Policy instruments for green-growth of clusters: Implications from an agent-based model

Kamath, Ram; Sun, Zhanli; Hermans, Frans Article — Published Version Policy instruments for green-growth of clusters: Implications from an agent-based model Environmental Innovation and Societal Transitions Provided in Cooperation with: Leibniz Institute of Agricultural Development in Transition Economies (IAMO), Halle (Saale) Suggested Citation: Kamath, Ram; Sun, Zhanli; Hermans, Frans (2022) : Policy instruments for greengrowth of clusters: Implications from an agent-based model, Environmental Innovation and Societal Transitions, ISSN 2210-4224, Elsevier, Amsterdam [u.a.], Vol. 43, pp. 257-269, https://doi.org/10.1016/j.eist.2022.04.003 , https://www.sciencedirect.com/science/article/pii/S2210422422000387 This Version is available at: https://hdl.handle.net/10419/266479 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc-nd/4.0/ Environmental Innovation and Societal Transitions 43 (2022) 257–269 Available online 20 April 2022 2210-4224/© 2022 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Policy instruments for green-growth of clusters: Implications from an agent-based model Ram Kamath, Zhanli Sun, Frans Hermans Leibniz Institute of Agricultural Development in Transition Economies (IAMO) Theodor-Lieser-Straße 2, 06120 Halle (Saale), Germany ARTICLE INFO Keywords: Decoupling Green-clusters Agent-based modelling Socio-technical transitions Policy instruments ABSTRACT Green-growth theory asserts that green-innovations will allow us to decouple economic growth from pollution. Policies that promote green-innovations are therefore paramount. Scholarship has become increasingly interested in how green-clusters can contribute to sustainable innovation and decoupled green-growth. In this study, we create an agent-based model that can simulate a cluster’s transition, as shaped by different policy instruments. We use this model to explore the effects of a) innovation grants, b) fines for pollution c) financial incentives for entrants, and d) an instrument mix of incentives and fines, on the green-growth of a peripheral-region cluster. Our results indicate that designing instruments for absolutely decoupled growth of peripheral-clusters is close to impossible; and demonstrate the inherent trade-offs in designing policies for relatively decoupled growth. Based on these results, we opine there should be more nuanced deliberation on the potential contribution of green-clusters to sustainable development,with greater focus on the possible trade-offs. 1. Introduction The notion of green-growth has emerged as a dominant policy response to climate change (Dale et al., 2016). Green-growth theory asserts that green-technological innovation will allow us to relatively or absolutely decouple economic growth from resource use and carbon emissions (Solow, 1973; Hickel & Kallis, 2020). Therefore, policies that promote the creation and adoption of green-innovations are paramount to achieving some level of decoupled green-growth. As a response, over the last decade, innovation policy literature has become increasingly focused on innovations targeting transitions and green-growth (Schot & Steinmueller, 2018). Scholarship has become increasingly interested in how green industrial clusters can contribute to sustainable innovation and green-growth (Hansen & Coenen, 2015; Boschma et al., 2017). Green-clusters grow in a decoupled fashion by developing and selling products or technologies that “reduce carbon emissions and pollution, enhance energy and resource efficiency, and prevent the loss of biodiversity and ecosystem services” (UNEP, 2011:16). In this study, we explore implications for policies to advance the green-growth of clusters. Here, we are specifically interested in policies for greening of clusters located in peripheral regions. According to the typology introduced by T¨ odtling & Trippl (2005), we can distinguish between metropolitan, specialized and peripheral regions. However, we have limited understanding of the prospects for the development of green-clusters in the various types of regions, and of the differences in required policies. Until recently, research has focused on the evolution of metropolitan or specialised regions, implicitly assuming that there is no innovation in peripheral areas. Growth of any form is difficult in peripheral regions because they are characterized by a lack of critical mass in any industrial E-mail address: [email protected] (R. Kamath). Contents lists available at ScienceDirect Environmental Innovation and Societal Transitions journal homepage: www.elsevier.com/locate/eist https://doi.org/10.1016/j.eist.2022.04.003 Received 6 September 2021; Received in revised form 22 March 2022; Accepted 4 April 2022 Environmental Innovation and Societal Transitions 43 (2022) 257–269 258 specialization, and by the lack of high-value, knowledge-intensive activities (Isaksen & Trippl, 2014). This view is being increasingly questioned, which is reflected by a rising number of papers, special issues, and edited volumes on innovation outside of agglomerations (Eder, 2019). To fill in the knowledge gap of green-growth in peripheral regions, we address the question, What policy instruments are most effective in causing green-growth of clusters in a peripheral region? To answer this question, we built an agent-based model (ABM) to simulate the transition of a peripheral cluster of firms. With this ABM, we aim to advance understanding of policy instruments that promote green-growth and innovation, and of the emerging practice of modelling socio-technical transitions (Holtz et al., 2015; K¨ ohler et al., 2019). The simulations are based on the idea that the emergence of green-growth in peripheral regions commonly involves the