Technological Forecasting & Social Change 172 (2021) 121060 Contents lists available at ScienceDirect Technological Forecasting & Social Change journal homepage: www.elsevier.com/locate/techfore Regulation at the source? Comparing upstream and downstream climate policies Joël Foramitti a, b, *, Ivan Savin a, c, Jeroen C.J.M. van den Bergh a, b, d, e a Institute of Environmental Science and Technology, Universitat Autónoma de Barcelona, Spain b Institute for Environmental Studies, Vrije Universiteit Amsterdam, the Netherlands c Graduate School of Economics and Management, Ural Federal University, Yekaterinburg, Russian Federation d ICREA, Barcelona, Spain e School of Business and Economics, Vrije Universiteit Amsterdam, the Netherlands A R T I C L E I N F O A B S T R A C T Keywords: Climate policies can be applied either upstream, where fossil fuels are extracted, or downstream, where emis­ Agent-based modeling sions are generated. Specific policy instruments can be defined for either level, and can take the form of a price Carbon leakage signal such as through a tax, or a quantity limit such as through direct regulation or a permit market. In this Carbon tax study, we present an agent-based model to compare the performance of these different instruments and regu­ Climate policy lation levels. Since policy coverage is often limited, i.e. not all firms being under the regulator’s control, we also Emission trading Quota examine the impact of incomplete coverage on relative policy performance. Our analysis shows that only up­ stream regulation leads to an increase in fossil fuel prices, which is benefitial under limited coverage as it also affects firms not directly affected by the policy instruments; that prices under quantity-based regulation can decline after an initial peak, stabilizing at a lower level than under the tax; and that direct regulation is more efficient when applied upstream. 1. Introduction et al. (2021) further show that a tax could be preferable to a permit market, as the unstable price of permits may favor emission-intensive The combustion of fossil fuels is the primary cause of global carbon producers and create windfall profits. emissions. To meet the target of limiting global warming to 1.5∘◦ C, the While climate policy has mostly been applied downstream, recent use of coal, oil, and gas needs to be largely phased out within the next studies have called for more attention to the upstream level (Collins and three decades (Masson-Delmotte et al., 2018). Regulatory measures can Mendelevitch, 2015). According to Lazarus and van Asselt (2018), up­ be applied at different levels in the system: upstream, which is where stream policies could reduce the overall costs of mitigation as they carbon first enters the economy through either extraction or imports of ‘widen the mitigation cost curve’, prevent carbon lock-ins (Seto et al., fossil fuels; midstream, where these fuels are refined and transported; or 2016), and increase the political pressure for climate action. Sinn (2012) downstream, where the actual emissions take place through the com­ suggests that upstream policy could prevent a ‘green paradox’, where bustion of fossil fuels (Goulder and Schein, 2013). the anticipation of downstream policies could lead to higher upstream On both the upstream and downstream level, policy makers can either production levels. And Piggot et al. (2018) argue that supply restrictions apply a market-based mechanism – e.g. through a carbon tax or a permit could be effective even if not all fuel-producing countries participate. trading system – or regulate the production levels of firms directly – e.g. Another important issue is the risk of ‘carbon leakage’, i.e. a situa­ through a quota on the extraction, import, or use of fossil fuels. Applied to tion where the emission reduction in a covered sector causes a rise of the downstream level, market-based mechanisms have the advantage emissions elsewhere. While the existence of such leakage is undisputed, that the regulator requires no perfect information about firms’ abatement its magnitude is debated (Collins and Mendelevitch, 2015). The study costs to achieve an efficient outcome (Perman et al., 2003). Foramitti of Erickson and Lazarus (2018) suggests that upstream policies address * Corresponding author at: Institute of Environmental Science and Technology, Universitat Autónoma de Barcelona, Spain. E-mail address: joel.foramitti@uab.cat (J. Foramitti). https://doi.org/10.1016/j.techfore.2021.121060 Received 27 November 2020; Received in revised form 24 June 2021; Accepted 18 July 2021 Available online 5 August 2021 0040-1625/© 2021 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). J. Foramitti et al. Technological Forecasting & Social Change 172 (2021) 121060 model are heterogeneous in terms of their production factors and trading behavior. Over time, they change their production level based on expected demand and adapt their mark-up based on experienced success. They submit market orders for fuels and permits, and adapt their trading prices based on experience. Downstream manufacturers further adopt less emission-intensive technology based on the costs of regulation and fossil fuels. The approach of this study is exploratory. Our model is not calibrated to a particular country or period, and instead looks at general dynamics under a wide range of parameter values and evaluation criteria. In this way, we identify potential drawbacks of each policy and discuss the underlying causal mechanisms. This puts our study in line with the approach of ‘reflexive possibilistic modeling’ (Edmonds and Aodha, 2019), which suggests to use ABMs for “identifying some of the possible ways a policy can go wrong (or indeed go right)”. The remainder of this article is organized as follows. Section 2 provides a detailed description of the model. Section 3 introduces the numerical experiment and the evaluation criteria of our analysis. Section 4 presents and discusses simulation results. Section 5 provides concluding remarks. Fig. 1. Overview of agents (white) and markets (black). 2. Model description carbon leakage better as supply restrictions can lead to higher prices An overview of the model structure is given in Fig. 1. The model is that will decrease demand outside the regulators jurisdiction. Fæhn made up of two industry sectors. One consists of ‘suppliers’ (S) of fossil et al. (2017), in contrast, find for the case of Norway that upstream fuels, which can represent coal mines, oil wells, or distributors that regulation would lead to higher leakage as domestic fuels could easily import fossil fuels. The other consists of ‘manufacturers’ (M), which use be substituted. fuels in order to create goods and emit carbon as a byproduct. Firms in The models that are used for such comparisons of downstream and both sectors are separated into regulated and unregulated firms, with the upstream regulation are mostly based on the assumptions of rational and latter not being covered by the regulator’s policies and emission target. representative agents with perfect knowledge. The real economy, in The following subsections provide a detailed description of the model. contrast, is a complex system characterized by unpredictable events, boundedly rational behavior, and heterogeneity (Arthur, 1999; Kirman, 2.1. Policy scenarios 2006; Mullainathan and Thaler, 2000). Another limiting assumption in many models