived from the firm’s technological watch. Thus the probability of adopting T2 positively depends on the cumulated stock of knowledge on T2 and also on how T2 has diffused in the market.

The decision to adopt T2 follows a two-step procedure. First, the firm compares its adoption index with an adoption threshold under which the firm will not adopt T2. If its adoption index is above the threshold, then the second step determines if the firm has a sufficient budget to bear the switching costs related to T2.

Firms that decide to adopt T2 can continue to produce and sell T1; they will have a technology portfolio constituted of T1 and T2. However, firms can decide to abandon T1 and focus only on the development of T2. Here we assume that each firm calculates the share hold by T2 in its total turnover:

𝑆ℎ𝑎𝑟𝑒𝑇2𝑖,𝑡 =𝑃𝑖,𝑡−1𝑇2 ×𝑄𝑖,𝑡

𝑇2

𝑇𝑢𝑟𝑛𝑜𝑣𝑒𝑟𝑖,𝑡−1 (6)

The share is compared with a firm-specific threshold that represents the more or less risky attitude of each firm toward T2. The higher the share or the lower the threshold, the higher the likelihood to bet only on T2 and to abandon T1.

Innovation process and green technological watch: each period, every firm

can improve the product performance in its portfolio by carrying out R&D and innovation activities.

18 The initial budget (B) and the initial fixed costs (FC) of the new firm are set in the same way as for the firms created at the start of a simulation run. The knowledge stock (K) and the switching costs (SC) of the new firm are functions of the industry average.

19

Every firm will allocate a certain proportion δ of its budget to R&D activities:

𝑅𝐷𝑖,𝑡 = 𝛿 × 𝐵𝑖,𝑡 (7)

Then, each firm is assumed to split its global R&D budget between both technologies T1 and T2:

𝑅𝐷1𝑖,𝑡 = 𝛿1 × 𝑅𝐷𝑖,𝑡 (8)

𝑅𝐷2𝑖,𝑡 = (1 − 𝛿1) × 𝑅𝐷𝑖,𝑡 (9)

where δ1 is the share of total R&D allocated to T1. For firms developing only T2, δ1 = 0. For firms developing both T1 and T2, δ1 = 0.5. For firms developing only T1, δ1 = 0.5 because they devote the other part to a technological watch on T2 (in that case, RD2 = RDwatch).

Green technology watch follows a stochastic process. Success occurs if the following condition is satisfied:

1 − 𝑒−𝛼𝑤×𝑅𝐷𝑤𝑎𝑡𝑐ℎ𝑖,𝑡 ≥ 𝑢(0,1) (10)

where αw is a scale parameter determining the speed at which the level of the current R&D expenditure allows knowledge accumulation. RDwatch represents R&D expenses allocated to a technological watch on T2. u(0,1) is a uniform random value selected between 0 and 1. The closer to 1, the more difficult it is to satisfy the condition (10) with a given R&D investment.

In case of success, new knowledge on T2 is accumulated and the switching costs linked to the potential adoption of T2 decrease:

𝐾𝑖,𝑡 = 𝐾𝑖,𝑡−1 + 𝛼𝐾 × 𝑢(0,1) × �𝐾𝑚𝑎𝑥 − 𝐾𝑖,𝑡−1� (11)

𝑆𝐶𝑖,𝑡 = 𝑆𝐶𝑖,𝑡−1 − 𝛼𝑠𝑐 × 𝑢(0,1) × (𝑆𝐶𝑡−1 − 𝑆𝐶𝑚𝑖𝑛) (12)

where αK and αSC are scale parameters and Kmax and SCmin are respectively maximum and minimum limits for K and SC.

The innovation process is similar to the previous procedure. Two steps are considered for each product characteristic (X, Cost, VOC, and Bio):

1. The innovation probability depends on the R&D investment allocated to the technology. Success of innovation depends on the following condition:

1 − 𝑒−𝛼𝐼×𝑅𝐷𝑘,𝑖,𝑡 ≥ 𝑢(0,1) (13)

20

where α1 represents the speed of the innovation process and RD the firm R&D expenses devoted to T1 or T2.

2. If condition (13) is satisfied, the new value for, respectively, X, Cost, VOC, or Bio is:

𝑋𝑘,𝑖,𝑡 = 𝑋𝑘,𝑖,𝑡−1 + 𝛽1 × 𝑢(0,1) × �𝑋𝑚𝑎𝑥 − 𝑋𝑘,𝑖,𝑡−1� (14)

𝐶𝑜𝑠𝑡𝑘,𝑖,𝑡 = 𝐶𝑜𝑠𝑡𝑘,𝑖,𝑡−1 − 𝛽2 × 𝑢(0,1) × �𝐶𝑜𝑠𝑡𝑘,𝑖,𝑡−1 − 𝐶𝑜𝑠𝑡𝑚𝑖𝑛� (15)

𝑉𝑂𝐶𝑘,𝑖,𝑡 = 𝑉𝑂𝐶𝑘,𝑖,𝑡−1 − 𝛽3 × 𝑢(0,1) × �𝐶𝑜𝑣𝑘,𝑖,𝑡−1 − 𝐶𝑜𝑣𝑚𝑖𝑛� (16)