deployment of green-technologies developed elsewhere, i.e. greening through importation (Grillitsch & Hansen, 2019). Mere availability of a green-technology does not guarantee widespread adoption; policy support is often a pivotal catalyst for adoption (del Río Gonz´ alez, 2005). In this regard, several studies have argued for the application of different policy instruments (Nauwelaers et al., 2009; IEA, 2011). Using the ABM, we explore the effectiveness of three individual instruments, and an instrument-mix, on the peripheral cluster’s green-growth. The instruments are 1) Financial incentives to attract external actors in a green industry, 2) Grants provided to qualifying innovation projects being run by local cluster actors, and 3) Imposition of fines on firms that cross a certain level of pollution. The instrument-mix we explore is a combination of incentives for foreign actors and fines. In the following section, we introduce the reader to the complex adaptive systems perspective that guides our ABM. In section three, we describe the characteristics of the agents and processes within our agent-based model. In the fourth section, we present and interpret the results of the simulation. In section five, we discuss the strengths and limitations of our model, lessons from our findings, and implications for how scholarship views green-clusters and decoupled growth. 2. Theoretical perspectives In this section, we discuss the perspective of complex adaptive systems (CAS), which underlies the approach of agent-based modelling. We detail why clusters can be viewed as CAS, and how green-growth can be viewed as the evolution of the clusterCAS’s macro-level properties. 2.1. Clusters as complex adaptive systems A CAS is composed of a dynamic network of actors that continually (re)act in response to the (re)actions of other actors, and to external stimuli (Waldorp, 1993). In a CAS, aggregate behaviour and various systemic characteristics are seen to emerge from the myriad actions and interactions of its constituent actors (i.e. the process of upward causation) (Gandolfi, 1999). At the same time, these characteristics may guide, and constrain, the actions of the actors making up the CAS (i.e. the process of downward causation). An industrial cluster is an interconnected, mutually dependant network of actors (such as firms, universities) and institutions, working in a particular field, concentrated in a particular geographical area (Porter, 1998). Industrial clusters exhibit characteristics that make them complex adaptive systems (Martin & Sunley, 2011; Nikolic et al., 2009), such as: 1 Emergence: In CAS, macro-level structures and dynamics emerge out of micro-level behaviors and interactions. Cluster-wide properties such as pollution-levels, financial assets emerge from the actions and interactions the cluster’s firms, universities etc. These properties then influence the actions of cluster members that (re)produce them. 2 Adaptation: CAS can adapt their structures and dynamics. Clusters can adapt (e.g. by reducing pollution-levels) in face of external or internal demands (Martin & Sunley, 2011). This enables clusters to survive, and even thrive, by evolving towards new (more sustainable) set of structures (Klepper, 2006). 3 Openness: CAS tend to be dissipative—subject to constant interaction and exchange with their environments. Clusters are open systems that engage in continual exchanges with its environment (Martin & Sunley, 2011). 2.2. Green-growth is the transition of a cluster CAS Green-growth implies that, amongst other changes, the cluster’s financial standing should improve as pollution-levels decrease. When we view a cluster as a CAS, its green-growth can be perceived as a transition of its macro-level properties (such as pollution, financial assets etc.); where these properties become more sustainable, and guide actor behaviour that drives further transition. These changes at the cluster-level will only emerge through the adoption and creation of green-technologies at the actor-level (i.e. upward causation). In parallel, cluster-level structures (such as the availability of grants) will attempt to shape actors’ choices that advance (or retard) decoupled growth (i.e. downward causation). Essentially, what we model in our ABM are these processes of upward and downward causation, and how they possibly lead to green-growth of the cluster. 2.3. Peripheral regions and green-growth Peripheral regions are characterised by low-levels of clustering, and a lack of industrial specialisation (Isaksen & Trippl, 2014; Grillitsch & Hansen, 2019). These regions are dominated by small-and medium-sized enterprises. They can host a few large firms (Grillitsch & Hansen, 2019). Peripheral firms engage in low-levels of collaboration, leading to little knowledge-exchange (Isaksen & Trippl, 2014; Grillitsch & Nilsson, 2015). Amongst the various types of regions, innovation rates are lowest in peripheral regions. R. Kamath et al. Environmental Innovation and Societal Transitions 43 (2022) 257–269 259 Furthermore, innovation by peripheral firms is mostly of an incremental nature (Shearmur, 2011; Isaksen & Trippl, 2014). Owing to the relatively weak support system for innovation, peripheral firms are unlikely to find all the resources necessary for innovation, within these regions. Consequently, to create green-growth paths, they must collaborate, with extra-regional actors, or use imported knowledge and technologies. Studies such as Grillitsch & Nilsson (2015), Grillitsch & Hansen (2019) argue that (green) growth in the periphery will be centred on the activities of a few innovative firms in the region, attracting foreign agents, and importing (green) knowledge and technologies. 