Fossil-extraction bans are not enough to achieve the Paris agreement but can facilitate it


 Given concerns about the ambition and effectiveness of current climate pledges, a case has been made for the integration of demand-side policies such as carbon pricing with supply-side bans on fossil fuel extraction. However, little is known about their interplay in the context of climate stabilization. Here, we present the first multi-model assessment quantifying the effectiveness of supply-side policies and their interactions with demand-side ones. We design narratives of fossil fuel bans and explore a variety of scenarios with four process-based integrated assessment models. We find that supply side treaties reduce emissions but not enough to stabilize temperature increase to well below 2°C. When combined with demand-side policies, supply side policies reduce the required carbon price to meet the Paris goals, dampen reliance on CO2 removal and increase investment in renewable energy while increasing fossil producer revenues. The results indicate the opportunity to integrate price-based policies with fossil bans when exploring climate neutrality pathways.

We used four Integrated Assessment Models (PROMETHEUS, REMIND, TIAM-UCL and WITCH) 29 that have been used to provide scenarios in key assessments such as the IPCC Assessment 30 reports 2,18 . As they differ in underlying modelling frameworks and assumptions as well as in the 31 representation of the economy and the energy system, the joint assessment provides robustness to 32 our results. 33 We designed a series of scenarios (Figure 1 (b)) to assess the effects of hydrocarbon extraction bans 34 on emissions and the energy system, both alone and coupled with demand-side policies (in the form 35 of carbon pricing). We model supply-side policies as a forced reduction of fossil fuel extraction by up 36 to 70% of 2020 production for all fossil fuels, a value that was found to be near the maximum feasible 37 ban level that all models could achieve. Given that the design allows for a modest residual 38 production, it depicts scenarios in which governments are unwilling or unable to completely shut 39 down their hydrocarbon extraction industries. 40 Results show that, while banning only coal is largely insufficient to deviate from NDCs trajectory, 41 extraction bans for all fossil fuels substantially reduce emissions if large producers participate in the 42 treaty. However, they can reach Paris consistent targets at a competitive cost only if coupled with 43 carbon pricing, as the combination of demand and supply-side policies produces synergies in policy 44 implementation and effectiveness. 45 Figure 1: (a) Narrative distribution for different fuels. For each fossil fuel, darker shades of the color identify groups that enter later the treaty. The beginning of the colored line identifies the year in which each group starts limiting production of the fossil fuel, while arrows mark the end of the transition period, after which the extraction limit is fully enforced at the final level. Maps are shown identifying the regional distribution of frontrunners, followers and laggards for each fossil fuel, using WITCH regions. (b) Scenario names and definitions.