is that policies are evaluated upon a single measure of The policy maker’s aim is to reduce overall emissions to the target costs.1 The above-mentioned study of Fæhn et al. (2017) even combines level e∗ . We test three policy instruments to reach this target: a carbon welfare costs on the downstream with costs of foregone profits on the tax, permit trading, and direct regulation through a uniform quota. Each upstream into a single measure. of these are applied either upstream, where the regulator will regard the Here we present an agent-based model (ABM) to compare the rela­ emissions embodied in fuels, or downstream, where the regulator will tive performance of different policy instruments and regulation levels regard the actual emissions caused by the use of fuels. We compare these for the reduction of emissions. ABMs are increasingly applied to the policies against a baseline scenario without any policy intervention. The analysis of climate policy (Castro et al., 2020), but have not yet been three distinct instruments work as follows: used for the comparative analysis of upstream and downstream regu­ lation. The method allows for the exploration of economic dynamics 1. Under carbon taxation, an emission price pet is set directly by the based on the continuous interaction of individual agents that have to policy maker at the beginning of each round. All covered firms in the make decisions under limited information about future prices and de­ target sector have to pay the same price. The quantity of emissions is mand (Farmer and Foley, 2009). Furthermore, ABMs can take into ac­ not defined directly, but is an outcome of market dynamics. count heterogeneous technological options and behavior among firms. 2. Under permit trading, the policy maker each round auctions a We compare seven different scenarios, which include no policy, a limited amount of permits u∗t to the targeted sector, as described in carbon tax, a permit market, and direct regulation through a uniform Section 2.4. Firms have to submit a permit for each unit of (embodied quota, with each of the policy scenarios being applied for either or actual) emissions they cause. This regulates quantity directly, extraction (upstream) or emissions (downstream). The aim is to under­ while the emission price is a result of trading dynamics.2 stand how these scenarios compare under bounded rationality, hetero­ 3. Under direct regulation, the policy maker sets a uniform quota that geneity, and dynamic markets. In the second part of the paper, firms in imposes a maximum emission limit eσi,t on a covered firm i. Similar the model are separated between covered and not covered, with only the former being under the regulators control. This distinction is imple­ to permit trading, this creates a direct quantity limit, although with mented to explore the possibility of carbon leakage, i.e. to compare the a different allocation, and without firms having to pay for their performance of the instruments when a share of companies does not share. This particular quota affects all covered firms equally (see have to obey national climate regulations. Appendix A.1), and thus does not favor more efficient firms. Our model builds upon Foramitti et al. (2021). It consists of fossil fuel suppliers, manufacturers of final goods, and consumers. The firms in the 2 Note that this model simplifies the permit trading to the interaction be­ tween the regulator and the firms to keep the complexity of the model manageable. The auctioning mechanism still captures two central aspects of 1 Common cost concepts include “change in GDP, change in consumption, permit trading. First, that permit prices are a result of firms’ willingness to pay. change in welfare, energy system cost, and area under marginal abatement cost And second, that firms with a higher willingness to pay receive a larger share of (MAC) curve” (Paltsev and Capros, 2013). permits. 2 J. Foramitti et al. Technological Forecasting & Social Change 172 (2021) 121060 The stringency parameters of these policies (pet , u∗t , eσi,t ) are gradually ( ) increased until they reach the target e∗ , as described in Appendix A.1. Bs,t+1 = Bs,t + qps,t ∗ β (4) The coverage of the policies (i.e. the fraction of firms that are regulated) can further be limited, with the fraction of regulated firms given by the Firms set their sales price pi,t to cover their production costs Bi,t and their parameter φ. This can either represent a situation where only specific emission and fuel costs cei,t . They further add a mark-up rate αi,t to their sectors are covered,3 or one where firms share the market with foreign costs per unit of production. competitors that are outside the regulator’s jurisdiction. ( ) ( ) pi,t = Bi,t + cei,t ∗ 1 + αi,t (5) 2.2. Order of events Firms emission and fuel costs cei,t are defined in Eq. (6), where pei,t relates The agents’ actions follow discrete time-steps t = 1,2,⋯,T, which are f to the price of emissions, pi,t refers to the price of fuels, and Am,t describes meant to roughly represent months. Each round is characterized by the the manufacturers fuel and emission intensity. For suppliers, the price of following chain of events: emissions regards the embodied emissions of their fuels. { e 1. The policy maker updates its climate policy. ps,t Supplierss 2. Consumption good firms form their production goals and order fuels. cei,t = ( e f ) (6) Am,t ∗ pm,t + pm,t Manufacturersm 3. Fuel suppliers form their production goals. 4. If a permit market is in place, covered firms trade permits. The mark-up rate αi,t reflects the dynamics of a ‘customer market’ where 5. Suppliers produce fuels and sell them to manufacturers on the fuel firms compete against each other over their market share ψ i,t . As market. 6. Manufacturers produce goods and sell them to consumers on the described in Dosi et al. (2010), they set a higher profit margin when they goods market. are successful - meaning that their market share is growing - and reduce 7. Manufacturers decide whether to invest in abatement technology. it when they are not. The magnitude of this adaptation is given by the 8. Manufacturers change towards more competitive fuel suppliers. factor ϑ. ( ) ψ − ψ i,t− 2 2.3. Production αi,t = αi,t− 1 ∗ 1 + ϑ ∗ i,t− 1 (7) ψ i,t− 2 At the beginning of each round t, every firm i ∈ S ∪ M (denoting both 2.4. Permit market and quotas fuel suppliers s ∈ S and manufacturers m ∈ M) sets their production goal qdi,t . Similar to Dosi et al. (2010), firms are demand-driven and myopic. There are two policy scenarios where the policy maker requires This means that they base their goal on expected demand D ̃ i,t . They regulated firms j to submit permits for their production (see Section 2.1). further add a desired inventory rate I and subtract the remaining in­ One targets suppliers (j ∈ S), forcing them to submit an extraction ventory from last round qi,t− 1 . permit for each unit of fossil fuel they introduce to the market. The other regulates manufacturers (j ∈ M), forcing them to submit an emission ̃ i,t ∗ (1 + I) − qi,t− qdi,t = D 1 (1) permit for each unit of fossil fuel they combust in their production process. Firms production qpi,t can be restrained by multiple factors. In case of Permits are distributed at the beginning of each round through a direct regulation, by the quota eσi,t . In case of permit trading, by their discriminatory auction.4 Every round, firms