2.4. Sustainability treadmill In our ABM, the agents are essentially on a type of technological treadmill (Cochrane, 1958). In its original form, the treadmill theory refers to a race to continually update technology in possession, to lower production costs. Failure to update technology leads to relatively higher costs, and to bankruptcy. In our model, we apply this idea with a modification – agents are in a race to continually innovate, update green-knowledge and green-technologies, lower pollution and become more sustainable (while becoming financially richer). If they fail to innovate, their pollution rises (on both a relative and absolute basis), as the performance of their green-technologies deteriorates with time. With failure to innovate, the agents become (relatively) dirtier, fall behind in the sustainability treadmill, and eventually, fall off and perish (as their dirtier products lose share in the market). 3. Methodology The most effective and accurate model of a CAS is one which is itself, a CAS. Therefore, the most suitable option to model cluster evolution is agent-based models. To answer our research question, we created an ABM in the Netlogo 6.1 modelling environment. Netlogo has an easy-to-use GUI, and does not require a lot of programming effort, letting us focus on the model’s logic. ABMs can be relied upon to examine how a system will respond to developments such as the introduction of new policy. There are multiple instances of ABMs being used to explore various emergent phenomena in industrial clusters. Dilaver et al. (2014) investigated the relationship between the entrepreneurial character of a region and the emergence of clusters. Agents in the model run innovation projects to create new knowledge. When these projects succeed, agents reap financial rewards. Successful projects can also lead to the creation of spin-offs. Experiments revealed a positive correlation between the entrepreneurial nature of the region and cluster development (because of higher spin-off formation). Canals et al. (2008) investigated the link between the willingness of actors to collaborate (to share knowledge, resources), and the intensity of clustering in a region. In the model, firms randomly collaborate. Experiments revealed that spatial clustering was more intense when the willingness to collaborate was high. Vermeulen & Pyka (2014) simulated the mediating effects of agents being able to collaborate with extra-regional actors. Agents in this ABM collaborate and share inputs for innovation. Rewards from successful innovation, which are abilities to create more advanced technological artefacts, are equally shared amongst the collaborators. The model demonstrated that when agents are able to collaborate with external actors, the artefacts created are much more radical. Finally, Zhang (2004) modelled how the emergence of clusters begins with the appearance of some inspiring entrepreneurial agents. In the model, an entrepreneurial firm randomly appears and makes a large profit, which inspires other firms to enter the landscape. Each firm is given some level of technological assets. Firms that fail to make a profit fail to innovate, leading to their exit. These processes combine to cause spatial clustering of firms. While our model has been inspired by the features of these models, unlike preceding ABMs, we do not model for innovation that merely leads to richer agents; rather, we model for innovation that makes the agents (and the cluster) not only richer, but also greener. Moving away from extant models of cluster-based innovation, we simulate processes of environmentally-sustainable innovation. Firmagents collaborate and contribute resources to projects aimed at effecting incremental or radical green-innovation. Succesful projects help agents not only reap financial rewards but also new green-knowledge and improved green-technologies; which help agents lower their pollution, and become more sustainable. Additionally, we make a further contribution to the field of modelling cluster-based innovation - by conducting experiments to determine which instruments can effectively stimulate green-growth. 3.1. Structure of the agent-based model We describe below, the variables in the ABM; the main agent behaviours, and processes, within the ABM; and the macro-level parameters that we track. For greater detail on the ABM, we welcome the reader to view the Overview, Design concepts, Details (ODD) protocol in the supplementary materials. Agent-properties and global variables In our ABM, there are a number of agents (i.e. firms), clustered in a landscape with characteristics of a peripheral region. The behaviours of these agents are governed by two sets of variables - agents’ own and global. The main agent variables are 1) Financial capital, 2) Knowledge capital, 3) Reputational capital, 4) Pollution-levels, 5) Radical project experience, and 6) Incremental project experience. The main global variables are 1) Probability of collaboration 2) Probability of radicalness 3) Probability of innovating 4) Innovation potential 5) Cluster-size 6) Decay rate. The first five global variables are informed by the postulations of T¨ odtling & Trippl (2005) and Isaksen & Trippl (2014), and data from the European Innovation scoreboard for 2019–2020. The definitions of these variables are given in Table 1 below. We can use this model to simulate cluster-transition in any region, by controlling the global variables. To answer our research R. Kamath et al. Environmental Innovation and Societal Transitions 43 (2022) 257–269 260 question, we simulate the greening of a cluster in a peripheral region. We operationalise the features of the periphery, as detailed in Section 2.3, by controlling values for the following global variables: •Percentage of large firms: we only have a small percentage of large firms in the