𝐵𝑖𝑜𝑘,𝑖,𝑡 = 𝐵𝑖𝑜𝑘,𝑖,𝑡−1 − 𝛽4 × 𝑢(0,1) × �𝐵𝑖𝑜𝑘,𝑖,𝑡−1 − 𝐵𝑖𝑜𝑚𝑖𝑛� (17)

where β1, β2, β3, and β4 are scale parameters and u(0,1) is a uniform random value selected between 0 and 1, which reflects the efficiency of the R&D activity and thus affects the innovative outcome. The last term of the equation represents the distance to the technological frontier associated with each product characteristic.19

Rebuy: each client j is assumed to use one single product at the same time and to renew its purchase every period. The client can choose to keep the same supplier or to leave the supplier following the two-step process:

1. Each client randomly chooses one product characteristic. The probability of a characteristic being chosen is proportional to the client-specific preferences a, b, c, d.

2. When the selected characteristic in step 1 is respectively X, P, VOC, or Bio, the client will make the following decision: if 𝑋𝑘,𝑖,𝑡

𝑗 ≤ 𝑚𝑎𝑥𝑋𝑘,𝑡 ×

𝑢(0.8,1.2), 𝑃𝑘,𝑖,𝑡𝑗 ≥ 𝑚𝑖𝑛𝑃𝑟𝑖𝑐𝑒𝑘,𝑡 × 𝑢(0.8,1.2), 𝑉𝑂𝐶𝑘,𝑖,𝑡

𝑗 ≥ 𝑚𝑖𝑛𝑉𝑂𝐶𝑘,𝑡 ×

𝑢(0.8,1.2), or 𝐵𝑖𝑜𝑘,𝑖,𝑡𝑗 ≥ 𝑚𝑖𝑛𝐵𝑖𝑜𝑘,𝑡 × 𝑢(0.8,1.2), the client leaves her

current supplier and chooses another one through the purchase procedure; otherwise, the client keeps the same supplier. maxX, minPrice, minVOC, and minBio are the best industry performances achieved at the current period. u(0.8,1.2) is a random number from a uniform draw allowing a certain zone of tolerance according to which a client may accept variation within a range of performance.

19 When the level of a given product characteristic comes closer and closer to the limit of what is achievable with the considered technology, a given R&D expenditure will achieve less and less progress (lower technological opportunities and R&D decreasing returns).

21

3.3 The regulation module Here we specify how we have formally taken into account the two main mechanisms underlying REACH: authorization process and extended producer responsibility.

3.3.1 The authorization process The purpose of the authorization process (Fig.5) is to progressively replace SVHC by other substances that are economically and technically viable. Two action leverages are considered in our model.

Fig. 5. Flow diagram for authorization

First, target-thresholds for techno-economic performances of alternative solutions are incorporated. The mechanism is the following: at the sunset date, if T2 reaches both thresholds of technical and economic performance (X* and Cost*), then the public authorities can consider that viable solutions exist and

N

Y

Y N

Authorization is granted. Possibility to use the substance for a

certain period of time Revision date is granted

R&D watch on alternatives?

STOP The substance is not authorized

and cannot be placed on the market or used after the sunset

date.

Suitable alternatives techno-economically

available?

Revision of authorization

22

can prohibit the use of T1.20 On the contrary, if T2 does not reach the target thresholds, public authorities can consider that there are no techno-economically viable alternatives. In that case, authorizations are granted and firms can keep on marketing T1 after the sunset date, but only if they prove that they carry out serious analyses of alternatives providing information on their R&D activity. In fact, in our model the budget allocated to the R&D watch on T2 is used to check whether the firm is searching for new alternatives. Below a certain threshold, authorization will not be granted, T1 is forbidden for the considered firm. Above the threshold, authorization is granted, and the firm can keep on developing and marketing T1 until the next revision date. The threshold for R&D watch depends on the average R&D watch performed in the industry multiplied by a parameter (αRDwatch*) reflecting the severity of the regulation.21

The timing of regulation is the second action leverage for public authorities. An early sunset date associated with close revision dates can be considered strict. On the contrary a late sunset date and distant revision dates impose softer constraints. In order to take timing into account, we assume that the probability of adopting T2 is modified as follows:

At the revision date, a similar sequential checking is made.

𝐴𝑑𝐼𝑛𝑑𝑒𝑥𝑖,𝑡𝑇2 = 𝐾𝑖,𝑡−1 × (𝑀𝑠𝑡−1𝑇2 ) × ��1 + αRi � × tT� (5’)

with T the sunset date (T=Tsunset) if t≤Tsunset, otherwise T is the revision date (T=Trevision). αR is a parameter reflecting the credibility that the firm confers to regulation.22

3.3.2 Extended responsibility

Equation (5’) implies that regulation positively influences the adoption of T2: the earlier and closer to the sunset date, the higher the adoption index; the more frequent the revision of authorization, the higher the adoption index; the higher the credibility given to regulation, the higher the adoption index.

By extending the responsibility principle, REACH aims to change the demand of downstream users of chemical products toward less toxic and harmful

20 In the model we compare the average technical performance of T2 with X* and the average cost of T2 with Cost*; if the average technical performance of T2 is above X* and the average cost of T2 is below Cost*, T1 is prohibited. 21 The closer to 1, the stricter the regulation; the closer to 0, the softer the regulation. 22 αR ∈ [0; 1].