Emission pathways and carbon budgets
We begin by exploring the emission and climate consequences of the scenarios (Figure 2). The 2°C 46 scenarios imply a sharp reduction of global emissions in the first half of the century, reaching net-47 zero between 2060 and 2075, depending on the model. These results are consistent with previous 48 assessments 2,19 . 49 Banning only coal extraction decreases cumulative emissions to 2100 by 2.6%-9.4% (model range) 50 over the reference scenario. On the one hand, this is due to a rebound effect (see Figure 3) on 51 unbanned fossil fuels. On the other hand, the reference scenario already implies a gradual phase-52 out of coal due to the NDC effort extrapolation to 2100. Moreover, the extraction ban design allows 53 for 30% of residual hydrocarbon production, a level aligned with long-term projections for the 54 Reference scenario. 55 Supply-side scenarios that limit production of all fossil fuels tend to bridge the current level of effort 56 with the emission pathways of a 2°C, reducing global CO2 emissions in 2050 by up to 59%. 57 When all suppliers join the treaty at the same time the emission range overlaps with the well below 58 2°C emission pathways range up to 2045, suggesting that a strong commitment from all suppliers 59 could drive emissions to an optimal decarbonization pathway towards PA goals in the first part of the 60 century, but not after that. In fact, both supply policy scenarios reach similar emission levels by the 61 end of the century of around 12 Gt CO2, a level consistent with the residual production of fossil fuels 62 allowed by these scenarios, while carbon tax-based scenarios reach net-zero emissions. Long-term 63 emissions are higher in supply-side scenarios even though in terms of electrification, renewables 64 penetration, fossil phase-out, and energy efficiency improvements extraction bans outperform 65 carbon pricing (see Additional results A). This is because extraction bans do not foster Carbon 66 Capture and Storage (CCS), Biomass with Carbon Capture and Storage (BECCS) and other 67 Negative Emissions Technologies (NET) needed to offset the residual CO2 emissions from industrial 68 processes and other hard-to-abate sectors, as well as land use and non-CO2 emissions from 69 agriculture and other sources. Therefore, while an extreme policy with 100% cuts to fossil fuel 70 extraction would reduce emissions from fossil fuels to zero, net zero GHG emissions could still not 71 be reached with supply-side policies alone. 72 The combination of carbon pricing and extraction bans anticipates decarbonization because the 73 production constraint created by the extraction ban in the supply regions is more binding than the 74 implicit constraint produced by carbon pricing. Thus, integrating fossil-extraction bans with global 75 carbon pricing increases the mitigation effort early on and reduces the reliance on NETs (Figure 2  76 (b)), a desirable feature both because these technologies are currently expensive and unavailable 77 at scale and because lower budget overshoot reduces climate risk 20 . 78 This is especially visible for the WITCH model (11% reduction of carbon sequestered by Biomass 79 with Carbon Capture and Storage, 36% reduction of carbon avoided by fossil CCS, 96% reduction 80 of carbon sequestered by Direct Air Capture, cumulative values to 2100) which relies more in 81 negative emissions, but holds true also for TIAM-UCL (3% reduction in carbon sequestered by 82 BECCS and 6% reduction of carbon avoided by fossil CCS), REMIND (35% reduction of carbon 83 sequestered by DAC, 35% reduction of carbon avoided by fossil CCS and 2.5% reduction of carbon 84 sequestered by BECCS), and PROMETHEUS (9% reduction of Fossil CCS and 18% reduction of 85

BECCS in 2050). 86
Lower reliance on CCS and NETs implies an increase in renewable penetration, electrification, 87 energy efficiency improvement and lower investments on fossil fuel power plants and upstream 88 sector with respect to a demand-side only scenario (see Additional results A). 89 Prices, trade, and producer support While both types of policies tend to decrease emissions, the underlying mechanism is different: 90 carbon pricing affects the consumers price of carbon-intensive goods and fuels but depresses fossil 91 fuel prices at the international market level, while an extraction ban produces fuel supply scarcity 92 and increases the price of the fuel banned ( Figure 3 (a)). 93 Uniformity of participation to the treaty governs the speed of the price increase, as can be seen by 94 the substantial difference between the two supply side scenarios prices. In the narrative scenario 95 the greatest impact for all fuels is seen when the laggards, that account for most low-cost producers, 96 initiate their ban. Only a coalition representing a large enough share of global hydrocarbon 97 production has a meaningful effect on prices: for coal, where frontrunners are a larger group, the 98 effect on prices is visible early on; for oil and gas, frontrunners action has a negligible effect on 99 prices, while when followers join the treaty the price of gas increases by 12% in the following time 100 step. A global participation in the supply treaty is necessary for oil prices to increase. The same 101 effect is visible when demand and supply-side policies are implemented together. In this case, the 102 price change can be compared to the price change in the demand-side scenario, on top of which the 103 supply side treaty is implemented: prices increase only when all regions participate the bans. 104 The increase in prices clearly reduces the primary energy use of the fossil fuels ( Figure 3 (b)). 105 Our results suggest that the ban of only one fuel may cause a visible increase in the use of other 106 fuels if it is not coupled with another type of policy, as seen by the increase in oil and gas primary 107 use when only coal is banned (6.2% increase in primary energy for gas and 2.6% for oil in 2050, 108 SUPCOAL model median). This rebound effect accounts for 16% to 31% (range across models) of 109 cumulative emissions avoided from reduced coal consumption. 110 Early and coordinated supply policies reduce by 76% and 74% the global gas and oil use by 2050 111 (model median), much faster than demand side scenario trajectories imply. For coal, the price 112 increase generated when all-regions act as frontrunners align the supply levels with the Paris 113 agreement compatible pathways up to 2035. After that, the primary energy consumption is bounded 114 by the residual production allowed in the scenario design, while demand side policies can further 115 reduce coal supply because carbon pricing puts a higher additional cost on coal, the most carbon 116 intensive fossil fuel. 117 With extraction ban policies, trade patterns and hydrocarbon producer revenues are influenced by 118 two opposing forces: higher fossil prices increase revenues per output, but shrinking demand 119 reduces volume traded. Figure 3 (c) shows that the first effect dominates and produces a large 120 increase of trade revenues if all regions are frontrunners, especially evident for coal (+143% of NPV 121 value of global exports, model median), but relevant also for gas (56%) and oil (39%). In the narrative 122 scenario, the effect is less evident because prices increase more slowly as countries join the treaty 123 at different points in time, but still relevant for gas (+28%) and coal (+16%). If carbon pricing and 124 extraction bans are implemented together, the net value of trade is in line with the carbon tax 125 scenario. In any case, in international energy markets supply side policies show the opposite 126 tendency of demand side policies, that tend to depress both prices and demand. 127