have to submit bids that are inventory of permits ui,t . And for manufacturers, by their inventory of accepted by the regulator until u∗t is reached or no more bids are left. fossil fuels fm,t . Manufacturers’ restrictions further depend on their fuel Permits are then sold at the respective bid-price, and unsold permits are intensity Am,t , which describes the amount of fuel needed to produce one kept for next round’s auction. Firms bidding volume is based on their unit of output. Units are normalized so that the combustion of one unit desired amount of permits udj,t , which depends on firms monthly pro­ of fuel leads to one unit of emissions, which means that Am,t also rep­ duction goal qdj,t , their rationing time t∗ , and their existing inventory of resents firms’ emission intensity. Finally, firms also ration their use of permits uj,t− 1 . permits and fuels over the following t∗ rounds to avoid sudden ⎧ shortages. ⎨̃ Ds,t ∗ t∗ − us,t Supplierss ⎧ ( ) d uj,t = (8) d σ us,t ⎩D̃ m,t ∗ t∗ ∗ Am,t − um,t Manufacturersm ⎪ min qs,t , es,t , t∗ ⎪ ⎨ Supplierss p qi,t = ( ) (2) ⎪ eσ Firms have an idiosyncratic emission price pej,t , which reflects how much ⎩ min qdm,t , m,t , um,t ∗ , fm,t ∗ ⎪ Manufacturersm Am,t Am,t ∗ t Am,t ∗ t they think a unit of emissions is worth and how much they are willing to pay for a permit at the auction. Depending on whether firms are able to Firms produce and trade according to the order of events in Section 2.2. trade their desired amount (success) or not (failure), they adapt their When they try to sell their goods at the respective market, they offer emission price based on the adaption rate μj for future rounds.5 their current inventory qi,t , which consists of both their latest production ( ) qpi,t and past inventory qi,t− 1 . { e pj,t − μj ∗ 1 + pej,t Success pej,t+1 = ( ) (9) qi,t = qpi,t + qi,t− 1 (3) pej,t + μj ∗ 1 + pej,t Failure Suppliers’ production costs Bs,t increase by a factor β with every unit of fossil fuel they extract. This reflects that fuel supplies become more scarce and difficult to extract as reservoirs deplete. 4 A comparison of discriminatory pricing auctions with uniform pricing auctions and grandfathering (i.e. no auction) is presented in Foramitti et al. (2021). 3 5 The European emission trading system, for example, only covers around A factor 1 is added to Eq. (9) to avoid the permit price being locked-in if it is 45% of European emissions (World Bank Group, 2019) close to zero. 3 J. Foramitti et al. Technological Forecasting & Social Change 172 (2021) 121060 Hence, firms that want to buy permits increase their emission price until 2.6. Fossil fuel market they receive their desired amount of permits. If their trades are suc­ cessful (i.e. they received their desired amount of permits), they reduce The fossil fuel market is organized through an order-based system. d the bidding price in the hope to spend less on future bids.6 Manufacturers calculate their desired amount of fuels fm,t in the same Under direct regulation, firms do not have to pay for permits, but still way as their desired amount of permits in Eq. 8: have an incentive to charge consumers an emission price if there is more d fm,t ̃ m,t ∗ t∗ ∗ Am,t − fm,t =D (18) demand than they are allowed to produce. In other words, if firms are not allowed to increase production in reaction to high demand, they Each manufacturer m has a preference dm,s,t ∈ [0, 1] for each supplier s raise prices instead. This represents a scarcity rent (Kalkuhl and Brecha, that defines what percentage of their desired fuels they will order from 2013). They thus apply the same adaption as in Eq. 9, with their con­ it. A supplier’s demand Ds,t thus becomes the sum of these orders: ditions of success being defined as follows: ∑ { σ Ds,t = d fm,t ∗ dm,s,t (19) es,t Supplierss ̃ Di,t ≤ (10) m∈M eσm,t ∗ Am,t Manufacturersm Suppliers try to produce enough to meet this demand, as described in Section 2.3. Their sales and unfilled demand are calculated like on the 2.5. Consumption good market consumption good market (Eqs. (16) and (17)). If there is unfilled de­ mand, it will be reduced from all orders in an equal share. Each firms’ The goods market follows the same evolutionary dynamic as in market share ψ s,t is defined as their share in total sales: Foramitti et al. (2021), following Dosi et al. (2010), where demand gradually moves towards more competitive producers. Competitiveness q∗s,t km,t is given by a firm’s sales price pm,t and unfilled demand from last ψ s,t = ∑ ∗ (20) s∈S qs,t g round lm,t− 1 (Eq. (17)). The first term implies that a firm’s competi­ tiveness falls with increasing sales prices. The second term ensures that At the end of each round, manufacturers adapt their list of preferred firms lose customers when they are unable to fulfill their demand. suppliers. Similar to the change of demand on the consumption good market, each supplier’s competitiveness ks,t is calculated as in Eq. (11). km,t = − pm,t − lm,t− 1 (11) The preferences dm,s,t of each manufacturer then change based on the replicator dynamics in Eq. (12) with the adaption speed χ S . These factors of competitiveness define the evolution of firms’ market shares ψ m , where χ denotes how fast consumers shift towards more competitive firms. 2.7. Abatement ⎛ ⎞ As in Foramitti et al. (2021), manufacturers can decide to adopt a M km,t − kt ⎠ ⎝ ψ m,t = ψ m,t− 1 ∗ 1 − χ ∗ (12) new technology that will reduce their emission intensity and increase kt their production costs. These technological options could represent the installation of more emission-efficient machines, a shift from fossil fuels The average competitiveness kt is given by the weighted competitive­ to renewables as an energy source (e.g., electricity), or new production ness of each firm, using the last rounds market shares ψ m,t− 1 as weights. routines that reduce fossil fuel dependency. ∑ Each firm has a different set of possible technological options x that kt = ψ m,t− 1 ∗ km,t (13) allow for a particular reduction in emissions ax at an extra cost bx per unit of production. The marginal abatement costs of this technological m∈M The level of total demand follows a simple declining curve that depends step, i.e. the additional production costs of emitting one unit less, are g defined as: on the average price pt . This means that consumers tend to buy less of the good if the overall price rises. The price sensitivity of demand is bx cλm = (21) given by the factor γ. ax (14) g Dgt = D0 ∗ e− Every round, firms examine the next possible technological option x γ∗pt with the lowest cλm . A technological improvement is implemented by This total demand for goods is then allocated according to firms’ market comparing the marginal costs of abatement to the sum of the price of shares: f emissions pem,t and fuels pm,t , which can be seen as the marginal damage Dm,t = ψ m,t ∗ Dgt (15) costs of causing one unit of emission.7 Since technological investment is a long-term decision involving uncertainty about future cost of abate­ Firms actual sales q∗m,t are then either limited by their demand or their ment, firms add an idiosyncratic profitability target ηm to this condition inventory: to reflect their different risk attitude. ( ) q∗m,t = min Dm,t , qm,t (16) f cλm ∗ (1 + ηm ) < pem,t + pm,t (22) If firms have produced too little, they are left with a certain amount of Under permit trading, a manufacturer decides to invest in technological unfilled demand lm,t that will translate into reduced competitiveness in improvements once the sum of the price of emissions and fuels is higher the following round: than the cost of abatement. This reduces their emission intensity, which lm,t = Dm,t − q∗m,t (17) in turn affects their demand and trading price for permits in the next round. Manufacturer’s trading and abatement behavior thus reflects a balance between the cost of permits and the cost of abatement options. 