23

substances. In order to take into account such a change in our model, we will now consider that the technology portfolio held by suppliers matters in the clients’ decisions such that the following become true:

- The score (U) given to a product (purchase module) will tend to decrease for suppliers for whom the portfolio is constituted only by technology T1; in that case U is weighted by a factor ��1 − αR

j � × tT�,

where αR is a parameter reflecting the credibility that the client gives to regulation and T represents alternatively the sunset date or the revision date.

- The decision made by a client to leave her current supplier (rebuy module) will be subject to a probability of defection based on the factor ��αR

j � × tT� when the supplier’s portfolio is constituted only by T1.

According to these changes, the closer the sunset date or the revision date, the lower the score in the purchase module for suppliers holding a portfolio with only T1 and the higher the defection probability of their current clients in the rebuy module.

4 RESULTS In this section we first expose the simulation protocol we have followed to study and analyze the results of the baseline simulations and the consecutive impact of regulation on industrial dynamics.

4.1 The simulation protocol Based on several indicators characterizing industrial dynamics, we compare a benchmark configuration without regulation with two other configurations that include the regulation module.

4.1.1 Main indicators characterizing industrial dynamics The following indicators are used to exhibit the main characteristics of industrial dynamics:

- The inverse Herfindahl-Hirshman index of concentration (1/HHI),23

- The cumulated number of failures, which takes into account the number of exiting firms in each period. In our model, the higher the

which is valued between 1 (monopoly) and N (atomicity). The higher the 1/HHI, the higher the degree of competition, and inversely.

23 HHI is the sum of the squares of the firms’ market shares.

24

number of failures, the higher the number of new entrants that come and replace the exiting firms. Three main causes of failures can occur: (1) lack of budget; (2) ban of T1, which forces firms with portfolio T1 to exit; and (3) lower size of the market when clients are not satisfied by the price-quality ratio of T2.

- The respective market share of T1 and T2. - A global environmental indicator that traces back the stock of VOC

emissions at the industry level. We consider the following equation:

𝑉𝑂𝐶𝑆𝑡𝑜𝑐𝑘𝑡 = 𝑉𝑂𝐶𝑆𝑡𝑜𝑐𝑘𝑡−1 − 𝐴𝐵𝑆 + ∑ 𝑉𝑂𝐶𝑘,𝑡𝑗 × 𝐵𝑖𝑜𝑘,𝑡

𝑗𝑀𝑗=1 (18)

where ABS stands for the assimilative capacity of an ecosystem receiving pollution (VOC emissions) during each period.24

4.1.2 Configurations under examination

It is set exogenous and constant over time. According to equation (18), the current stock of VOC emissions depends on the previous stock of VOCs (the “history” of pollution flows) minus assimilated emissions by the ecosystem plus the current flow of emitted VOCs. Such a global environmental indicator enables grasping the ability of the industry to decrease its VOC emissions over time. This decrease in VOCs can result from two effects: a qualitative effect through innovation (decrease in the VOC and/or bio characteristics) and a quantitative effect through lower market size in cases when clients cannot afford a product (too costly and/or too low quality).

As presented previously (see 3.2.5), initial conditions for the benchmark configuration are based on empirical data differentiating solvents and biosolvents regarding the four considered characteristics. Other parameters have been set to calibrate the model in a consistent way.

Batches of 500 simulations are carried out in order to cope with stochasticity. For each indicator, the average over the 500 simulations is computed at different time steps: 1, 50, 100, 150, 200, and 250.

Concerning the initial conditions for the regulation module, we study two opposite configurations (Table 2) depending on the level of the target thresholds (X*, Cost*, and R&Dwatch*) and on the timing of regulation (Tsunset and Trevision): the low-stringency configuration combine soft target

24 The assimilative capacity of an ecosystem receiving pollution can be defined as the ability “to receive a determined level of residues, to degrade them and to convert them in non- damaging and even beneficial products” (Pearce and Turner, 1990, p. 38).

25

thresholds and late sunset and rare revision dates; the high-stringency configuration combine tight target thresholds and early sunset and frequent revision dates.25

Table 2. Policy variables and parameters in the REACH model

Target Thresholds Timing Techno-economic

performances R&D watch Sunset date Revision date

X* Cost* αRDwatch*

Low stringency High (93.2)

Low (0.66)

Low (0.1)

Late Tsunset=150

Rare ∆Trevision=10

High stringency Low (38.8)

High (2.74)

High (0.9)

Early Tsunset=50

Frequent ∆Trevision=2

Results are presented in the following section and Monte Carlo simulations

have subsequently been carried out to give robustness to the results.

4.2. The impact of regulation on industrial dynamics The following figures depict the evolution of each indicator in the different cases: the benchmark (black line), the less stringent scenario (grey dotted line), and the most stringent scenario (gray full line).

The benchmark is characterized by the following properties: - Slight decrease in market concentration (Fig. 6) due to a “success

breeds success” process: initially firms exploit the technological potential of T1, which is especially important regarding environmental dimensions. Thus a few firms perform well on T1 and gain market shares because of incremental environmental innovations.

- The stronger the market concentration, the faster the technological change in particular environmental innovations.

- Early established dominant positions are challenged by the introduction and development of T2 around period 36 on average (Fig. 7): clients require higher environmental performance and they make more defections through the rebuy module.