Costs, carbon prices and co-benefits
In both the scenarios that implement a global carbon tax, the global carbon budget is constrained at 128 1000 GTCO2 by 2100. If combined, the supply-side policy interacts with carbon pricing by carrying a 129 part of the shadow cost of the energy transition. As a result (Figure 4 (b)), the optimal carbon price to 130 reach the target budget is reduced by 6.2% in 2050 and 5.9% in 2100 (model median, SUPDMD vs 131 DMD). 132 The higher cost of supply side scenario with early participation can be explained by three factors. As 138 discussed, hydrocarbon extraction bans incentivize a narrower portfolio of mitigation options with 139 respect to carbon pricing. Moreover, the prescribed linear reduction for fossil fuel bans does not 140 follow a least-cost-option-first approach, introducing inefficiencies in the decarbonization 141 process. Finally, the simultaneous ban of all fossil fuels provides a greater shock to the energy 142 system sooner in time when the discount effect for the future is lower. 143 Banning only coal, while providing only incremental emission reductions over the Reference 144 scenario, is very cheap over the century (0.05% GDP loss, model median). 145 Similarly, joining supply and demand action provides consistent improvements in decarbonizing the 146 energy system and meeting the Paris goals, while at the same time introducing only a small amount 147 of additional costs for the society (1.1% vs 1% of GDP loss when cost optimal DMD scenario is 148 implemented). 149 Global deaths from air pollution decrease in the more aggressive supply side scenario by almost 150 700.000 people per year against 450.000 people per year due to demand-side policies (relative to 151 the reference, see Additional Results D: Air pollution). While not explicitly estimated, lower costs 152 from reduced air pollution damages could counterbalance the higher GDP loss seen in scenarios 153 with supply side policies. 154

Conclusions
We have shown that a global supply side treaty can improve current climate ambition and induce 155 emission reductions, substituting carbon pricing up to 2050 in case of early and uniform participation. 156 After mid-century, however, supply policies alone fail to phase-in CCS and NETs and reach stringent 157 budgets compatible with Paris goals, and they are significantly less cost-effective than a carbon tax. 158 This suggests that, if not backed by a carbon tax of similar ambition, limiting extraction of fossil fuels 159 might not be an efficient solution to reach deep mitigation targets by the end of the century. 160 Banning only coal proves cheap but largely ineffective in increasing the level of ambition relative to 161 current NDCs and stated policies, in part due to a significant rebound effect to the other fossil fuels. provided that their effectiveness will depend on the share of fossil production suppliers they 169 include. Further analysis is needed to assess minimum effective coalition size and quantify positive 170 spill-over effects. 171 The increase in fossil fuel prices provides producing regions with sustained revenues from 172 international hydrocarbon trade, counterbalancing reduction in the traded energy volumes. This 173 could lead energy producing regions to root for this kind of policy rather than a carbon tax but could 174 cause opposition from importing countries that would suffer from the increase in fossil fuel prices. 175 However, the fact that both importing and exporting countries would, at least in the short term, benefit 176 from exceeding their allowed production quotas poses a challenge to the stability of the treaty. 177 If supply-side policies are integrated with carbon tax, in a 2°C scenario the reliance on expensive, 178 high-risk, and currently immature CCS and negative emission technologies and investments in fossil 179 fuels are reduced at a marginal additional cost for the economy and with a lower carbon tax. These