cluster, at initialisation. •Probability of collaboration: we have used the range of values for modest regions (0 to 27%) from the European RIS scoreboard. •Probability of radicalness: we have used the range of values for modest regions (0 to 55%) from the European RIS scoreboard. •Probability of innovation: At initialisation, we have used the lowest value for modest regions (2.2%) from the European RIS scoreboard. We use data for “modest innovator regions” because they resemble most closely, peripheral regions. They are regions whose overall innovation scores are below 50% of the European average. The ranges for probabilities for collaboration, innovation, radicalness are lowest for modest regions. To simulate conditions of metropolitan or specialized regions, we simply change the range of values for the above global variables (using data for “strong innovators” or “innovation leaders” from the scoreboard). As the model runs, the ranges for probability of collaboration and probability of radicalness are always adhered to. This way, the agents always collaborate and engage in radical innovation at rates that are characteristic of peripheral firms. We do not maintain a range for the probability of innovating, and for the percentage of large firms, as we treat them as an emergent variables shaped by agents’ innovation activities. Agents’ action-space The objective of the agents in the simulation is to survive, and grow, for as long as possible. The agents exist in an environment where demand for their products is positively correlated with their sustainability (applying the idea of the sustainability treadmill from Section 2.4). Consequently, each agent’s survival depends on its ability to increase knowledge-capital, financial capital, and reputational capital. The challenge facing agents is that with time, the capital stocks they hold diminish, while pollution climbs. In the model, assets decrease by the formula C(1 – dr) , while the pollution increases by the formula P(1 +dr). “C” is the value of the capital-asset in the prior time-step, “P” is the pollution-level of the agent in the prior time-step, and “dr” is the decay rate. The agents’ knowledge capital (which is green-knowledge, green-technologies and associated machinery and equipment possessed by an agent) decreases with each time-step. As the sustainability performance of its green-technologies decreases with time (due to deterioration of machinery and equipment), the agent’s knowledge capital loses value. With their physical technology assets deteriorating, and the technology becoming relatively inefficient, agents’ products and processes become dirtier, as the pollution-level of each agent increases (both in relation to other agents’ products and processes, and on an absolute basis). When an agent’s pollutionlevel crosses 100, the agent will cease to exist (an emulation of the government forcing a highly polluting firm to close). Agents’ financial capital decreases when demand for their products comes down. As agents’ knowledge capital decreases, the sustainability of their products and processes come down. This causes the relative competitiveness of products to go down, reducing demand for them. Lower demand forces agents to use their reserves to survive; hence the reduction in financial capital each time-step. When the financial capital for an agent falls under 10, the agent will cease to exist. With declining value of green-technology and knowledge, declining sustainability for their products, the reputational capital of the agents decreases each time-step. When the reputational capital for an agent falls under 10, the agent will cease to exist. Agent survival and growth requires innovation at a rate that staves off rising pollution and decreasing assets. New knowledge and Table 1 The agents’ own and global variables that guide agent behaviour. Variable Definition Agents’ own Financial capital Financial capital in possession of an agent. Knowledge capital Green-knowledge and green-technologies possessed by an agent. Reputational capital Reputation, sustainability credentials of an agent. Pollution-level Represents how polluting an agent’s operations are. Radical project experience The number of successful radical innovation projects an agent has participated in. Incrementalproject experience The number of successful incremental innovation projects an agent has participated in. Global Probability of collaboration This is the probability of firms collaborating for a (green) innovation project.Defined as ratio of number of firms collaborating in an innovation project to the total number of firms. The range of values for this variable is obtained from the range of values for "Innovative SMEs collaborating with others" for modest regions, from the European RIS 2019–2020 scoreboard. Probability of radicalness The probability of a (green) innovation project being a radical one. Defined as the ratio of number of successful radical innovation projects to total number of successful projects. The range of values for this variable is obtained from the range of values for "Sales of new-to-market and new-to-firm innovations" for modest regions, from the European RIS 2019–2020 scoreboard. Probability of innovating The probability of a (green) innovation project succeeding. Defined as the ratio of number of successful projects to total number of projects. The initial value for this variable is obtained from the lowest value for "Product or process innovators" for modest regions, from the European RIS 2019–2020 scoreboard. There is no range used for this variable. Innovation potential This is the ratio of the number of successful projects to the total number of firms at a point time. Cluster-size This is the number of agents in the cluster, at any given point in time (note: split agents are counted as one agent). Decay rate This is the rate at which the agents’ three capital asset stocks decline with each time-step, and also the rate at which the agents’ pollution rises with each time-step. R. Kamath et al. Environmental Innovation and Societal Transitions 43 (2022) 257–269 261 technology assets are created or adopted through green-innovation projects. Successful projects endow agents with rewards in the form of capital assets and lower pollution. Initially, most projects are incremental in nature, and green-technologies are assumed to be available “off-the-shelf” (applying the idea of greening through importation (Grillitsch & Hansen, 2019)). After multiple instances of incremental innovation, the peripheral agents can develop more advanced green-technologies through radical innovation projects. Through (incremental) innovation projects, agents can obtain new knowledge and tech, which help them (slightly) lower pollution. While innovation lowers pollution on an absolute basis, it can also make them cleaner in comparison to other agents; which increases demand for the agent-firms’ products. So agents also reap financial asset rewards through innovation. Along with green-innovation success, comes greater reputation. So, the agents also gain reputational capital. With successful radical projects, the firm-agent moves onto a new green-technology that surpasses the sustainability of technologies in possession. The asset rewards and pollution-reduction in this case, therefore, are higher than from incremental innovation. However, radical innovation is rarer, with lower chance of success. Engaging in radical innovation projects requires agents to first succeed in several incremental innovation projects. Without green-innovation, agents turn dirty, with higher levels of pollution and lower levels of capital. This leads to firm-agents ceasing to exist, and the cluster dying. Green-innovation in the model has three steps – initiating projects, innovating, and reaping rewards for successful innovation. Initiating green-innovation projects To start an innovation project, an initiator agent collaborates with between one and five agents that possess the highest levels of knowledge capital. Once a project is initiated, links are formed between the project partners. Any firm-agent can initiate a greeninnovation project. At a time, an agent can only be a member of one project (unless it splits into two divisions, in which case the agent can participate in two projects simultaneously). For this exploratory model, we have not factored in the effects of any proximity dimensions (such as geographical, social, institutional, cognitive or organisational (Boschma, 2005)) in the collaborative process; we aim to incorporate these dimensions in future versions. The number of projects successfully initiated in a time-step is the product of probability of collaboration and a random positive integer below the variable “number of projects in a tick”, which is input by the observer. Some of the projects are randomly chosen to be radical innovation projects. In the beginning stages, most projects are incremental innovation projects. The chances of an initiated project being a radical project are positively correlated with the innovation potential. With time, if there are several successful incremental innovations, the innovation potential rises, and so can the number of radical projects. The act of innovation After the agents have collaborated on a project, they innovate using green-technologies that present the possibility of new capital rewards, and reduction in pollution. To execute the project, project-partners commit capital to the project. The chances of project success are directly proportional to the total knowledge, financial, and reputational capital committed to the project, the total number of successful radical/incremental projects that the partners have been a part of, and the global variable probability of innovating. Chances of project success are lower when the project is a radical one. The capital an agent commits to the project is proportional to its capital stocks, and the number of successful projects it has been part of. This way, richer, more experienced agents are able to commit more capital. A project succeeds or fails within one time-step. When a project succeeds, a few things happen. Firstly, the probability of collaboration increases (by the ratio of the number of partners in the project to total number of agents). The increment occurs as long as the probability of collaboration stays under the upper limit of its range (see Table 3). Secondly, the probability of innovating is updated (by calculating ratio of successful projects to total number of projects undertaken), as long as it stays under the value of one. Thirdly, the radical project experience or incremental project experience for each partner is increased by one. Next, if the project that succeeded is a radical one, probability of radicalness is updated (by calculating ratio of successful radical projects to total number of successful projects). The updating occurs as long as the probability stays under the upper limit of its range (see Table 3). By maintaining the ranges for probability of collaboration and radicalness, we ensure that the agents always behave as peripheral firms. Fifth, if the successful project is a radical one, a spin-off firm-agent is created from the initiator agent (like in Dilaver et al. (2014)). This spin-off will have 25% of the parent’s knowledge and financial capital, pollution-level, and past experience with successful projects; and 50% of reputational capital. Finally, the links that were formed between the project-partners are removed (which is the first event that occurs when a project fails). Rewards