25 The values for techno-economic performances are chosen in relation to the size of the techno-economic potential, which can be covered by innovators. In the low stringency configuration, 90% of the potential must be covered, whereas in the high stringency configuration 10% of the potential needs to be covered.

26

- However T2 does not diffuse significantly and remains doomed to a niche (4.7% on average at t = 250). As a consequence the VOCs stock keeps on increasing, even if its growth rate decreases (Fig. 8).

Fig. 6. Evolution of the inverse HHI (average for 500 simulations)

Fig. 7. Evolution of T2 market share (average for 500 simulations)

1 2 3 4 5 6 7 8 9

10

0 50 100 150 200 250

InvH

HI

Period

benchmark early_freq_H late_rare_L

0%

20%

40%

60%

80%

100%

0 50 100 150 200 250

Mar

ket s

hare

T2

Period

benchmark early_freq_H

late_rare_L 50%market share

27

Fig. 8. Evolution of the global stock of VOCs (average for 500 simulations)

When introducing regulation, we can notice that market concentration tends to increase all the more strongly when regulation is more stringent (Fig. 6). This is due to the pressure put on firms with only T1 in their portfolio. The closer the regulatory threat, the higher the defections of clients toward suppliers providing only T1, but also the more drastic the selection when technology T1 is prohibited.

We thus observe that only the most stringent scenario allows domination of T2 due to an early ban of T1, whereas the less stringent scenario results in higher market shares for T2 (20% on average at t = 250) but still coexisting with T1 (Fig. 7).

As a consequence, the global stock of VOCs (Fig. 8) shows a kind of inverted U curve in the most stringent scenario as VOC emissions rise in the first place and then fall with the emergence and diffusion of radical environmental technology T2. By contrast, the less stringent scenario exhibits small differences with the benchmark until period 150 before undergoing deceleration and decrease of VOCs. In that case, the regulatory threat is so distant in the future that it lacks efficiency to stimulate radical technological change. Firms specialized in T1 are disturbed by regulatory pressure during the whole time period without yet experiencing complete prohibition of T1.

Regarding the evolution of failures (Fig. 9), regulation tends to increase the number of failures. This is all the more significant in the most stringent

0,0 1,0 2,0 3,0 4,0 5,0 6,0 7,0 8,0 9,0

0 50 100 150 200 250

VOCS

tock

Mill

ions

Period

benchmark early_freq_H late_rare_L

28

scenario when a combination of different effects happens: (1) forced exit due to an insufficient R&D watch on T2 or to a ban of T1 or (2) insufficient sales (and thus insufficient budget) due to a decrease in market size that results from demand rationing because many clients are not satisfied any more with the new dominant design (high price and/or low quality of T2).

Fig. 9. Evolution of failures (average for 500 simulations)

Our simulation results are in line with empirical studies depicting that the efficiency of policy depends primarily on the way they are designed and applied (Ashford et al., 1985; Blazejczak et al., 1999; Kemp and Pontoglio, 2011; Norberg-Bohm, 1999a, 1999b). More precisely our model shows that technological substitution (from T1 to T2) requires open-ended standards based predominantly on the technology within reach of a vigorous R&D effort (strong target thresholds) combined with a cut-off date (sunset date). In our model, stringent target thresholds imply inflexible deadlines because they lead to a high probability of a T1 ban on early checking dates. Such a policy design appears to be efficient in terms of radical innovation and thus in terms of pollution reduction. On the contrary, flexible deadlines combined with lenient techno-economic performances are not suited to promote radical innovation. In this soft scenario, the failure to ban T1 inhibited development of T2. Suppliers are able to comply with the standard simply by improving the existing technology T1 and by carrying incremental environmental innovation. By failing to prohibit T1, regulation effectively inhibits

0

5

10

15

20

25

30

0 50 100 150 200 250 Cum

ulat

ed n

umbe

r of f

ailu

res

Period

benchmark early_freq_H late_rare_L

29

technology substitution and radical environmental innovation. We have here an illustration of what Ashford et al. (1985) have empirically observed: “contrary to the widely held belief that too stringent a regulation inhibits innovation, in some cases a standard not stringent enough may inhibit innovation” (p. 464).

What our model shows further are the economic consequences in terms of market concentration and number of failures that go with technological transition. Technology substitution is possible only if regulation is stringent enough but after many sacrifices. The selective pressure related to a stringent regulation is able to bring win-win effects26

4.3. Sensitivity analysis

(Porter and van der Linde, 1995) for a small group of surviving firms but is also conducive to a significant number of losing firms. The successful firms prove to be diversified in their technology portfolio (T1 and T2) or be green risk-taking pioneers. Meanwhile demand is affected by too stringent a regulation. Indeed severe performance targets force disruptive technology to enter the market (T2) and provide a product with lower performance than the incumbent (in terms of price and technical performance), which exceeds the requirements of certain clients, thereby temporarily decreasing the size of the market until technology improvements take place. These results stress the conflict between environmental and techno-economic dimensions related to radical environmental change. From a policy point of view, these findings show the importance of finding the right balance between incentive to innovate and disruptive selection to encourage technology transition. There are necessary trade-offs to support technological transition and substitution of chemical substances.