6 7 If firms have more permits then they want or if they manage to buy all The cost of fuels is covered here, which is not the case in textbook treat­ available permits, they will treat that round as successful. ments of abatement costs. 4 J. Foramitti et al. Technological Forecasting & Social Change 172 (2021) 121060 Fig. 2. Quantity and price dynamics for a single run under full coverage. 3. Numerical experiment 1. Technology adoption: The share of abatement10 that is achieved through the adoption of more emission-efficient technology (tech­ In our experiment, we simulate a setting of 30 upstream and 30 nological abatement). downstream firms over the time-span of 200 rounds. The model initiates 2. Compositional change: The share of abatement that is achieved over 50 rounds, then a climate policy is linearly introduced over the through a restructuring of the sector that gives emission-intensive following 100 rounds,8 followed by 50 more rounds where the policy is firms a lower market-share. held at a constant level. The policies are compared under the same level 3. Production decline: The share of abatement that is achieved through a of effectiveness, i.e. their stringency is set so that they all meet the reduction of firms’ production levels. emission target e∗ . The permit markets reach this target automatically, 4. Sales price: The average sales price pt,m of downstream manufacturers. while the emission tax and quota are calibrated. 5. Consumer impact: The average financial burden on consumers, We repeat this experiment 64,000 times to look at a wide range of assuming that policy revenue is recycled and given back to con­ different parameter combinations, which are described in Appendix A.1. sumers. Calculated as the average sales price minus the policy rev­ The code for the simulation is written in Python 3 and available on enue per good. GitHub.9 The parameter variation is based on the sampling scheme of Saltelli et al. (2010) and operationalized through the Python package Since policies are compared with the same effectiveness, the first SALib (Herman and Usher, 2017). The same package is used to calculate three criteria can be seen as different formulations of efficiency. High Sobol sensitivity indices (Sobol, 2001), which are presented in technology adoption and compositional change regard the emission- Appendix A.3. efficiency of the economy (i.e. average emissions per unit of produc­ The baseline and the six policy scenarios are compared between two tion), while a low production decline represents an efficient economy in distinct settings (see Section 2.1). In the first setting, all firms are regard to economic output. The other two criteria - sales price and regulated by the climate policy (full coverage, Section 4.1). In the sec­ consumer impact - are desirable to be low in regard to the equity impacts ond, only a fraction of firms is regulated and the rest is outside the and political feasibility of the policy. regulator’s jurisdiction (limited coverage, Section 4.2). In addition, we control for the following six indicators in our model. To analyze and compare the relative impact of each instrument, we These help us to identify the causal link between a certain policy and its consider five different evaluation criteria. Each criterion regards the performance along the evaluation criteria. Their exact definition is also state of the model at the last ten rounds of the simulation, and describes given in Appendix A.2. The indicators are as follows: how well a given policy performs relative to the other six scenarios. Their mathematical definition can be found in Appendix A.2. The 1. Emissions: The overall emissions of the downstream sector. criteria are as follows: 2. Emission price: The average emission price pet,j of regulated firms. 3-4 Profit shares: The amount of profits in the downstream and up­ stream sector. 5-6 Market concentrations: The distribution of market shares in the 8 This means that the policy is strengthened over time along a linear path downstream and upstream sector. over 100 rounds to avoid extreme changes in the economy from one round to another. This is in line with climate policy design in countries like Argentina, Canada, and South Africa (World Bank Group, 2019). 9 Link to the repository: https://github.com/JoelForamitti/UvsD_ABM 10 As derived in Appendix A.2, abatement is decomposed into three contrib­ uting factors. This results in three shares which sum up to the total amount of abatement that is equal among the policy scenarios (as they reach the same emission target). 5 J. Foramitti et al. Technological Forecasting & Social Change 172 (2021) 121060 4. Results and discussion regulation can create a competitive advantage for less emission- efficient producers. The consistent and usually higher price of the tax Results are presented in two parts. In Section 4.1, we explore the case thus creates a stronger advantage for emission-efficient firms. of full coverage, meaning that all firms in the target sector are covered As the uniform quota does not favor more emission-efficient pro­ by the climate policy. This demonstrates the differences between the ducers, compositional change and market concentration are very small policies when no carbon leakage is possible, e.g. in a closed economy for direct regulation on the downstream level. However, the policy with full coverage, or under a global agreement between connected does not lead to a stronger production decline as the quota increases economies. In Section 4.2, we consider the contrasting case of limited scarcity and stimulates more technology adoption, thus increasing coverage. In both parts, there is one sub-section that shows dynamic emission-efficiency. results over time for a single run of the model, and a second one that While other types of quotas could increase the efficiency of quotas on presents average results over multiple runs and varied parameters. A the downstream level, our scenario demonstrates that the inefficient sensitivity analysis can be found in Appendix A.3. selection (i.e. low compositional change) of direct regulation is not a disadvantage when applied upstream. On the contrary, compositional 4.1. Full coverage change and market concentration are slightly higher than under up­ stream permit trading. This is because the less efficient selection of 4.1.1. Single run dynamics upstream firms through the quota can lead to higher fuel prices and thus Fig. 2 shows the quantity and price dynamics over time for a single create a stronger selective effect (i.e. high compositional change) run, using mid-point parameter values from Table Appendix A.2 . We downstream. can see that all policies, while following different paths, end up leading Since there is no policy revenue under direct regulation, the quota to a similar reduction of fuels, emissions, and sales, as well as a similar creates particularly high profit rates and consumer