from green-innovation Successful innovation immediately rewards all participating agents equally (like in Vermeulen & Pyka (2014)), with capital-assets and lower pollution. After each successful project, the agents will have their financial, knowledge and reputational capital doubled (in case of a radical project), or rise by 25% (in case of an incremental project). Pollution is halved (in case of a radical project), or decreases by 25% (in case of an incremental project). See Table 4, for the equations that govern capital increments, and pollution decrements. As firms become richer, they can be involved in multiple innovation projects. If any of an agent’s three capital-stocks go above 100, the agent is allowed to participate in two innovation projects within the same time step. The model enables this by splitting the agent into two separate agents that represent internal divisions that can independently carry out innovation-projects (conceptually, we think of this splitting as a rich, large firm reorganising into independent divisions that are focused on one particular product-line, or a certain market). The parent agent’s financial and knowledge capital, its pollution, and its past experience with successful projects are split into two, and each half is inherited by the two divisions. Since there are no new firms being created, but rather independent internal R. Kamath et al. Environmental Innovation and Societal Transitions 43 (2022) 257–269 262 divisions, the two new agents are counted as one when the model calculates cluster-size. Agent death Rewards are tempered by decreasing capital and increasing pollution in each time-step. See Table 4, for the equations that govern capital decrements, and pollution increments. If the financial or reputational capital falls under the value of 10, or if the pollution-level for an agent goes above 100, the agent will cease to exist. With the death of an agent, the innovation potential of the cluster is updated (as the ratio of the number of successful projects surviving agents have executed, to the number of agents). Macro-level patterns and end-state of interest We track the following macro-level variables, to gauge the progress of the cluster’s green-growth: 1 Pollution-levels: This is the sum of the pollution-levels of all the agents that make up the cluster. 2 Cluster-size: This is the number of agents existing at any point in time (note: split agents are counted as one agent). 3 Knowledge capital: This is the sum of the knowledge capital stock of every agent in the cluster. 4 Financial capital: This is the sum of the financial capital stock of every agent in the cluster. 5 Reputational capital: This is the sum of the reputational capital stock of every agent in the cluster. 6 Ratio of pollution to financial capital: Green-growth requires economic growth to decouple from pollution. We track this ratio to see if as the cluster’s pollution-levels go down, its financial capital stock goes up. Besides these six parameters, we also track the number of successful projects and the number of successful radical projects. The desired end state, where the simulated cluster has transitioned and decoupled, is one where pollution has declined, cluster-size has increased or at least remained the same, capital stocks have increased, and the ratio of pollution to financial capital is trending down. The least desired state is when the agents have died out as pollution increased and capital stocks have been depleted. 3.2. Experiments with instruments shaping green-growth In the ABM, we conducted experiments to simulate the effects of three individual instruments, and an instrument-mix, on peripheral agents’ innovating behaviour, and on the green-growth of the cluster. The three instruments were 1) Financial incentives used to attract external actors having experience with the imported green-technologies, 2) Grants provided to qualifying innovation projects being run by cluster actors, and 3) Imposition of fines on cluster actors that cross a certain level of pollution. We chose these instruments because we wanted a group of instruments with different purposes. While grants and incentives for foreign actors are both economic instruments for technology-push, fines are a regulatory instrument for demand-pull. The instrument-mix we explored is incentives and fines being applied simultaneously. This combination ensured that while our study did not analyse the effect of so-called “sermons” (i.e. communicative policy tools that stimulate stakeholder participation, learning and collaboration (Wieczorek & Hekkert, 2012)), we covered the “carrots” and “sticks” of innovation policy. Incorporating communicative tools is something we aim to explore in future research. The four experiments we conducted are summarised in Table 5. Model initialisation for experiments For each of the experiments, we initialised the model with a total of 50 agents, with 3% of the firm-agents being “large” firms. For the large firms, financial capital and reputational capital were assigned random initial values between 50 and 100. For the small firms, financial and reputational capitals were random values between 10 and 50. The model was initialised with agents having random pollution-levels between 0 and 100. For agents with a pollution-level of under 50, knowledge capital is a random value between 25 and 75. These greener agents are given the colour green. For all other agents, knowledge capital is 25. These “dirtier” agents were given the colour red. At initialisation, the radical project experience and incremental project experience for each agent was set to 0. We assigned probability of collaboration a value of 14% (roughly mid-point of the range) and probability of radicalness, a value of 28% (roughly mid-point