Tackling Transport-Induced Pollution in Cities: A Case Study in Paris

Get Complete Project Material File(s) Now! »

Judging the merits of diﬀerent regulations

There are broadly two types of public policies to regulate pollution: command-and-control regulations, which consist in setting standards on the type of technology to be used (in the case of technology mandates) or on the maximum level of pollution (in the case of performance standards), and monitoring enforcement; and market-based or incentive-based solutions, which consist in putting a price on pollution to incentivize polluters to reduce emissions. Market-based solutions can be further distinguished between those regulating the quantity of emissions (such as cap-and-trade markets setting a cap on the number of emission permits in circulation) and those regulating their price (such as Pigouvian taxes).
The merits of diﬀerent regulations can be judged based on diﬀerent criteria. Goulder and Parry (2008) cite economic eﬃciency, cost-eﬀectiveness, the distribution of benefits and costs, the ability to address uncertainties and political feasibility as potential evalu-ation criteria.
Regarding eﬃciency, theoretical evidence suggests that market-based solutions are more cost-eﬀective than command-and-control regulations in the first best: this is be-cause polluters generally diﬀer in their ability to reduce pollution – measured by their marginal abatement cost – (Newell and Stavins, 2003), and only incentive-based instru-ments allow to equate marginal costs and marginal benefits across all emitters if pol-luters know their abatement costs but the policy-maker does not. Another theoretical advantage of market-based instruments is that they exploit all pollution reduction chan-nels, while command-and-control regulations neglect some, in particular output reduction (Spulber, 1985). Finally, incentive-based instruments generally generate revenues (except when emission permits are freely distributed), which, if appropriately recycled, can re-duce other distortionary taxes and create eﬃciency gains. This recycling could generate a “double dividend », both improving environmental quality and reducing the net welfare cost of environmental policy (Baumol and Oates, 1988; Pearce, 1991; Chiroleu-Assouline and Fodha, 2014).
However, in the presence of multiple market failures, the superiority of market-based instruments can be questioned, and a combination of diﬀerent instruments may be appro-priate, including market-based, standards but also public subsidies. For example when the administrative costs of monitoring emissions is high, mandates may be superior (Goulder and Parry, 2008). Goulder et al. (2016) also show that in the presence of pre-existing factor market distortions, clean energy standards are more cost-eﬀective than price-based instruments because they represent a smaller implicit tax on factors of production. Public support to innovation in low-carbon technologies, beside carbon pricing, is also justified by the public good nature of innovation: in a setting where the gains from innovation cannot be fully appropriated, investment is too low without public support (Fischer and Newell, 2008). The road transport sector illustrates well the multiplicity of market fail-ures and the need for second-best policies, since road transport contributes to multiple externalities: CO2 emissions, local air pollution, but also noise, accidents, and congestion (Parry et al., 2007).
As for the choice between price and quantity instrument, Weitzman (1974) suggests that the relative merits of price vs. quantity instruments depend on how steep the marginal damage function is, when there is uncertainty on the aggregate costs of pol-lution reduction – which is often the case in the real world. Price instruments are superior when the marginal damage curve is flat and quantity instruments are superior when the marginal damage curve is steep. The intuition behind this result is that it is all the more important to get the quantity right when damage costs increase a lot for a small change in pollution. Because the marginal damage curve of climate change is rather flat, economists have favoured a carbon tax over a carbon market on these theoretical grounds. At the global level, Weitzman (2015) also argues in favor of a uniform carbon tax rather than internationally tradeable permits, because of uncertainty regarding country-specific abatement cost profiles.
Regarding equity, distributional concerns may arise because some individuals bear a disproportionate cost of the regulation. For example, carbon taxes have often been found to be regressive in high-income countries if the receipts of the tax are not redistributed (Poterba, 1991). This is because poorer households dedicate a higher share of their con-sumption expenditures to carbon-intensive goods. Studies have found that carbon taxes can become progressive if tax receipts are redistributed in the form of lump-sum trans-fers to all households (Metcalf, 2009a; Cronin et al., 2018; Douenne, 2020; Berry, 2019). However, lump sump transfers are unlikely to correct horizontal equity issues, because of the high degree of heterogeneity in tax incidence within income deciles (Sallee, 2019; Douenne, 2020; Berry, 2019). Standards have been found to be more regressive than a carbon tax with lump-sum transfers in some cases (Davis and Knittel, 2018; Levinson, 2018), while Zhao and Mattauch (2020) show that standards are more equitable when consumers exhibit a preference for high-carbon technology attributes – and they verify that this is the case in the US.
Political support and feasibility will likely be aﬀected by the objective distributional properties of the proposed regulations. Beyond these objective properties, however, recent research highlights the importance of the perception of policies (Douenne and Fabre, 2021; Maestre-Andres et al., 2019), and of contextual factors such as political trust (Rafaty, 2018).