impacts. In that increase in goods prices. The model thus demonstrates that, under full sense, direct regulation is very similar to a permit system where per­ coverage, a reduction of emissions will lead to a similar change in prices mits are allocated for free, only that firms have no influence on the independent of the policy instrument and whether it is applied upstream distribution of permits. The tax, due to its higher price, causes higher or downstream. mark-ups and thus more profits than the permit market in its target One key difference between the two regulation levels is that upstream sector.11 Finally, consumer impacts are lower for both tax and permit policies lead to an increase in fuel prices, while downstream regulation trading if these are applied downstream, as the higher emission price leads to a price close to or even below that under no policy. This is (see Section 4.2.1) leads to more policy revenue which is then recycled because when downstream regulation reduces the amount of emissions, it back to the consumers. automatically also reduces the demand for fossil fuels. This is good for climate mitigation, in the sense that it reduces the profits that can be 4.2. Limited coverage made from selling fossil fuels, but it is also problematic in the sense that firms who are not covered by the downstream policy could get easy ac­ In the following subsections, we assume that only a fraction φ (see cess to cheap fuels, which can lead to carbon leakage (see Section 4.2). Table Appendix A.2 in Appendix A.1) of firms on the market are covered Fig. 2 further shows that quantity-based regulation (i.e. permit by the respective climate policy. This allows for the possibility of carbon trading and direct regulation) displays a dynamic where prices stabilize leakage, i.e. that the climate regulation in the covered sector causes at a lower level than the tax or even decline after an initial peak emissions to increase elsewhere. (henceforth referred to as ‘overshoot-decline dynamic’). This dynamic has been identified as a key difference between downstream tax and 4.2.1. Single run dynamics permit trading in Foramitti et al. (2021), and can be explained by the fact In Fig. 4, we can see the emissions over time, separated between that successful abatement makes production more emission-efficient, embodied (upstream) and caused emissions (downstream), as well as which in turn leads to less demand for emissions and fuels, and thus between covered and total (covered plus not covered) emissions. We see lower emission and fuel prices. that regulation on one level affects outcomes on the other only weakly, i. Finally, upstream regulation tends to reach the same abatement with e. reaching the local target downstream reduces upstream extraction a lower emission price, particularly in the case of taxation. This is only by a small amount, and vice versa. because upstream regulation creates higher profit rates as firms set their Emissions under quantity-based regulation fluctuate strongly as they mark-up as a percentage of their costs per unit of production. Down­ mirror the fluctuation in emission prices. This is because the competition stream firms then have to pay this additional mark-up for every unit of between firms that are covered and those who are not causes emission fuel, to which they than also apply their own mark-up rate. This means prices to change fast over time. Under the full coverage setting, emission that an upstream policy can apply the same pressure as a downstream price changes only affected firms’ demand because of differences in policy with a lower emission price. emission intensity and changes in overall demand. Under limited coverage, in contrast, there is an additional dynamic as covered firms 4.1.2. Average results over multiple runs can loose market shares to firms that are not regulated. In Fig. 3, the evaluation measures and additional indicators are Total emissions appear slightly lower for upstream regulation. This is presented as average results over all parameter combinations tested. As because of the fuel price increase that only happens under upstream mentioned in Section 3, the six policy scenarios are compared under regulation, as discussed in Section 4.1.1. Under limited coverage, fuel equal effectiveness, i.e. the levels of stringency are set so that all policies price changes affect all firms (both covered and not covered), and thus reach the same emission target. Like already observed for a single run, also incentivizes emission reduction outside the reach of the regulator. production decline and sales price increase are fairly similar (Fig. 3). Downstream regulation, in contrast, only creates such incentives for Fig. 3 further shows that technology adoption tends to be slightly covered firms. higher for quantity-based policies. This happens due to the unstable emission prices seen in Section 4.2.1, which leads to a temporary price levels above that of the carbon tax until enough technological im­ 11 In Foramitti et al. (2021), it has been shown that permit trading can create provements have taken place to drive the price down again. This, in higher profit rates than a tax because firms could receive permits at a low price turn, results in lower scarcity of goods, and thus lower sales prices. and make profit by selling them to competitors. This is not the case in current Regarding compositional change on the goods market, both tax policies model because permit trading has been simplified so that permits are only perform best. This is because the unstable prices of quantity-based auctioned and not traded between firms. 6 J. Foramitti et al. Technological Forecasting & Social Change 172 (2021) 121060 Fig. 3. Policy evaluation criteria (top panel) and additional indicators (bottom panel) under full coverage. Values indicate the performance of each scenario relative to the others (see Appendix A.2). Error bars report standard deviations. Fig. 4. Emissions over time under limited coverage. Note: The y-axis describes the percentage of emissions relative to the emission level before the introduction of the policy. 7 J. Foramitti et al. Technological Forecasting & Social Change 172 (2021) 121060 Fig. 5. Policy evaluation criteria (top panel) and additional indicators (bottom panel) under limited coverage. Values indicate the performance of each scenario relative to the others (see Appendix A.2). Error bars report standard deviations. 