Corresponding Authors
Correspondence to Pietro Andreoni. 194

SUPPLEMENTARY INFORMATIONS Methods
We have designed six scenarios (Table 1). First, we defined a counterfactual scenario (REF) that 195 reflects current established and planned policies, including the NDCs (as submitted by mid-2010) 23 . 196 The reference scenario is based on the socio-economic assumptions of SSP2 (middle-of-the-road 197 scenario) 24 . Then, we build supply-side narratives where the production of coal, oil and gas is cut, starting from 204 different enforcement years (scenario SUP), up to at least 70% of 2020 production level ( Table 2). 205 Table 1: target level for different fuels.
The residual production takes into account the challenges to fully phase out fossil fuels in hard-to-206 decarbonize sectors (e.g. heavy industry, aviation, maritime) and the difficulties of countries to 207 completely shut down their resource extraction industries. We use a systematic approach to design 208 this narrative according to each region position in energy trade, reserves and resources, fossil fuel 209 dependency and climate policy commitment. While we are aware that supply-side narratives may be 210 hard to enforce given the status-quo, we keep a realistic approach based on qualitative and 211 quantitative information that can help us hypothesize how such policies would unfold. The world 212 regions are classified into ``front-runners'', ``followers'' and ``Laggards'', defining the speed at which 213 the region will enforce the production cuts ( Figure 1). 214 Recognizing that some fossil fuels are more difficult to ban (oil) and more important for the energy 215 transition (gas) than others (coal), the timing for the phase-out varies with the fuel: laggards for coal 216 finish banning in 2055, while the extraction ban for oil completely enters into force in 2060 and in 217 2065 for gas. 218 To analyse the effects of different regional timings in the production cuts, we analyse a supply-side 219 policy scenario where all the regions are front runners. Furthermore, we design a SUPCOAL 220 scenario in which only coal is phased out following the same narrative as in SUP.

Narrative design
To design the Narrative for the SUP scenario, participating regions were categorized by the 227 dimensions described in Table 2. According to the rationale explained in the same table, each of 228 these dimensions favors or hinders the participation to the supply treaty. Each dimension was 229 parametrized by a numerical indicator that served as a starting point to assign a total score to each 230 region/country, measuring the estimated propensity to join the treaty for each fuel. According to this 231 aggregate indicator, countries were assigned to followers, frontrunners, or laggards (see Table 3). 232  Table 3 were chosen because they are relevant as producers of 233 at least one fossil fuel, large energy consumers, or because they hold large hydrocarbon reserves. 234

Countries and regions analyzed in
Regional disaggregation of the models, however, differs from that in Table 3 as well as among each 235 other. The countries analyzed were thus translated into the model regions as closely as possible by 236 each team. 237 All results are then reaggregated to the 17 regions of the WITCH model using GDP weighting, to 238 provide coherent plotting and figures. 239 Table 4 shows the share of total demand and supply for each fossil fuel, distributed among the 240 narrative groups, relative to the reference scenario. 241 For oil and gas, frontrunners account for 7.8% and 5.9% of total production respectively, and 242 followers for 14.7% and 11.0% of total production in 2020. Laggards thus represent most of the oil 243 and gas producers, as well as the largest portion of total demand. 244 to the other fossil fuels. A major reason for this is the US, which is modelled as a frontrunner for 247 Coal, given that both consumption and production are historically declining, but a laggard for oil and 248 gas, because of the shale revolution and the renewed role of the United States as a major oil and 249 gas producer as well as a key consumer. This reflects the reality that coal is a less powerful industry 250 than oil and gas and the political feasibility of banning it may be higher than the other two fossils.  The price of fossil fuels and exhaustible resources (oil, gas, coal and uranium) is determined by the 301 marginal cost of extraction, which in turn depends on current and cumulative extraction. A regional 302 mark-up is added to mimic different regional costs including transportation costs. supply to form market equilibrium at the regional (for ten distinct regions) and global energy 309  32 . 336 The regional developments of energy demand and supply, the inter-fuel substitutions, energy and 337 climate policies and hydrocarbon resource assumptions influence the evolution of international fossil 338 fuel prices. The price of crude oil depends on the petroleum production cost by region, the reserves 339 to production ratio, the capacity supply and role of OPEC in the global market. Gas prices are 340 influenced by oil price developments, gas resources and production prospects and costs to produce 341 and transport conventional and unconventional gas in each region. The model includes different gas 342 pricing mechanisms and can represent oil indexation, gas-to-gas competition or their combination 343 (e.g. in the EU). Coal price is driven by global coal demand, coal reserves and production prospects 344 and is partly linked to oil price developments as observed in energy markets. Overall, the dynamics 345 of international fossil fuel prices are influenced by energy demand and supply, thus they are directly 346 influenced by supply constraints. 347 The modelling combines economic foundations with the representation of agents' behavior (e.g. Agricultural Production and its Impacts on the Environment). 372