Regulating air pollution and CO2 emissions in practice

In practice, local pollutants and CO2 emissions have been regulated via a combination of instruments. Command-and-control instruments have historically been more common and have consisted in standards with specific requirements to use the best available technology (BAT) or other specific technology mandates (Metcalf, 2009b). Nowadays, diﬀerent types of regulations co-exist depending on the jurisdictions and on the pollutant: local pollutants from industrial installations (including the power sector) are regulated by standards in the European Union (Directives on industrial emissions) and in the US (Clean Air Act); by cap-and-trade programmes in some US states (California Regional Clean Air Incentives Market (RECLAIM)); by taxes in some European countries (the TGAP in France or the NOx fees in Sweden or Norway (Bonilla et al., 2018)). Carbon emissions are also regulated via a combination of command-and-control regulations, for example in the form of fuel eﬃciency standards (such as the US CAFE) or emission standards (in the European Union) in the transport sector, and market-based instruments, for example the European carbon market (EU ETS) or the Chinese ETS. Market-based instruments have increased in the past decade, with 64 carbon pricing instruments across the world in 2021 against only 21 ten years before in 2011 (World Bank, 2021).

Evidence on the eﬀectiveness of existing policies

The eﬀectiveness of several of the above-mentionned regulations has been estimated em-pirically (e.g, Fowlie et al. (2012) for RECLAIM; Colmer et al. (2020) for the EU ETS; Currie and Walker (2019) for the Clean Air Act; Andersson (2019) for the Swedish carbon tax). However, such evaluations are relatively scare, in particular concerning carbon pric-ing instruments and their eﬀectiveness. One reason is simply that these instruments are rather recent. Another reason is methodological: beyond the amount of data required, causally estimating the impact of a carbon pricing on emissions requires overcoming the “fundamental problem of causal inference » (Holland, 1986a), whereby in countries with a policy in place, we only observe the evolution of pollution in the presence of the policy, but not in the counterfactual situation where the policy is absent. This problem is all the more pronounced in the case of carbon pricing instruments, typically targeting an entire sector or region. Of the 21 empirical evaluations of existing carbon prices listed in Rafaty et al. (2020), four evaluations are on the manufacturing sector, five on the transport sector alone, six on both the power and manufacturing sectors12 and three pool several country-level instruments on diﬀerent sectors. Only three papers (including the first chapter of this thesis) focus on the power sector alone and they all consider the same instrument. Thus, more evidence is needed on the eﬀectiveness of carbon pricing in general, but also in the power sector in particular, given its high contribution to worldwide emissions. In addition, there is a wide heterogeneity in estimates of carbon price eﬀectiveness (Rafaty et al., 2020; Green, 2021). This calls for more systematic evidence on the factors ex-plaining such diﬀerences, whether these are due to contextual factors or methodological diﬀerences in the estimation strategy and scope considered.