4.2.2. Average results over multiple runs mainly reduces emissions by driving covered firms out of the market. Fig. 5 presents the same evaluation measures and additional in­ Small changes in the emission price can thus have a strong competitive dicators as in Section 4.1.2, but this time for the setting of limited effect, which can be seen in the high standard deviation for the emis­ coverage. Note that these measures regard all firms (covered and not sion price in Fig. 5. covered). In line with the discussion above, we can observe that upstream regulation leads to slightly less emissions and more compositional 5. Conclusions change, as the increase in fuel prices affects all firms downstream (i.e. even those that are not within the regulator’s reach). This study presented an agent-based model to compare the perfor­ We can further see that quantity-based regulation leads to more mance of upstream and downstream regulation for climate change technology adoption when applied downstream. This is due to the same mitigation, considering the instruments of taxation, permit trading, and ‘overshoot-decline’ dynamic introduced in Section 4.1.1. The reason direct regulation through a uniform quota. The model takes into account that this dynamic does not appear here for upstream regulation is that heterogeneous agents using heuristic decision rules on constantly fuel suppliers cannot innovate, while the policy impacts are diffused changing markets, as well as the difference between full and limited with the unregulated fuel supply sector (i.e. overall fuel prices fluctuate coverage. Results were presented for multiple runs with a wide variation less because only part of the fuel suppliers are affected) before it reaches of parameter values, and evaluated upon multiple measures. downstream firms. In our analysis, we compare the performance of each policy and Market concentration is increased by policies only within their regulation level under an equal level of effectiveness. This suggests that regulation level, i.e. upstream policies increase upstream market con­ none of the tested instruments can be ruled out as an adequate instru­ centration and downstream policies downstream market concentration. ment to reduce emissions. However, we do identify several dynamics This can be understood through the competition between firms. As that lead to relevant differences between the policies. firms in the regulated sector become less competitive in comparison to In line with Foramitti et al. (2021), we show that quantity-based unregulated firms, their market shares will shrink and the overall regulation can lead to an overshoot in prices followed by a decline market concentration will increase; and succesful firms will apply after successful abatement. In comparison to a tax, this dynamic can lead higher profit rates. to a higher level of technological adoption and production costs, while Due to this competitive effect, the emission price necessary to reach compositional change towards emission-efficient firms is smaller as a the target is very low. Essentially, under limited coverage, a policy low emission price can make such firms less competitive. 8 J. Foramitti et al. Technological Forecasting & Social Change 172 (2021) 121060 A key difference that is found between the two regulation levels is that Some suggestions for future research are to account for different fuel downstream policies reduce the demand for fuels, which means that only types and trading strategies, emissions from fossil fuel extraction, agent upstream policies lead to an increase in fuel prices. This can be both learning (Yu et al., 2020), and additional abatement options like beneficial and detrimental for the success of climate policy. On the one end-of-the-pipe technologies. Further features like heterogeneous con­ hand, it can reduce the profits that can be made from selling fossil fuels, sumers, inter-sectoral interactions, finance, and labor markets, could on the other it allows firms who are not covered by the downstream allow for an assessment of impacts on additional criteria like economic policy to buy fuels at a cheap price - increasing the risk of carbon leakage. stability and equity. We further demonstrate that the level of regulation matters partic­ ularly for direct regulation. Applying a uniform quota downstream CRediT authorship contribution statement creates little abatement through compositional change as it does not advantage more emission-efficient firms. However, when applied up­ Joël Foramitti: Conceptualization, Software, Writing – original stream, this inefficient selection leads to an increase in fuel prices, which draft. Ivan Savin: Conceptualization, Supervision, Writing – review & can only strengthen the shift towards more emission-efficient firms editing. Jeroen C.J.M. van den Bergh: Conceptualization, Supervision, further downstream. Writing – review & editing, Funding acquisition. These insights complement the results from traditional models like Erickson and Lazarus (2018) and Fæhn et al. (2017) by identifying po­ Acknowledgments tential policy dynamics that are overlooked under the assumptions of rational and representative behavior and economic equilibrium. Further This study has received funding through an ERC Advanced Grant dynamics could be identified under additional assumptions, and cali­ from the European Research Council (ERC) under the European Union’s bration towards particular cases would be necessary to test their likeli­ Horizon 2020 research and innovation programme (grant agreement no. hood. The flexible structure and open-source nature of our model makes 741087). I.S. acknowledges financial support from the Russian Science it very suitable for such extensions. Foundation (RSF grant number 19-18-00262). Appendix A A1. Parameter values This section presents the parameter values used in our numerical experiment. Table Appendix A.1 presents the fixed parameters, which are being held constant throughout the whole simulation. Table Appendix A.2 presents variable parameters, which are varied for each of the 64,000 simulation runs based on the sampling scheme of Saltelli et al. (2010). There are further three policy parameters pet , u∗t , and σ t , which are gradually changed over the policy implementation period T ∗ , with their final values calibrated to reach the emission target e∗ . In the first round of permit trading, the amount of permits u∗t is multiplied by t∗ to avoid extreme fluctuations in the initial rounds. Under direct regulation, the individual quotas of each firm are calculated based on their emissions before the start of the policy: eσi,t = σt ∗ ei,T0 − 1 (A.1) Firms technological abatement options are calculated as follows. Firms have i = 1, ⋯, Nλ technological options, each with an abatement potential of a = λ/Nλ and a marginal abatement cost of cλi = bi /a = θ ∗ a ∗ i. This means that firms’ marginal abatement costs will linearly increase with every abatement step that they take. A2. Performance criteria The definition Yn,z of each measure n and scenario z is given in Table Appendix A.3. The policy evaluation criteria (n = 0 − 5) and additional indicators (n = 6 − 11) are described in Section 3). They are calculated as a sum of the upstream (j ∈ S) or downstream (j ∈ M) sectors activity in the last ten rounds of the simulation, as given by the function S: ∑ T Nj ∑ ( ) S yj,t = yx,t (A.2) t=T− 10 j=1 The relative measures Cn,z that are shown in Section 4ww describe the results for each scenario z and measure n in relation to the other scenarios as given in Eq. Appendix A.3. This means that the eleven measures describe relative performance between the scenarios and sum up to 1. Yn,z Cn,z = ∑6 ⃒ ⃒ (A.3) ⃒Yn,z ⃒ z=1 The first three criteria critera are based on the decomposition of downstream abatement as put forward in Foramitti et al. (2021, Appendix 2). A manufacturers’ change in emissions from round 0 to round t can be decomposed into changes in production level and changes in emission-intensity, using the functions xt = (xt +x0 )/2 and Δxt = xt − x0 . Δem,t = em,t − em,0 = qm,t ∗ Am,t − qm,0 ∗ Am,0 (A.4) = Δqj,t ∗ Am,0 + ΔAm,t ∗ qj,0 + Δqm,t ∗ ΔAm,t = Δqm,t ∗ Am,t + ΔAm,t ∗ qm,t 9 J. Foramitti et al. Technological Forecasting & Social Change 172 (2021) 121060 The change in production Δqm,t can further be decomposed further into contributions from a shift of relative shares of