In this section, we perform a sensitivity analysis of results to regulation parameters to give more robustness to the previous results and exhibit additional findings. The analysis is based on a set of 10,000 simulations carried out with a Monte Carlo procedure.27

26 In our model win-win effects are materialized by higher market shares and environmental performances.

We process results with regression trees. A regression tree (Venables and Ripley, 1999) establishes a hierarchy between independent variables using their contributions to the

27 This methodology enables us to run a high number of simulations with a random setting of the values of the parameters (see Table B.1 in Appendix B).

30

overall fit (R²) of the regression. The tree depicts a hierarchical sequence of conditions on the variables of the model: the higher the role of a condition in the classification of the observed case, the higher its status on the tree.28

When comparing extreme branches of each tree, one can notice great differences. This confirms our previous finding that regulation outcomes depend dramatically on policy design. The market share of T2 is only 12% (Fig. 10) and the probability of T1 prohibition is less than 3% (Fig. 11) when stringency

29

is low. This market share is five times higher when target thresholds are strict and the probability of banning T1 rises to 70%. The combination of stringent target thresholds and early sunset dates can reduce pollution by 41% (Fig. 12), whereas soft regulation leaves it unchanged. Similar conclusions can be made about failures (Fig. 13): increasing regulation stringency will multiply failures by 2.26 to 3.53, depending on a late or early sunset date.

Fig. 10. Regression tree of T2’s market share

28 For each condition, the left branch shows the cases for which the condition is true and the right branch indicates cases compatible with the complementary condition. If we consider, for instance, the tree in Fig. 10, on the left branch, we have all observations for which X* ≥ 45.46. On the right branch, we have all observations for which X* < 45.46. When X* < 45.46 and Cost* < 1.452, the expected value for the market share of T2 is 22.15%, and we have n = 756 observations corresponding to this case. 29 Stringency is reflected by X*, Cost*, and αRDwatch*: the lower the X* and/or the higher the Cost* and/or the higher the αRDwatch*, the more stringent the regulation.

0.1206 n = 8052

0.2215 n = 756

X* ≥ 45.46

0.6095 n = 1192

Cost* < 1.452

yes

yes

no

no

31

Fig. 11. Regression tree of probability of T1 prohibition

Fig. 12. Regression tree of global stock of VOCs

Fig. 13. Regression tree of cumulated number of failures

0.02779 n = 7556

Cost* < 1.472

X* ≥ 40.53

X* ≥ 48.86

0.1627 n = 1002

0.3922 n = 696

0.7011 n = 746

yes

no

yes

yes

no

no

X* < 40.69 yes

no

αRDwatch* ≥ 0.5571

7.292e+06 n = 3894

8.005e+06 n = 4848

yes

no

Tsunset < 109.5

4.719e+06 n = 265

6.855e+06 n = 305

yes

yes

no

no

7.288e+06 n = 688

Cost* ≥ 1.834

X* ≥ 42.85

6.192 n = 4665

14.71 n = 880

14 n = 24.07

αRDwatch* < 0.5566

21.84 n = 2048

Tsunset ≥ 116.5 yes

yes

yes

no

no

no

32

Fig. 14. Regression tree of inverse HHI

Trees highlight that stringency is the most determining feature of policy design. Timing30 is also decisive but it appears to be of secondary importance. Tsunset will significantly affect the stock of VOCs and the number of failures only when regulation is stringent.31

Trees confirm the inevitable trade-offs between environmental and techno-economic dimensions related to radical environmental change. They show that technological transition is fostered by stringent target thresholds for techno-economic performances (Fig. 10). The probability of a T1 ban is much higher when target thresholds are tight: more than 70% with X* < 40.53 and Cost* ≥ 1.472 (Fig. 11). This leads to a significant reduction of pollution (Fig. 12), but also to a significant increase of market concentration (Fig. 14) because of stronger pressure on firms. This is reflected by the higher number of failures when X* is low (Fig. 13), even if firms are pushed out of the market primarily because of their insufficient R&D watch on T2.

This observation gives support to the empirically based argument that regulatory stringency is the most important factor influencing environmental innovation (Ashford et al., 1985; Johnstone, 2007; Johnstone et al., 2012; OECD, 2007), and it qualifies the conclusion of Sartorius and Zundel (2005) that the time element is a fundamental design issue for the success of an innovation-oriented environmental policy.

32

30 Timing is reflected by Tsunset and ∆Trevision: the lower the Tsunset and/or the ∆Trevision, the higher the timing pressure.

31 ∆Trevision seems to be not significant because it does not appear on any tree. 32 We remind that failures can have three causes: insufficient budget, the absence of T2 in the firm’s portfolio while T1 is prohibited, and an insufficient R&D watch on T2. The last seems to be the main cause of failure because regulatory pressure on R&D watches stops only if T1 is prohibited. Consequently, this strong pressure (forced exit) will generally come into play over a long period of time.

8.595 n = 7550

7.443 n = 1445

X* < 48.9

yes

no

8.323 n = 1005

Cost* ≥ 1.472

yes

no

33

5 CONCLUSION This article intends to contribute to a better understanding of the relationship between policy design and eco-innovation through an ABM. Stringency, flexibility, and timing of regulation are crucial to spur eco-innovation. These are key aspects to consider in the REACH regulation, especially to foster the development of alternative substances (such as biosolvents) to replace toxic and harmful substances (such as organic solvents).