Policy recommendations and paths for future research

Most emission pathway scenarios consistent with the Paris Agreement impose drastic declines in coal power by 2030 and full phase-out by 2050 (Jewell et al., 2019). My first chapter shows the potential of carbon pricing in helping to achieve this transition. Three factors arguably enabled to achieve high emission reductions in the UK with limited leakage outside the UK – a high fuel switching potential, a limited interconnection with other countries, and stringent air pollution regulations making coal use uneconomical. Carbon pricing initiatives in the power sector could be all the more successful if these conditions are met. Future research could keep the UK as a case study and examine whether such coal phase-out also entails significant health and environmental benefits, as found in ex ante modelling studies on the entire world (Rauner et al., 2020). Another research avenue would consist in investigating the political economy factors that facilitated this transition, which seem to contrast with other contexts such as the German, Polish or Australian cases.
The conclusions from our second chapter call for caution in assessing the priority sectors to tackle in terms of local pollution. Although we show that cruise vessel traﬃc has a significant impact on city-level ambient air pollution concentrations, it is unclear whether tackling this source of pollution should be a priority over other sources of pollution such as road traﬃc. The answer also depends on the abatement costs in these two sectors, which should be examined. Future research could also evaluate the impact of maritime traﬃc on health using neighbourhood-level health data. Another avenue would be to apply a similar causal inference pipeline to a context where command-and-control regulations mandating a decrease in the sulphur content of vessel fuel has been implementing and estimate the eﬀectiveness of such policies. Finally, in our analysis there was a trade-oﬀ between reducing data imbalance and obtaining precise estimates. In light of this trade-oﬀ, it would be interesting to characterize the conditions under which exact pair matching fares better than other matching methods using simulation methods.
In the third chapter, we show that several levers are necessary to reduce pollution from daily mobility in cities. Given the importance of high-distance trips in total emissions, modal shift could only reduce emissions by a fifth for a given public transit network. Our work however suggests that electric cycling has among the highest potential as an alternative to car, and should be targeted by public policies. Encouraging a shift to electric vehicles (EVs) by allocating EV adoption subsidies in priority to those unable to shift modes is also deemed necessary. Finally, we show that some professional groups appear to be more reliant on cars and could be disproportionately aﬀected by driving restrictions. Local policy-makers should be aware that if alternatives to car use are not oﬀered, this group could form a powerful coalition opposing any restriction in car use. A natural follow-up of this work would be to examine the distributional impacts of existing transport policies tackling car use, such as the low-emission-zones being currently rolled out in several French cities. Future research could also run similar analyses on other cities to characterize the city-level factors influencing inequalities in emissions and modal shift potential.

READ Multi-objective Metaheuristics

The UK carbon tax: context and expected eﬀects

The Carbon Price Support (CPS) was first introduced in April 2013. This domestic carbon tax was proposed in a double context of low prices on the EU carbon market, and the obligation for the UK to meet national targets for greenhouse gas emissions as defined in the 2008 Climate Change Act. The Climate Change Act set an emission target for 2050 and implemented a system of 5-year carbon budgets. Under the second carbon budget, running from 2013 to 2017, the UK had to reduce its total emissions by 236 MtCO2e compared to the first carbon budget (covering 2008 to 2012). Low prices on the EU carbon market were perceived as potentially too low to eﬀectively decrease emissions in the sectors covered by the ETS. In this context, the UK Government announced in March 2011 that a Carbon Price Floor (CPF) would be implemented in the power sector for the 2013/2014 budget year3. Under this price floor, power installations located in Great Britain (GB)4 would have to pay a tax called the Carbon Price Support (CPS), for which annual rates would reflect the diﬀerence between the desired level of carbon price floor and the expected carbon price on the EU ETS. The announced goal of the CPF was to tackle price uncertainty on the EU ETS and encourage investment in low-carbon technologies in the generation sector; in oﬃcial communication documents, the CPF was labelled “support and certainty for low-carbon investment »(Hirst, 2018). The price floor was expected to increase over time, with a total carbon price target of £30 (around e35) by 2020.