of the range). At initialisation, the model gave probability of innovating a value of 0.022 (lowest possible value), and innovation potential, value of 0.01 (meaning only 1 out of 100 agents can come up with a successful innovation). The number of projects initiated in a tick is the product of probability of collaboration and a random positive integer below five (number of projects in a tick). A project could have between 2 and 6 partners (we chose to use this range based on an overview of projects in a cluster we previously studied. Between 2014 and 2019, there were 27 collaborative projects in this cluster, only four of which included 6 or more partners). We set the decay rate at which assets decline and pollution rises at 0.1%. For the range of possible variable values at initialisation, see Table 3. Once started, the simulation runs till either of two conditions is met - when the number of agents in the cluster is less than the number of projects in a tick (5), or when the number of ticks is greater than 300. Imposition of fines on polluting firms Fines refer to financial capital that is instantly reduced from an agent’s stock when its pollution-level is equal to, or crosses, 33. Every time-step, the model finds agents that do not satisfy this criterion, and fines them. Fine levels range from 0% to 25% of each agent’s financial capital. For this scenario, we varied fine levels from 0% to 25%, with jumps of 5%. At each fine level, we executed 100 simulation runs, for each of the macro-level patterns of interest. R. Kamath et al. Environmental Innovation and Societal Transitions 43 (2022) 257–269 263 Introducing grants for local innovation projects Grants refer to financial capital that is given to a particular project, which is added to the financial capital that has already been committed to a project, by project partners. Grant levels range from 0% to 25% of the financial capital possessed by the project. In each time-step, during the act of innovation, the model will check if particular projects can receive grants. The chances of a project receiving a grant are positively correlated with capital stocks committed to it, and the number of successful incremental/radical projects that the partners have been part of (following Banal-Estanol et al. (2016)). For this scenario, we varied grant levels from 0% to 25%, with jumps of 5%. At each grant level, we executed 100 simulation runs, for each of the macro-level patterns of interest. Incentives for attracting entrants Incentives refer to financial capital that is bestowed to the entrant, immediately upon entrance. Incentives range from 0% to 25% of the financial capital an entrant possesses at entry. The probability of a new agent entering the cluster is positively correlated to the incentives set. Only one entrant may enter the cluster in each tick, and it may be a large or small agent. Entrants upgrade the knowledge capital of the cluster by bringing in new, higher-value green-technology and knowledge. Entrants’ knowledge capital, ranging between 75 and 100, is higher than of cluster agents. An entrant’s pollution-level is the same as the lowest pollution-level amongst clusteragents. Entrants’ experience with successful incremental and radical innovation is random values between the mean and maximum experience amongst the cluster-agents. Once they enter the cluster, entrants will behave like any other cluster agent. If they fail to innovate, they can turn dirty, become asset poor, and cease to exist. For this scenario, we varied incentive levels from 0 to 25%, with jumps of 5%. At each incentive level, we executed 100 simulation runs, for each of the macro-level patterns of interest. Introducing an instrument-mix of incentives and fines In this final scenario, every time-step, some entrants may enter the cluster and receive incentives, and some agents may get fined. For this scenario, we varied both incentive and fine levels from 0 to 25%, with jumps of 5%. At each combination of incentives and fines, we executed 100 simulation runs, for each of the macro-level patterns of interest. 4. Results We now present the variations of the macro-level patterns of interest for each scenario. We first discuss results for the individual instruments, then for the instrument-mix. Agent-based modelling often entails, as in our case, stochastic settings, and may generate rich and complex patterns. To investigate the uncertainty of the modelling results and how the uncertainty or variations of the model outputs can be attributed to various input variables, a sensitivity analysis needs to be conducted symmetrically (Ten Broeke et al., 2016). We used RNetLogo, and its extended R packages, nlexperiment and nlrx, to execute a sensitivity analysis. We welcome the reader to view the setup and results of this analysis in the supplementary data. 4.1. Green-growth shaped by fines, grants, incentives We started off our experiments by first analysing the effects of the three policy instruments individually. In the simulation where we imposed fines on polluting firms, we saw that as fines increase, the number of successful (radical) projects decline. We also saw that the cluster-size drops below the initial value at all fine levels, with the reduction in size trending slightly up with fine-levels. Asset-stocks stay below initial levels at all fine-levels, with the reduction in stock trending up with fines. Decreasing innovation success causes a decline in asset values, leading to a shrinking cluster. Pollution stays below initial values, and drops more with increasing fines. However, with declining innovation, this drop in pollution has to be attributed to shrinkage in cluster size. Decreasing pollution causes the ratio of pollution to financial capital to trend down. Overall, fines seem to