 The objective and the outline of the talk today, first, I'm going to try to convince you that pollution and environmental regulation is an important issue, not only in general, but it's an important issue in economics and for economists. And I'm going to argue, particularly, that it is important to perform policy evaluation. And this is something, as economists, we are able to do. So the talk will present a methodology to evaluate environmental regulations, which is called the use of structural models. And structural models combine economic theory and econometrics. And at the end, after presenting an overview of this methodology, of course, I will not be too technical, but I'll try to give you some overview of this methodology. I will present some results on the effects of environmental regulation on the automobile market. These results, I will not claim they are exhaustive. And I will have to admit that I will mainly rely on my own results I have achieved in this area. So as an introduction, I should try to motivate you why should we care, why pollution is an important issue, and I don't think I have to spend a lot of time. And instead, I'm going to show you three pictures. This is a picture of Paris in a very clear day in December 2016. You can see very well the Eiffel Tower. Another motivation for why should we care about pollution is this bear on a very big iceberg, as you can see. And the last motivation for why should we care about pollution is it affects our life, it affects our society. And this is very recently, just last month, we could see a heat wave in France. And so if you were not in Brittany, you probably experienced high temperature. So the main conclusion of this three picture is that air pollution matters and affects our society, affects our lives. So I think it's clear that we should care about it. Pollution is a word, important terminology, that actually can be split in two. We mean by pollution, either global pollution, but also local pollution. And these two are quite different actually. So we talk about global pollution coming from carbon emissions. Carbon emissions, carbon gas is the main greenhouse gas. And this greenhouse gas we know are responsible for climate change. So basically this pollution is global because when you emit emissions somewhere, it doesn't stay here. But instead it goes through the ozone and has consequences globally. And as opposed to global pollution, we talk about local pollution. And as its name suggests, it is local. And it has a direct effect on air quality, the quality of air we breathe and ultimately translate into health consequences. So those local pollution are not coming from carbon emissions but are coming from nitrogen oxide, carbon monoxide, gas and particulate matter. And these two sources of pollution, global pollution and local pollution, can be found in transport sector. Transportation sector, I can give you a few figures here to illustrate why we should care about transport when talking about pollution. It's because cars are responsible for around 12% of carbon emissions in Europe, which is a significant share. Moreover, there was a study that estimated 3 million tons of nitrogen oxide are released by the transport in 2013, which is a quite large number. And NOx is known to have direct effect on health and have some adverse effect on health. In particular, we can also quote this study, recent study of the World Health Organization, that estimated to 400,000 premature deaths are caused every year by air pollution. So air pollution is important and pollution at least some of it is caused by the transport sector and by cars. So this is why car regulations are important to analyse. And what kind of regulations we can find in Europe? So we can find actually some regulations both at the European level but also at the national level. In terms of global pollutants, as I said, carbon emissions, so I will discuss it later. But we have regulations that target carbon emissions and we have regulations that target local pollutants. The two are separated. For local pollutants, they have been regulated through car emissions standards, Euro norms, I don't know if you heard about this. So basically all the new car models have to have emissions below a certain standard. And the European Commission set these standards. In terms of, so okay, this talk will not analyse regulations that target local pollutants. But instead I will focus on regulation that target carbon emissions. Why is simply because there was more regulation targeting carbon emissions than local pollution for so far. And so it was more relevant to analyse these types of regulations. A remark here just to be clearly on the same page is that basically when you're driving, you emit CO2 emissions and these are directly proportional to fuel consumption. So the more you consume fuel, the more you emit CO2 emissions. So basically what it means is that regulating carbon emissions is strictly equivalent to regulate fuel efficiency of cars basically. So what kind of carbon regulation we could see in Europe. So actually the European Commission started to care about carbon emissions of cars quite early in the early 90s. But there was no regulation attached. It was basically the European Commission was setting long-term targets for average carbon emissions of new cars. So basically what they were saying is oh it would be great if by 2005 we would reach 161 grams of carbon emissions for new cars. And there was no regulatory aspect in this and no regulation attached. What happened instead is that the European Commission was sitting with car manufacturers and car manufacturers were saying okay okay we agree we're going to do our best to reach this target by 2005 and that's what we call a voluntary agreement. But there was no mandatory things, no incentive regulation with this voluntary agreement. But in 2010, these voluntary agreements were transformed into carbon emission standards. So instead they moved from voluntary agreement and long-term targets to mandatory emission reduction targets. And so since 2015 actually, so in 2010 they decided to impose a carbon emission standard and the standard was actually effective only starting in 2015. So what they say is from 2015 every car manufacturer has to reach an average of at least not more than 130 grams per kilometer of carbon emission for the cars they're selling across Europe. And what happens is if you don't meet the standard, if your cars you're selling emit more than what's required you have to pay a penalty. So the penalty is said to be non-linear and so basically for the first gram you have to pay five euros on each car you're selling in Europe. So even if five euros seem smaller for a car if you multiply by the total number of cars that are sold in Europe it can be a large amount of