The ABM we propose in this article is a first attempt in stylizing policy mechanisms in accordance with the ones underlying REACH, especially the authorization procedure and the extended producer responsibility. Authorization takes place sequentially and involves revisable guidelines in order to force the search for viable alternatives through R&D efforts. Extended responsibility affects the demand of downstream users who give more weight to the suppliers’ technology portfolio and to the timing of regulation. Given the complexity and the sequence of events behind these mechanisms, an ABM is a consistent way to examine the effects of such open-ended standards on the dynamics of innovation and the associated shift to alternative substances.

The main results of the model are as follows: (1) stringency is the most determining feature of policy design (timing is also decisive but it appears to be of secondary importance); (2) technology substitution that brings radical technological change and significant pollution reduction is possible only if regulation is stringent enough but after many sacrifices, especially in terms of market concentration and number of failures; (3) soft regulation does not lead to technology transition because of weak incentives and selection effects. Lack of an irreversible product ban dilutes incremental changes over time and does not result in significant pollution reduction. This calls into question the legitimacy of a not-stringent-enough regulation. Finally, these results stress the necessary trade-offs involved by different degrees of stringency. This very first modeling exercise calls for further research into the impact of REACH on innovative activities.

ACKNOWLEDGMENTS The authors would like to thank the Aquitaine Regional Council for its financial support in the ECOCHIM project and the Regional Council of PACA.

34

REFERENCES Arrow, K., 1962. Economic welfare and the allocation of resources for

innovation, in: Nelson, R. (Ed.), The Rate and Direction of Inventive Activity. Princeton University Press, Princeton, NJ.

Ashford, N.A., Ayers, C., Stone, R., 1985. Using regulation to change the market for innovation. Harvard Environmental Law Review 9 (2), 419–466.

Béal, S., Deschamps, M., Ravix, J.-T., Sautel, O., 2013. Les effets d’une réglementation sur la concurrence et l’innovation: première analyse de la réglementation européenne REACH. Economie et Prévision 197–198, 63–79.

Belis-Bergouignan, M.C., Oltra, V., Saint-Jean, M., 2004. Trajectories towards clean technology: The example of volatile organic compounds (VOC) emission reductions. Ecological Economics 48 (2), 201–220.

Berkhout, F., Ilzuka, M., Nightingale, P., Voss, G., 2003. Innovation in the Chemicals Sector and the new European Chemicals Regulation - A Report for WWF-UK. WWF-UK, Godalming, UK.

Blazejczak, J., Edler, D., Hemmelskamp, J., Jänicke, M., 1999. Environmental policy and innovation - An international comparison of policy frameworks and innovation effects, in: Klemmer, P. (Ed.), Innovation Effects of Environmental Policy Instruments. Analytica, Berlin, 9–30.

Brouillat, E., Oltra, V., 2012. Extended producer responsibility instruments and innovation in eco-design: An exploration through a simulation model. Ecological Economics 83, 236–245.

Chiaromonte, F., Dosi, G., 1993. Heterogeneity, competition and macroeconomic dynamics. Structural Change and Economic Dynamics 4 (1), 39–46.

Cowan, R., Cowan, W., Swann, P., 1997. A model of demand with interactions among consumers. International Journal of Industrial Organization 15 (6), 711–732.

EEA, 2011. EEA Technical Report. Environmental Tax Reform in Europe: Opportunities for Eco-innovation 17. Publications Office of the EU. Copenhagen.

Eppstein, M.J., Grover, D.K., Marshall, J.S., Rizzo, D.M., 2011. An agent-based model to study market penetration of plug-in hybrid electric vehicles. Energy Policy 39 (6), 3789–3802.

35

European Commission, 2013. General Report on REACH. Report from the Commission to the European Parliament, the Council, the European Economic and Social Committee and the Committee of the regions.

Eurostat, 2009. The REACH Baseline Study. Methodologies and Working Papers. European Communities. Luxembourg.

Faber, A., Valente, M., Janssen, P., 2010. Exploring domestic micro-cogeneration in the Netherlands: An agent-based demand model for technology diffusion. Energy Policy 38 (6), 2763–2775.

Fuchs, O., 2011. REACH: A new paradigm for the management of chemical risks. Health and Environment Report 4, Institut français des relations internationales (IFRI). Paris (France).

Grimm, V., Berger, U., Bastiansen, F., Eliassen, S., Ginot, V., Giske, J., Goss-Custard, J., Grand, T., Heinz, S., Huse, G., Huth, A., Jepsen, J.U., Jørgensen, C., Mooij, W.M., Müller, B., Pe’er, G., Piou, C., Railsback, S.F., Robbins, A.M., Robbins, M.M., Rossmanith, E., Rüger, N., Strand, E., Souissi, S., Stillman, R.A., Vabø, R., Visser, U., DeAngelis, D.L., 2006. A standard protocol for describing individual-based and agent-based models. Ecological Modelling 198, 115–126.

Grimm, V., Berger, U., DeAngelis, D.L., Polhill, J.G., Giske, J., Railsback, S.F., 2010. The ODD protocol: A review and first update. Ecological Modelling 221, 2760–2768.

Hahn, R.W., 1989. Economic prescriptions for environmental problems: How the patient followed the doctor’s orders. Journal of Economic Perspectives 3 (2), 95–114.