Selecting countries entering the donor pool

The “donor pool » designates the set of countries not aﬀected by the CPS that will poten-tially enter the composition of the synthetic UK. The starting pool of countries consists of the twenty European countries included in the European Union (EU), other than the UK, described in the data section. Restricting the donor pool to EU countries rather than including other OECD countries has several advantages and one drawback. The main advantage is that over the period considered, the UK and other EU countries are subject to the same EU-level policies (in particular the EU ETS and the LCP directive, but also other energy policies). European countries would also have been more likely to be aﬀected in a similar way by global shocks on the energy market, such as the 2011 US shale gas revolution. One drawback is that such geographic proximity and sectoral integration makes spillovers between treated and synthetic unit more likely.
Starting with this initial pool of twenty countries, it is important to discard the coun-tries that are likely to be poor counterfactuals (Abadie et al., 2010). This essentially describes three country types: first, countries that suﬀered idiosyncratic shocks to the outcome of interest, either by directly introducing a policy targeting the power sector or via a more generic exogenous shock likely to aﬀect the electricity sector; second, coun-tries more likely to have been directly aﬀected by the CPS; and third, countries with very diﬀerent characteristics compared to the UK, which may cause severe interpolation biases.
By 2017, no other European country had adopted a carbon tax or a carbon price floor that would interact with ETS pricing in the power sector (Metcalf and Stock, 2020).23 The most radical change in other European countries’ power sectors is the case of Germany, which unexpectedly decided to phase out nuclear energy following the 2011 Fukushima nuclear accident. I therefore exclude Germany from the donor pool. Since the European debt crisis significantly aﬀected the Greek economic environment over the period, I also exclude Greece. However, including them in the donor pool does not change the results, as shown in appendix A.1.10.
Regarding the second country type, tension can occur between excluding countries from the donor pool countries whose outcomes are aﬀected by the treated unit and iden-tifying those suﬃciently comparable to the treated unit (Abadie, 2021). While I do not exclude any country based on the risk of spillover, I do discuss this risk and oﬀer an estimation the amount of potential spillovers in section 1.4.4.
Finally, to avoid including countries that diﬀer too greatly from the UK, I eliminate Estonia, a country where high emissions per capita are due to the unusual use of oil shale for power generation, a high-emitting input fuel. I also exclude the two other Baltic countries, Latvia and Lithuania, which unlike the UK do not use coal for power generation (see Figure A.1.3). Since coal-to-gas fuel switching is expected to be an important driver of decarbonisation, it is relevant to restrict the analysis to countries with the capacity to do so.
In the end, the donor pool includes 15 EU countries. Appendix A.1.10 shows that changing the composition of the donor pool does change the composition of the synthetic UK and the estimates, but not their order of magnitude. To ensure that building a convex combination of countries (having positive weights) that closely reproduce the UK’s values for predictors and emissions is possible, there needs to be common support between the distribution of the predictors in the donor pool and in the UK. I check that this is the case for all variables (See the histograms in appendix A.1.6).

Risk of spillovers

For the synthetic control method to identify the causal impact of the intervention, candi-date units for the synthetic control group should not be aﬀected by the intervention. As an overlapping policy to an existing carbon market, The CPS could spill over to other European countries’ power sectors via two channels highlighted by Perino et al. (2019): internal leakage, that is, an increase in UK net electricity imports from other European countries; or a waterbed eﬀect, that is, an increase in emissions from European power plants not subject to the CPS, due to the negative eﬀect of the CPS on ETS permit prices under a fixed emission cap. Quantifying the magnitude of these two eﬀects for the EU carbon market as a whole goes beyond the scope of this paper, which focuses on the impact of the CPS on UK emissions. What I endeavor to assess is the risk of spillovers to countries entering the synthetic UK, given that they serve as a counterfactual for the evolution of UK emissions in the absence of a CPS.
I first estimate the amount of emissions from countries in the synthetic UK potentially caused by import spillovers. This amount is naturally bounded by the limited interconnec-tion capacity of the UK with the rest of Europe. I then estimate the amount of emissions in the synthetic UK potentially caused by a waterbed eﬀect. The two eﬀects combined represent 11% of the estimated abatement of the lower bound.
Risk of spillover via increased electricity imports : UK net electricity imports per capita are generally low compared to other European countries (see Figure A.1.2b), representing 2% of gross electricity consumption in the 2005-2012 period. However, net imports increased to 5% of gross electricity consumption in the 2013-2017 period. If this increase was caused by the CPS, it could threaten the identification strategy because two of the UK trading partners, Ireland and the Netherlands, have a combined weight of 63% in the synthetic UK. The increase in UK net imports would then increase the synthetic UK’s emissions as a result of the CPS and contaminate the counterfactual. The question is how large in magnitude this contamination is, relative to the estimated abatement. I calculate the maximum amount of synthetic UK emissions that may have been directly caused by the CPS, considering that the increase in UK electricity imports from France, the Netherlands and Ireland after 2012 was entirely caused by the CPS29. I estimate the emissions associated with these exports for Ireland and the Netherlands (those countries entering the synthetic UK). In Appendix A.1.12, I run another test where I exclude all interconnected countries from the donor pool to assess whether the presence of Ireland and the Netherlands in the synthetic UK would drive up the results. The estimated abatement is 14% lower without interconnected countries, but the balance in predictors’ characteristics is also less good. First, I calculate the excess electricity generation in the Netherlands and in Ireland which can be imputed to CPS-induced exports to the UK: to do so, I simply calculate, for every post-treatment year, the diﬀerence between electricity exports to the UK that year and average electricity exports to the UK in the pre-treatment period. I use electricity trade statistics from Ofgem to determine quarterly trade flow for each interconnector with the UK.30 I estimate that on an average year between 2013 and 2017, the Netherlands produced an excess of 2,965 GWh, and Ireland produced an excess of 382 GWh, compared to the pre-treatment period.