production ρm = qm,t /Qt within ∑ m the sector and a decline in total production Qt = Nm=1 qm,t . Δqm,t = ρm,t ∗ Qt − ρm,0 ∗ Q0 = Δρm,t ∗ Qt + ΔQt ∗ ρm,t + Δρm,t ΔQt (A.5) = Δρm,t ∗ Qt + ΔQt ∗ ρm,t Consumer impact describes average sales price minus the regulators’ revenue per unit of goods sold. The revenue Ri,t from each firm is given by either the tax or permit payments of each round: ⎧ ⎪ ⎨0 ⎪ ⋯No policy or direct regulation Ri,t = ei,t ∗ p e t ⋯Tax (A.6) ⎪ ⎪ ⎩ pui,t ∗ uti,t ⋯Permit trading Finally, the profit πi,t in criterion 10 and 11 describes the sum of revenue from sales minus the sum of expenses for production, fuels, and regulation. { 0 ⋯Suppliers π i,t = q∗i,t ∗ pi,t − qi,t ∗ Bi,t − Ri,t − (A.7) em,t ∗ Am,t ⋯Manufacturers A3. Sensitivity analysis To understand the sensitivity of the results towards each parameter, we perform a Sobol sensitivity analysis (Saltelli et al., 2010; Sobol, 2001) for each measure and varied parameter. The calculated first-order Sobol Sensitivity Indices are presented in Fig. Appendix A.1 and Appendix A.2 for full and limited coverage. Extensive data on the average evaluation measures over different values for each parameter can further be found in the sup­ plementary material. Here we provide a summary of some of the more pronounced sensitivities. Let us first regard full coverage. A higher heterogeneity of the production factors (ΔA0 , ΔB0 ) makes the compositional change towards more emission-efficient firms less sensitive to price fluctuations, which reduces the difference between the tax scenarios and quantity-based regulation. Consumer impact is further increased in the tax and permit market scenarios, while it is decreased under direct regulation. Another relevant factor is the emission price adaption rate (μ), which decreases technology adoption for the tax scenarios, and increases it for downstream direct regulation. This regards the ‘overshoot-decline’ dynamic of quantity-based regulation, which explained the lower technology adoption of the tax scenarios in Section 4. The magnitude of this dynamics is reduced when firms adapt their prices more slowly, i.e. if the emission price adaption rate is low. Technology adoption is also affected by the profitability target (η), although the effect is small. A high value of this parameter leads to lower adoption under the tax scenarios and the upstream permit market. Downstream direct regulation, in contrast, shows an increase of adoption and a decrease of compositional change. Consumer impact is further decreased for all revenue-based instruments (i.e. tax and permit market). Downstream market adaption speed (χ M ), while having only a small effect, increases technology adoption and compositional change in all sce­ narios except the downstream quota. Consumer impact is further increased for all revenue-based instruments. The upstream speed (χ S ) increases technology adoption for tax and permit market, while decreasing it for direct regulation. It also decreases production decline for tax and permit scenarios, while increasing it under upstream direct regulation. The mark-up adaption rate (ϑ) reduces technology adoption for downstream policies and increases it for upstream policies, while generally increasing compositional change in all policy scenarios and leading to less production decline for quantity-based regulation. The heterogeneity of the abatement cost factor (Δθ) increases compositional change for both tax scenarios as well as for downstream direct regulation. Many parameters are particularly sensitive under downstream direct regulation. Increased technology adoption and decreased compositional change with high abatement cost factor (θ) and demand sensitivity (γ). Lower sales price and production decline are found for low heterogeneity (ΔA0 , ΔB0 ) as well as high abatement potential (λ), permit price adaption rate (μ), and mark-up adaption rate (ϑ). Further sensitivities that are similar for all scenarios are found for the demand response to prices (γ), which decreases the sales prices, the pro­ duction cost increase (β), which decreases sales price and production decline, and the abatement cost factor (θ), which reduces consumer impact (except for direct regulation). Regarding the limited coverage setting, the coverage factor (φ) is also shown to affect the results (see, e.g. pp. 63-64 in the supplementary ma­ terial). This sensitivity appears in particular for upstream regulation and for the measures of consumer impact and sales price. However, no simple linear relationship is found between the parameter values and the evaluation measures. A more detailed analysis of this relationship is left for future research. Other sensitivities that do not appear under full coverage are found for the upstream production cost increase (β) and the mark-up adaption rate (ϑ), but also do not show a clear relationship. This is likely because the price increase of fuels from unregulated suppliers plays a key effect in the results discussed in Section 4.2, with different levels of upstream production cost and mark-up increases also affecting competitive dynamics. 10 J. Foramitti et al. Technological Forecasting & Social Change 172 (2021) 121060 Table Appendix A.1 Values of fixed model parameters. Parameter Symbol Value Simulation length T 200 Policy implementation period T∗ 100 Initialization length T0 50 Number of firms in each sector Nm , Ns 30 Number of abatement options Nλ 20 Initial production factors A0 , B0 1 Maximum demand D0 1 Desired inventory rate I 1 Emission target e∗ 0.1 Initial mark-up m0 0.1 Initial emission price pe0 0.1 Table Appendix A.2 Value ranges of variable model parameters. Parameter Symbol Minimum value Maximum value Abatement cost factor θ 15 20 Forecasting factor t∗ 3 10 Abatement potential λ 0.5 0.9 Fraction of covered firmsa φ 0.3 0.7 Profitability target η 0.1 0.5 Price sensitivity of demand γ 0.4 0.5 Mark-up adaptation rate ϑ 0.1 0.5 Market share adaptation rates χM , χS 0.1 0.5 Emission price adaption rate μ 0.05 0.1 Heterogeneity factors ΔA0 , ΔB0 , Δθ, Δμ, Δη 0.1 0.5 Upstream production cost increase β 0.01 0.1 a Notes: The fraction φ is set to 1 in Section 4.1 (full coverage) and only varied in Section 4.2 (limited coverage). Table Appendix A.3 Definition of policy evaluation criteria and additional indicators. n Criteria Y n,z 1 Technology adoption S(ΔAm ∗ qm ) 2 Compositional change S(Δρm ∗ Am,t ) 3 Production decline S(ΔQt ∗ ρm,t ∗ Am,t ) g 4 Sales price S(qm,t − lm,t ) g g 5 Consumer impact S(sm,t ∗pm,t )/10 − S(R)/S(qm,t − lm,t ) 6 Emissions S(em,t ) 7 Emission price S(ej,t ∗ pej,t ) 8 Upstream profit rate S((πs,t ))/S((q∗s,t )) 9 Downstream profit rate S((πm,t ))/S((q∗m,t )) 10 Upstream Market concentration S((ψ s,t )2 ) 11 Downstream Market concentration S((ψ m,t )2 ) Notes: S() and R are given in Eqs. Appendix A.2 and Appendix A.6. 11 J. Foramitti et al. Technological Forecasting & Social Change 172 (2021) 121060 Fig. Appendix A.1. Sobol sensitivity indices for full coverage. 12 J. Foramitti et al. Technological Forecasting & Social Change 172 (2021) 121060 Fig. Appendix A.2. Sobol sensitivity indices for limited coverage. Supplementary material Supplementary material associated with this article can be found, in the online version, at 10.1016/j.techfore.2021.121060 References Simulation XIX. MABS 2018. Lecture Notes in Computer Science, vol 11463. Springer, Cham. https://doi.org/10.1007/978-3-030-22270-3_1. Erickson, P., Lazarus, M., 2018. Would constraining US fossil fuel production affect Arthur, W.B., 1999. Complexity and the economy. Science 284 (5411), 107–109. https:// global CO2emissions? A case study of US leasing policy. Clim. Change 150 (1-2), doi.org/10.1126/science.284.5411.107. 