Hanley, N., Mourato, S., Wright, R.E., 2001. Choice modelling approaches: A superior alternative for environmental valuation? Journal of Economic Surveys 15, 435–462.

Hoyos, D., 2010. The state of the art of environmental valuation with discrete choice experiment. Ecological Economics 69, 1595–1603.

IRSST, 2010. La substitution des solvants par les esters méthyliques d’acides gras d’huiles végétales. RAPPORT B-079. Bilans de connaissances Institut de Recherche en Santé et en Travail. Montréal (Québec).

Jaffe, A.B., Newell, R.G., Stavins, R.N., 2003. Technological change and the environment, in: Mäler, K.-G., Vincent, J.R. (Eds.), Handbook of Environmental Economics, vol. 1. Elsevier Science, Amsterdam, pp. 461– 516.

36

Jaffe, A., Palmer, K., 1997. Environmental regulation and innovation: A panel data study. The Review of Economics and Statistics 79, 610–619.

Jänicke, M., 2012. Dynamic governance of clean-energy markets: How technical innovation could accelerate climate policies. Journal of Cleaner Production 22 (1), 50–59.

Jänicke, M., Lindemann, S., 2010. Governing environmental innovations. Environmental Politics 19 (1), 127–141.

Johnstone, N. (Ed.), 2007. Environmental Policy and Corporate Behaviour. OECD, Paris.

Johnstone, N., Haščič, I., Poirier, J., Hemar, M., Michel, C., 2012. Environmental policy stringency and technological innovation: Evidence from survey data and patent counts. Applied Economics 44 (17), 2157–2170.

Kemp, R., Pontoglio, S., 2011. The innovation effects of environmental policy instruments - A typical case of the blind men and the elephant? Ecological Economics 72, 28–36.

Lancaster, K., 1971. Consumer Demand: A New Approach. Columbia Studies in Economics 5. Columbia University Press, New York.

Lanjouw, J., Mody, A., 1996. Innovation and the international diffusion of environmentally responsive technology. Research Policy 25, 549–571.

Leibenstein, H., 1950. Bandwagon, snob, and Veblen effects in the theory of consumers’ demand. The Quarterly Journal of Economics 64 (2), 183–207.

Malerba, F., Nelson, R., Orsenigo, L., Winter, S., 1999. “History-friendly” models of industry evolution: The computer industry. Industrial and Corporate Change 8 (1), 3–40.

MEI Report, 2007. Measuring Eco-innovation. UM MERIT. The Netherlands. Norberg-Bohm, V., 1999a. Stimulating “green” technological innovation: An

analysis of alternative policy mechanisms. Policy Sciences 32, 13–38. Norberg-Bohm, V., 1999b. Creating incentives for environmentally enhancing

technological change: Lessons from 30 years of US energy technology policy. Technological Forecasting and Social Change 65, 125–148.

Nordbeck, R., Faust, M., 2003. European chemicals regulation and its effect on innovation: an assessment of the Eu's White Paper on the strategy for a future chemicals policy. European Environment 13 (2), 79-99.

37

OECD, 2007. Impacts of Environmental Policy Instruments on Technological Change. COM/ENV/EPOC/CTPA/CFA(2006)36/FINAL. OECD. Paris (France).

Pearce, D.W., Turner, R.K., 1990. Economics of Natural Resources and the Environment. Johns Hopkins University Press, Baltimore.

Popp, D., 2005. Lessons from patents: Using patents to measure technological change in environmental models. Ecological Economics 54, 209–226.

Porter, M., van der Linde, C., 1995. Toward a new conception of the environment - Competitiveness relationship. Journal of Economic Perspectives 9, 97–118.

Rennings, K., 2000. Redefining innovation - Eco-innovation research and the contribution from ecological economics. Ecological Economics 32, 319–332.

Sartorius, C., Zundel, S. (Eds.), 2005. Time Strategies: Innovation and Environmental Policy. Edward Elgar, Cheltenham, UK.

Schwarz, N., Ernst, A., 2009. Agent-based modeling of the diffusion of environmental innovations - An empirical approach. Technological Forecasting and Social Change 76 (4), 497–511.

Schwoon, M., 2006. Stimulating the adoption of fuel cell vehicles. Journal of Evolutionary Economics 16 (4), 435–472.

Scott, J., Trubek, D., 2002. Mind the gap: Law and new approaches to governance in the European Union. European Law Journal 8 (1), 1–18.

Sopha, B.M., Klöckner, C.A., Hertwich, E.G., 2011. Exploring policy options for a transition to sustainable heating system diffusion using an agent-based simulation. Energy Policy 39 (5), 2722–2729.

Sopha, B.M., Klöckner, C.A., Hertwich, E.G., 2013. Adoption and diffusion of heating systems in Norway: Coupling agent-based modeling with empirical research. Environmental Innovation and Societal Transitions 8, 42–61.

van der Vooren, A., Brouillat, E. Evaluating CO2 reduction policy mixes in the automotive sector. Environmental Innovation and Societal Transitions. to be published.

Venables, W., Ripley, B.D., 1999. Modern applied statistics with S-PLUS. 3rd ed. Springer, New York.

Wolf, O., Delgado, L., 2003. The Impact of REACH on Innovation in the Chemical Industry. Technical report series. EC-JRC, EUR 20999 EN.