retard the green-growth of the cluster by failing to encourage green-innovation. In the simulation of grants for qualifying projects, we saw that increasing grant levels lead to higher levels of innovation (unlike with fines). Like with fines, asset values and cluster-size drop below their initial values at all grant-levels. However, we saw that with rising grants, the magnitude of reduction in size and assets diminish. We observed that with rising grants, the drops in pollution levels increase. With cluster-size increasing with grants, we can attribute this increasing reduction in pollution to rising levels of innovation. We also saw that rising grants causes the ratio of pollution to financial capital to decline. For this scenario, we ran another set of simulations, with a larger grant range of between 0 and 200%. We did see a positive effect where the magnitude of reduction in clustersize and assets decreases; but even with a grant-level of 200%, an inflection to rising cluster-size and stock-levels does not occur. This tells us that simple financial grants alone are not sufficient to cause a richer, decoupled cluster (unless perhaps when they reach extraordinary levels that dwarf private investment). In the simulation of incentives for entrants, what we saw is that with increasing incentives, the number of successful (radical) projects increase. Rising innovation leads to cluster-size and capital assets moving up from their initial values. However, this is only after incentives cross certain threshold. Below these thresholds, innovation is not high enough to lead to increasing size and capital stocks. Although pollution drops below initial values at all incentive-levels, raising incentive levels diminishes the amount of reduction in pollution. This demonstrates that we cannot optimise for pollution-reduction and capital-augmentation simultaneously. We will have to either give up a bit of capital or some pollution-reduction, in choosing one of many states where pollution has dropped, and capital stock has risen. High incentives seem to aid decoupling, as the ratio of pollution to financial capital fall with increasing incentives. We also ran a second set of simulations for this scenario, where entrants’ experience with successful innovation was the same R. Kamath et al. Environmental Innovation and Societal Transitions 43 (2022) 257–269 264 as the maximum experience amongst the cluster-agents. What we saw was that with more experienced entrants, the inflection points where asset-values and cluster-size start increasing from initial values, are at lower incentive-levels than in the first case. This is because of higher innovation levels, which also lead to higher magnitude of asset and cluster-size augmentation. We welcome the reader to view the graphs we have provided in the supplementary data, which depict the variation of the macrovariables of interest in each of the above scenarios. 4.2. Green-growth shaped by an instrument-mix of incentives and fines We now come to the scenario where entrants are incentivised to come into the cluster, and where agents can be fined. Fig. 1 depicts how pollution in the cluster varies with increasing levels of incentives and fines. Fig. 2 depicts how the ratio of pollution to financial capital of the cluster varies. Fig. 3 depicts the variation of cluster-size, knowledge capital of the cluster, financial capital of the cluster, reputational capital of the cluster, the number of successful radical projects, and the number of successful projects. From Fig. 1, we see that the pollution has dropped below its initial value throughout, but the reduction in pollution increases significantly after the fine-level of 2.5%. Above this level of fines, incentives do not have a major effect on the magnitude of pollutionreduction. Below this level, we see increasing incentives lead to lower reduction in pollution. The figure suggests we keep fines above 2.5% for optimum reduction in pollution. Looking at changes in the cluster’s capital-assets, graphs b, c, d in Fig. 3, we see that capitalstocks are maximised when fines are kept below 2.5%, and incentives are kept at the highest level. Coming to innovation rates, looking at graph f in Fig. 3, we see total innovation success is highest when fines are kept at zero, and incentives at 25%. This combination of incentives and fines also maximises cluster-size, as per graph a. For radical project success, it is better to keep both fines and incentives high (see graph e in Fig. 3). Finally, increasing fines and incentives both cause the ratio of pollution to financial capital to decline. The decline due to increasing incentives is more gradual than due to increasing fines. Fig. 2 suggests we keep both incentives and fines at the highest level to optimise for this ratio. 5. Discussion Over the past decade, the field of innovation policy has seen increasing emphasis on studying the design of policies promoting green-clusters, clusters that undergo decoupled green-growth. There is however, limited knowledge on which policies can instigate green-growth in clusters. To contribute to this knowledge gap, we created an agent-based model. This ABM stitches macro-level and micro-level perspectives together, enabling us to see how a cluster decouples through agents’ green-innovation. Our model is a virtual laboratory that allows for experimentation and projection – by modifying the set-ups of extant instruments or through the introduction of new instruments. We can for instance, explore the effects of reducing the pollution threshold above which agents are fined. We can also, with some changes in the model, explore the effects of other instruments such as Fig. 1. Variation of pollution in the cluster, with increasing levels of incentives and fines. R. Kamath et al.