29–42. https://doi.org/10.1007/s10584-018-2152-z. Castro, J., Drews, S., Exadaktylos, F., Foramitti, J., Klein, F., Konc, T., Savin, I., van den Fæhn, T., Hagem, C., Lindholt, L., Mæland, S., Rosendahl, K.E., et al., 2017. Climate Bergh, J., et al., 2020. A review of agent-based modeling of climate-energy policy. policies in a fossil fuel producing country: demand versus supply side policies. Wiley Interdiscip. Rev.: Clim. Change e647, 1–26. https://doi.org/10.1002/ Energy J. 38 (1), 77–102. https://doi.org/10.5547/01956574.38.1.tfae. wcc.647.May 2019 Farmer, J.D., Foley, D., 2009. The economy needs agent-based modelling. Nature 460 Collins, K., Mendelevitch, R., 2015. Leaving coal unburned: options for demand-side and (7256), 685–686. https://doi.org/10.1038/460685a. supply-side policies. Technical Report. Foramitti, J., Savin, I., van den Bergh, J.C.J.M., 2021. Emission tax vs. permit trading Dosi, G., Fagiolo, G., Roventini, A., 2010. Schumpeter meeting Keynes: a policy-friendly under bounded rationality and dynamic markets. Energy Policy 148, 112009. model of endogenous growth and business cycles. J. Econ. Dyn. Control 34 (9), https://doi.org/10.1016/j.enpol.2020.112009. 1748–1767. https://doi.org/10.1016/j.jedc.2010.06.018. Goulder, L.H., Schein, A.R., 2013. Carbon taxes versus cap and trade: a critical review. Edmonds, B., Aodha, L.n., 2019. Using agent-based modelling to inform policy what Clim. Change Econ. 04 (03), 1350010. https://doi.org/10.1142/ could possibly go wrong ? In: Davidsson, P., Verhagen, H. (Eds.). Multi-Agent-Based S2010007813500103. 13 J. Foramitti et al. Technological Forecasting & Social Change 172 (2021) 121060 Herman, J., Usher, W., 2017. Salib: an open-source python library for sensitivity analysis. coordination among abatement options. Eur. J. Oper. Res. 286 (3), 1113–1128. J. Open Source Softw. 2 (9), 97. https://doi.org/10.21105/joss.00097. https://doi.org/10.1016/j.ejor.2020.03.080. Kalkuhl, M., Brecha, R.J., 2013. The carbon rent economics of climate policy. Energy World Bank Group, 2019. State and Trends of Carbon Pricing 2019. 10.1596/978-1- Econ. 39, 89–99. https://doi.org/10.1016/j.eneco.2013.04.008. 4648-0725-1. Kirman, A., 2006. Heterogeneity in economics. J. Econ. Interact. Coord. 1 (1), 89–117. https://doi.org/10.1007/s11403-006-0005-8. Joël Foramitti is a PhD candidate at the institute ICTA-UAB of the Universitat Autónoma Lazarus, M., van Asselt, H., 2018. Fossil fuel supply and climate policy: exploring the de Barcelona (Spain), and the Vrije Universiteit Amsterdam (Netherlands). His work is part road less taken. Clim. Change 150 (1-2), 1–13. https://doi.org/10.1007/s10584- of the ERC project on behavioral-evolutionary analysis of climate policy (EVOCLIM). Joël 018-2266-3. holds a master’s degree in Ecological Economics from the Universitat Autónoma de Bar­ Global warming of 1.5∘ C. An IPCC Special Report on the impacts of global warming of celona (Spain) and a bachelor’s degree in Physics from the Vienna University of Tech­ 1.5∘ C above pre-industrial levels and related global greenhouse gas emission nology (Austria). His research focuses on climate policy analysis, agent-based modeling, pathways, in the context of strengthening the global response to the threat of climate and the question of human wellbeing within environmental limits. change, sustainable development, and efforts to eradicate poverty. In: Masson- Delmotte, V., Zhai, P., Pörtner, H.O., Roberts, D., Skea, J., Shukla, P.R., Pirani, A., Moufouma-Okia, W., Péan, C., Pidcock, R., Connors, S., Matthews, J.B.R., Chen, Y., Ivan Savin is a postdoctoral research fellow within the ERC project ’Behavioral-evolu­ Zhou, X., Gomis, M.I., Lonnoy, E., Maycock, T., Tignor, M., Waterfield, T. (Eds.), tionary analysis of climate policy: Bounded rationality, markets and social interactions’ 2018. IPCC - Intergovernmental Panel on Climate Change.In press (EVOCLIM) at ICTA-UAB. Before that, he served within the KIT-BETA project devoted to Mullainathan, S., Thaler, R.H., 2000. Behavioral Economics. Working Paper. general purpose technologies, creativity and sustainability (2015-2018) and was a postdoc Paltsev, S., Capros, P., 2013. Cost concepts for climate change mitigation. Clim. Change for four years at the Friedrich Schiller University and the Graduate College ”The Eco­ Econ. 04 (Suppl. 1), 1340003. https://doi.org/10.1142/S2010007813400034. nomics of Innovative Change” (2011-2015). He received his PhD from the Department of Perman, R., Ma, Y., McGilvray, J., Common, M., 2003. Natural Resource and Economics of the Justus Liebig University of Giessen (Chair of Econometrics, supervisor Environmental Economics, 3. Pearson Education Limited, Harlow. Prof. Dr. Peter Winker) in 2011. In 2017 I habilitated at KIT with the thesis titled ”Ap­ Piggot, G., Erickson, P., van Asselt, H., Lazarus, M., 2018. Swimming upstream: plications of Evolutionary Optimization and Modelling in Economics”. Given all that, his addressing fossil fuel supply under the UNFCCC. Clim. Policy. https://doi.org/ research interest naturally include agent-based models, economics of innovation and 10.1080/14693062.2018.1494535. optimization methods. Saltelli, A., Annoni, P., Azzini, I., Campolongo, F., Ratto, M., Tarantola, S., 2010. Variance based sensitivity analysis of model output. Design and estimator for the Prof. Dr. Jeroen C.J.M. van den Bergh ICREA Research Professor at Institute of Envi­ total sensitivity index. Comput. Phys. Commun. 181 (2), 259–270. https://doi.org/ ronmental Science & Technology, Universitat Autónoma de Barcelona. Honorary Professor 10.1016/j.cpc.2009.09.018. of Environmental & Resource Economics in School of Economics & Business and Institute Seto, K.C., Davis, S.J., Mitchell, R.B., Stokes, E.C., Unruh, G., Ürge Vorsatz, D., 2016. for Environmental Studies, VU University Amsterdam (VUA). Editor-in-chief of journal Carbon lock-in: types, causes, and policy implications. Ann. Rev. Environ. Resour. 41 Environmental Innovation & Societal Transitions. Previously, full professor of Environ­ (1), 425–452. https://doi.org/10.1146/annurev-environ-110615-085934. mental Economics (1997-2007) at VUA. A Master degree in Econometrics & Operations Sinn, H.-W., 2012. The Green Paradox - A Supply-Side Approach to Global Warming. MIT Research from Tilburg University, and a PhD from VUA. His work is much cited, indicated Press. by more than 20,000 citations in Google Scholar. He received the Royal Shell Prize 2002, Sobol, I.M., 2001. Global sensitivity indices for nonlinear mathematical models and their IEC’s (Sant Jordi) Environmental Prize 2011, an ERC Advanced Grant in 2017, and an Monte Carlo estimates. Math. Comput. Simul. 55 (1-3), 271–280. https://doi.org/ honorary doctorate from the Netherlands’ Open University in 2019. His latest book is 10.1016/S0378-4754(00)00270-6. ’Human Evolution Beyond Biology and Culture: Evolutionary Social, Environmental and Yu, S.-m., Fan, Y., Zhu, L., Eichhammer, W., 2020. Modeling the emission trading scheme Policy Sciences’ (Cambridge University Press). from an agent-based perspective: system dynamics emerging from firms’ 14