38

APPENDIX A Table A.1 Overview of processes, parameters, and default values of parameters

Parameter Value Number of firms 10 Number of clients 200 Number of products Number of characteristics 2 4 Potential for each product characteristic T1 Xmax, Costmin, VOCmin, Biomin 100; 0.4; 2; 1 T2 Xmax, Costmin, VOCmin, Biomin 100; 0.4; 0; 0.005 Initial values of each product characteristic T1 X0, Cost0, VOC0, Bio0 100; 0.5; 23; 20 T2 X0, Cost0, VOC0, Bio0 32; 3; 0.05; 0.1 Purchase decision Client’s preferences: a, b, c, d [0; 1] Parameters: A, B, C, D 0; 6; 25; 25 Bandwagon effect: e 0.05 Reserve price [0.75; 4.5] Minimum product quality [32; 100] Total budget Initial budget 15 Mark-up rate µ 0.5 Fixed costs CF 2 Exit/entry Absorptive capacity [0.8; 1.2] Technology portfolio Decision to adopt T2 Adoption threshold [0; 1] Initial switching costs 10 Decision to leave T1 Abandon threshold [0; 1] Innovation process R&D budget 3 R&D rate δ 0.2 R&D share allocated to T1 δ1 0 or 0.5 Speed αI 0.25 Scale parameters β1, β2, β3, β4 0.032; 0.133; 0.087; 0.05 R&D watch Speed αw 0.25 Knowledge accumulation Scale parameter αK 0.2 Kmax 1 Switching costs Scale parameter αsc 0,1 SCmin 10 Rebuy process Tolerance parameter [0.8, 1.2]

39

APPENDIX B Table B.1 Chosen domain of parameters in Monte Carlo procedure

Target Thresholds Timing Credibility Techno-economic

performances R&D watch Sunset date

Tsunset Revision period

∆Trevision αR X* Cost* αRDwatch*

Range of values [32;100] [0.4;3] [0;1] [1;249] [1;25] [0;1]

DOCUMENTS DE TRAVAIL GREDEG PARUS EN 2013GREDEG Working Papers Released in 2013

2013-01 Raphaël Chiappini Persistence vs. Mobility in Industrial and Technological Specialisations: Evidence from 11 Euro Area Countries2013-02 Kevin D. Hoover Was Harrod Right?2013-03 Kevin D. Hoover Man and Machine in Macroeconomics2013-04 Isabelle Corbett-Etchevers & Aura Parmentier-Cajaiba Toying with Regulation: ‘Strategizing Tools’ as Organizational Bricolage2013-05 Aura Parmentier-Cajaiba Research Diary Mapping: Enhancing Reflectivity in Process Research2013-06 Richard Arena Sraffa’s and Wittgenstein’s Crossed Influences: Forms of Life and Snapshots 2013-07 Christophe Charlier & Sarah Guillou Distortion Effects of Export Quota Policy: An Analysis of the China - Raw Materials Dispute2013-08 Cristiano Antonelli & Alessandra Colombelli Knowledge Cumulability and Complementarity in the Knowledge Generation Function2013-09 Marco Grazzi, Nadia Jacoby & Tania Treibich Dynamics of Investment and Firm Performance: Comparative Evidence from Manufacturing Industries2013-10 Anna Calamia, Laurent Deville & Fabrice Riva Dynamics of Investment and Firm Performance: Comparative Evidence from Manufacturing Industries2013-11 Lauren Larrouy Bacharach’s ‘Variable Frame Theory’: A Legacy from Schelling’s Issue in the Refinement Program?2013-12 Amel Attour Adoption et modèles de diffusion régionale de l’innovation dans les gouvernements locaux: le cas du développement de l’e-Gouvernement en Lorraine2013-13 Anaïs Carlin, Sébastien Verel & Philippe Collard Modeling Luxury Consumption: An Inter-Income Classes Study of Demand Dynamics and Social Behaviors2013-14 Marie-José Avenier & Catherine Thomas Designing a Qualitative Research Project Consistent with its Explicit or Implicit Epistemological Framework2013-15 Amel Attour & Maëlle Della Peruta Le rôle des connaissances architecturales dans l’élaboration de la plateforme technologique d’un écosystème en émergence: le cas des plateformes NFC2013-16 Evelyne Rouby & Catherine Thomas Organizational Attention Elasticity: An Exploratory Case of Cement Production2013-17 Małgorzata Ogonowska & Dominique Torre Residents’ Influence on the Adoption of Environmental Norms in Tourism

2013-18 Isabelle Salle & Pascal Seppecher Social Learning about Consumption2013-19 Eve Saint-Germes & Sabrina Loufrani-Fedida L’instrumentation de la GTEC au service de l’articulation entre compétences individuelles et employabilité : le cas de la plateforme eDRH062013-20 Francesco Quatraro & Marco Vivarelli Entry and Post-Entry Dynamics in Developing Countries2013-21 Dorian Jullien, Judith Favereau & Cléo Chassonnery-Zaïgouche Rationality and Efficiency: From Experimentation in (recent) Applied Microeconomics to Conceptual Issues2013-22 Nabila Arfaoui, Eric Brouillat & Maïder Saint-Jean Policy Design, Eco-innovation and Industrial Dynamics in an Agent-Based Model: An Illustration with the REACH Regulation