TIAM-UCL is the TIMES Integrated Assessment Model (TIAM) developed at the UCL Energy 373
Institute (code underlying the model available at https://github.com/etsap-TIMES/TIMES_model). 374 This is a global multiregional technology-rich bottom-up cost optimization model. It is a partial 375 equilibrium model that represents energy resource extraction through conversion processes 376 (refineries, electricity, and heat generation) and infrastructure to end-uses in the residential, 377 commercial, industry, transport and agriculture sectors. With perfect foresight over the modelling 378 period, the model designs a cost-optimal transition of the energy system so that future service 379 demands are met, while obeying technical, economic and policy constraints. 380 On the resource side, a total of eleven conventional and unconventional oil resource categories, 381 eight conventional and unconventional gas resource categories, and two coal resource categories 382 are specified. Each of these categories is specified with an individual supply cost curve within each 383 region. Table M

WITCH
In the WITCH model, the method for implementing production quotas on the supply side vary with 400 fossil fuels: the oil upstream sector (as well as investments) is completely endogenous in the model 401 33 , and limiting production simply requires the implementation of an upper bound on the variable 402 regulating extraction. 403 For coal and gas on the other end the extraction sector is modelled via global and regional extraction 404 costs curves, calibrated from (Rose project). Those curves are read after each iteration of the parallel 405 bunch of solutions for the coalitions, which implies that extraction price and regional production are, 406 within each iteration, fixed. This implies that, with the old algorithm, reducing production by simply 407 binding the extraction variable would not affect prices. Therefore, the production algorithm was 408 modified with two interventions: on the one hand, the reading of the curves was modified to allow for 409 the reallocation of regional production according to the prescribed production limits. On the other 410 hand, coal and gas market were integrated into the existing ADMM algorithm that grant market 411 clearance for global markets in the model, so that, if an imbalance exists between global demand 412 and available supply due to the extraction bans, the price of the fuel increase eventually leading to 413 market balance through iterations. 414 To gradually reduce the supply, all fuels were banned with linear trajectories: in the start year of the 415 ban, production is bound to 2020 level, while at the end year each region has to produce at most 416 30% of that value. In the intermediate periods, the reduction is linear. 417

PROMETHEUS
In the PROMETHEUS model, the production of crude oil, coal and natural gas is modelled via global 418 and regional extraction cost-supply curves, calibrated to data from the IEA, USGS and (Rose 419 project). These curves describe a non-linear relationship between fuel production cost and the 420 amount of available reserves and resources by region and fuel. Therefore, we added a constraint in 421 the production of fossil fuels (differentiated by region and fuel) to simulate the extraction bans in the 422 SUP scenarios. This simulates that in the first year of the ban, fuel production is bound to 2020 level, 423 while in the end year of the ban each region is constrained to produce at most 30% of this value (we 424 assume a linear annual reduction in the intermediate period). PROMETHEUS represents the 425 interactions in the international energy markets through market-derived prices to ensure equilibrium 426 between fuel demand and supply. Therefore, fossil fuel extraction cuts in the SUP or SUPALL 427 scenarios would directly lead to increased global fossil prices, which are also then reflected in 428 increased import prices and final consumer prices for oil, gas and coal. 429