Table of contents :

1 Carbon Pricing and Power sector Decarbonisation: Evidence from the UK
1.1 Introduction
1.2 The UK carbon tax: context and expected effects
1.2.1 The UK Carbon Price Support
1.2.2 Descriptive evidence
1.2.3 Potential confounders
1.3 Empirical strategy
1.3.1 The synthetic control method
1.3.2 The Data
Power plant-level emission data:
Country-level power sector characteristics:
1.3.3 Selecting the predictors
1.3.4 Selecting countries entering the donor pool
1.4 Results
1.4.1 Upper Bound
1.4.2 Lower bound
Potential confounders and emission decomposition:
Lower bound: counterfactual emissions of plants converted
to biomass if they had not converted:
1.4.3 Inference
In-Time placebo
Leave-one-out test
Permutation test
1.4.4 Risk of spillovers
Risk of spillover via increased electricity imports
Risk of spillover via a waterbed effect
1.5 Discussion
1.5.1 Channels contributing to emission reduction
1.5.2 Comparison of the results with existing estimates
1.6 Conclusion
2 Estimating the Causal Effects of Cruise Traffic on Air Pollution using Randomization-Based Inference
2.1 Introduction
2.2 Materials and Methods
2.2.1 Data
Vessel Traffic Data
Air Pollution and Weather Data
Road traffic Data
2.2.2 Method
Stage 1: Formulating Plausible Interventions on Vessel Traffic
Stage 2: Designing the Hypothetical Randomized Experiments
Stage 3: Analyzing the Experiments using Randomization-based
Inference
Point estimate
Randomization-based quantification of uncertainty
Stage 4: Interpreting the Results
2.3 Results
2.3.1 Matching Results
2.3.2 The Effects of Cruise Vessel Traffic on Air Pollutants
2.4 Discussion
2.4.1 Putting our Results into Perspective
2.4.2 Reflection on the Methods
2.4.3 Potential Paths for Future Research
2.4.4 Concluding Remarks
3 Tackling Transport-Induced Pollution in Cities: A Case Study in Paris
3.1 Introduction
3.2 Air pollution and transport emissions in Paris
3.3 Data and methodology
3.3.1 The Data
Individual transport:
Emission factors
Counterfactual travel time data
Charging stations for Electric Vehicles
3.3.2 Methodology
Building individual measures of contribution to pollution
Exact factor decomposition analysis
Individual characteristics associated with high emissions
3.4 Results
3.4.1 How unequal are contributions to emissions?
3.4.2 Are high emissions mostly due to high distances, high-emission modal shares or highly polluting cars?
3.4.3 Who emits pollution?
3.4.4 What are the options to reduce emissions?
Shift to low-emission modes:
Avoid travelling by teleworking:
Improve: shift to an Electric Vehicle:
3.5 Discussion
3.5.1 A 80-20 rule?
3.5.2 Traditional and less traditional factors associated with emissions
3.5.3 From modal shift potential to actual modal shift
3.5.4 Limits
3.6 Conclusion
Bibliography