TIAM-UCL
The fossil fuel upstream sector in TIAM-UCL incorporates the availability and costs of primary energy 430 resources, all extraction processes, and any processing required to produce, trade and distribute 431 energy products for use in end-use sectors. Individual supply cost curves for each type of reserves 432 (or potential resources) are estimated for each region. The distribution of resources assigned to 433 different cost categories varies by region and is influenced by technology maturity and technical 434 difficulty of extracting the resource. To introduce the supply policy in TIAM-UCL, we added 435 constraints limiting fossil fuel production differentiated by region and resource. The supply policy 436 main regional constraints are applied from a specific year (function of the group the region is included 437 and the fossil fuel type) and represent an upper limit of production equal to 30% of 2020 production 438 level in the region for the specific resource. Additional constraints are applied 10 years prior (5 years 439 only in the case of coal) and are equal to 55% of 2020 production levels. In between these two 440 constraints the supply levels are controlled following linear interpolation. 441 REMIND REMIND characterizes the exhaustible fossil resources (coal, oil and gas) in terms of region-specific 442 extraction cost curves that relate production cost increase to cumulative extraction 34,35 . In the model, 443 these fossil extraction cost input data are approximated by piecewise linear functions that are 444 employed for fossil resource extraction curves. The supply policy is simulated by introducing an 445 endogenous constraint that limits yearly extraction levels relative to 2020, for each region and fossil 446 resource, according to the extraction ban in the SUP scenarios. In the first year of the ban, the region 447 is constrained to produce at most 30% of the 2020 production. Previous years assume a non-448 necessarily binding, linear annual reduction extending two decades before the ban. 449 This can be explained by looking at Figure 5    As far as fossil share of primary energy is concerned, the 2030 values are significantly lower in 500 SUPALL for most models with respect to carbon tax scenarios, except for TIAM-UCL that uses a 501 slower transition trajectory before reaching the final extraction limit levels. Renewable share follows 502 the opposite trend. This shows that, while the level of ambition in the 2030s is similar for SUPALL 503 and DMD scenarios, the former targets first and more aggressively the fossil fuel sector. 504 In 2100, the same indicator shows a fossil fuel penetration in the energy system slightly higher in 505 supply scenarios (SUP and SUPALL) with respect to DMD, at around 18% of TPES. 506 This proves that the residual level of production allowed by the extraction ban design produces a 507 long-term energy system that is roughly comparable in terms of relative presence of fossil fuels, as 508 well as renewable penetration. 509 Interestingly, SUPDMD has a lower long-term level than all other scenarios: this result is driven 510 mainly by the WITCH model and its due to lower gas requirements in SUPDMD due to reduced DAC 511 deployment. 512   Figure 2 and Figure 6 with an overview of the 513 cumulative investments in various sectors of the energy system. Results are shown only for the 514 WITCH and REMIND models, because they are the only ones able to produce these variables. Both 515 models foresee a reduction in investments related to fossil fuel plants in the electricity sector with 516 both supply and demand side policies, as well as an increase in nuclear and renewables 517 investments. There is no accordance, however, on which type of policy produces the larger effect: 518 for Remind, SUP and SUPCOAL have very similar levels of cumulative investments in fossil fuels 519 power plants with respect to REF, while SUPALL, DMD and SUPDMD show a 20% reduction. The 520 WITCH model, on the contrary, is more responsive to supply side policies for fossil fuel investments 521 and shows a reduction of up to 70% in the SUPALL scenario. 522 The same trend is visible in oil upstream sector investments: scenarios containing extraction limit for 523 all fuels reduce significantly cumulative investments on the extraction sector with respect to both the 524 reference (-63% to -85%) and DMD scenarios. While produced only by the WITCH model and thus 525 lacking robustness, this result is significant because it implies that deploying supply side policies, 526 alone or coupled with carbon pricing, would halve the total size of the oil upstream sector with respect 527 to a scenario with carbon pricing only, reducing the relative importance and the lobbying power of 528 the extraction industry. 529 Banning only coal instead produces a small increase in total investments, because of higher 530 extraction requirements due to the rebound effect in consumption of oil and gas. 531 Additional results B: trade and revenues In general, hydrocarbon producing countries gain from supply side policy scenarios, while energy 534 importers worsen their exposure. 535 For coal, the biggest winners from supply side treaties (SUPALL) among large economies are the 536

USA. 537
The other scenarios tend to disrupt much less trade patterns that remain similar to the reference 538 scenario for most regions (given the fairly large uncertainty across models). 539 For gas, the general trend for demand side policies is to reduce exposition for importing countries 540 (China, Europe, India) and depressing revenues for exporters (MENA). 541 SUPDMD and DMD perform very similarly in this regard, slightly increasing revenues for some 542 producing and exporting regions (Russia, MENA) and worsening exposure for others, mostly large 543 gas importers (Europe). 544 Supply side policies tend to have the opposite effect, with most producers gaining a considerable 545 amount of trade value until mid-century regardless of the velocity of application of the extraction bans 546 (SUP vs SUPALL). 547 For oil, finally, the picture remains similar. Notable exception is the US, that in the SUPALL scenario 548 loses a considerable amount of trade position. 549 Figure 10: variation of total net present cost of the carbon tax from 2020 to 2100, discounted 3%. Figure 9 and Figure 10 show the trends for carbon taxes in different models and the total cost of the 551 carbon tax (REF relative) respectively. The total cost of the carbon tax is measured as the Net Present Value between 2020 and 2050 of the price of carbon times the CO2 emissions, discounted 553 at 3%. 554 Lower total costs of the carbon tax imply a lower distortion of the economy through taxation, but not 555 necessarily lower costs faced by consumers, because of the simultaneous increase in price due to 556 the production bans. 557 Both figures show that the reduction of the carbon tax in SUPDMD compared to DMD scenario 558 happens consistently in all four models, even if the size of the reduction is not extreme. 559 The combined lower price of carbon and decreased use of fossil fuels, however, contributes in 560 decreasing the total cost of the carbon tax by more than 10% in SUPDMD with respect to DMD. 561 Figure 11 shows a regional disaggregation of policy costs by scenario, measured in terms of NPV of 562 scenarios over 2020-2050 discounted at 3%, relative to REF. Overall, the SUPALL scenario is 563 consistently costlier for all regions, even if it's the only scenario for which certain models (TIAM-UCL) 564 foresee an overall net benefit for big producing regions (MENA, Russia, USA, and CANADA). In 565 average, SUP is less costly than a carbon budget scenario (but does not achieve the Paris mitigation 566 goal), while SUPDMD can increase the overall cost of climate policy for big importers (Europe, India, 567 China) or slightly decrease it for exporters (USA, Russia, Canada) with respect to DMD. 568 SUPCOAL is, for most regions, only slightly costlier than the REFERENCE but achieves very limited 569 emission reductions. 570 Globally, all l mitigation policies provide co-benefits as avoided deaths from air pollution. For supply 574 side policies, banning only coal provides a visible improvement in people's health, with results mainly 575 driven from India, while other big countries with air pollution problems like China show only 576 incremental benefits because coal is already being phased out at a similar pace in the reference 577

scenario. 578
The effect is much stronger if all fossils are banned together, and in 2030 the SUPALL scenario 579 shows a reduction in deaths much larger than all other scenarios, including the ones implementing 580 the carbon tax. 581 The stronger reduction of deaths compared to DMD in 2030 can be explained by the steeper 582 reduction in fossil fuels' consumption in the first part of the century, when the air pollution controls 583 are still weaker in many countries, especially in the less developed countries. The SSP2 baseline is 584 a current air pollution policy continuation scenario. It assumes a three speed world in terms of air 585 pollution control deployment, a full implementation of maximum feasible reduction end-of-pipe 586 technologies is assumed to be reached only after mid-century even in high income 587 regions. Furthermore, faster pollutant emission reductions are happening, via structural changes, in 588 highly polluted regions (e.g. Mena, Mexico and China) that do not have yet advanced air pollution 589 controls, thus structural measures may yield large co-benefits. 590 In some regions, however, banning only coal may cause an increase in air-pollution related deaths 591 in 2050, because the substitution of coal happens partially with biomasses that are also associated 592 with particulate matter emissions. 593 Finally, combining supply and demand side policies (SUPDMD scenario) results in long-term co-594 benefits for avoided deaths relative to DMD, because of reduced use of fossil fuels and CCS 595 (especially gas). The global results are mainly driven by China and India but are also robust across 596 developed countries as well. 597