tax that car manufacturers have to pay if they are not complying to the standard. So what about this emission standard from 2015 to 2018 is actually there's nothing very interesting here because all the car manufacturers were complying. So we're really below the standard. So in 2015 we had an average of 118 grams overall on overall car so we're really below the standard. So perhaps the standard was said to be too lenient, perhaps car manufacturers anticipated correctly the standard and reacted to meet. But basically it's not a regulation that was really how the binding nature so far. But what we have is a much tighter standard in 2020. So the standard in 2020 it will be 95 grams per kilometer which implies a quite a large decrease from 2015 to 2020. So this is what's going on at the European level to regulate carbon emissions of cars. At the national level basically the European Commission doesn't impose anything. They let each country to do what they want with their national carbon emissions for cars. And so I'm going to show you that actually all, I'm not going to show you that all but I'm going to tell you that all most of the European countries have adopted some tax incentives to reduce carbon emissions of cars. And those are actually quite diverse. So here I'm going to show you a few examples of national policies. So the first is France. We are in France. So it makes sense to start with this. Plus it's an interesting one. In France since 2008 we have a Peabate system. So if you have been buying a car in the past 10 years you probably know that each car is associated to a bonus malus or a malus. So basically if the car emits more carbon emissions at a certain level you have to pay an additional tax when you purchase a new car. On the opposite, if the car emissions are lower than a certain threshold you get a rebate. So you get basically a price discount when buying a new car. So what's kind of funny is that this Peabate was introduced so we have 10 years of Peabate. And this Peabate scheme has changed quite a lot and I think it's funny to compare the 2008 situation with the 2018. So in 2008 we could get a rebate 200 to 1,000 euros if your car emitted less than 140 gram. And you had to pay a tax of 200 to 2,600 euros if the car emitted more than 160 gram. As on the opposite in 2018 what you get is a rebate, a huge rebate, 6,000 euros for cars emitting less than 20 gram. So what does it mean 20 gram? It means that the car is electric basically. So the rebate is only targeting electric vehicles. And on the opposite you get to have to pay a fee as soon as your car emits more than 130 gram. So it's basically a lot of the cars. So this Peabate scheme progressively turned into a subsidy for electric vehicle and a tax on the rest of vehicles. So it's funny to see this. But I think a Peabate is quite an interesting policy. Why? Because it combines two incentives that goes in the same direction. First you subsidize fuel efficient cars. So you're going to push demand for fuel efficient. And on the other side you tax high carbon emission cars. So you're going to decrease, you're going to give incentives to substitute away from polluting cars. So it's two effects go in the same direction. In other countries in Europe it's not that fun. In Germany since 2009 consistently there was an annual circulation tax. So basically every year if you want to drive your car you have to pay a circulation tax. And this tax is based on how much carbon emission your car has. Belgium as always is divided. You have two different systems, one in Flanders, one in Wallonia. And so maybe not so surprisingly in Wallonia you have a Peabate system like in France. And in Flanders you have CO2 based registration tax. Quite simple. It's not the same as an annual circulation tax because you pay it only once when you register the car, when you buy the car and register it. In Spain you also have, let me try my Spanish, Impuesto Especial sobre Determinando Medios de Transporte. It's basically an additional tax based on the gross price of the car. And this additional tax, the tax rate is actually a function of how much emissions, how much carbon emissions the car has. So you can see various tax systems across European countries, but all of them are really targeting carbon emissions, want to improve carbon emissions of cars. So thinking about the regulations, the fact that there are some regulations to decrease carbon emissions, what does economics have to say? How can you contribute to the debate? What kind of question, as economists, you would be interested to ask and answer, try to answer at least, is what are the effects of regulations? Yes, basic. Are the regulations working? Are they doing what they want, which is to decrease carbon emissions? So this is a basic and first question. The second is already more difficult to answer, what is the question of efficiency of regulation? Are regulations efficient? And this already, you have this word efficient, which you have to put a meaning on it. And as economists, I think we are able to put a meaningful meaning on efficient. So basically try to say what is efficiency, what are the criteria to say a regulation is efficient. Another question is, given the various instruments to regulate the automobile market, we want to ask maybe whether all regulations are equivalent or not, are there some instruments that are more efficient than others? And again, if we think about efficiency of one instrument versus another, you have to think about what is my definition of efficiency. Another question economics can ask is, how is this regulation affecting agents, the different agents? And are they better off or worse off? What are the distributional effects of such regulations? Who are the consumers or agents that are gain with this regulation? Who are the ones that lose? What are the conditions we can address as economists, I think? And also, so this regulation targets you to emissions, but they have after effects. They change what's going on at the car market, and it's important to think about what are the after effects. For instance, what are the effects of carbon emission regulation on local pollutants? Local pollutants are not at all targeted by such policies, but they will be affected by it. And it's, I think, interesting to think about effects on other elements that are not directly the target of the regulations. And also you can think about the national regulation as effects on other countries. We know Europe is integrated. We have open border and open trade. So if the markets are integrated, one national regulation maybe is affecting other countries. And this is also something we want to know, whether this is happening. But knowing whether there are after effects, yes, no, I think it's not sufficient. As economists, we want to say, by how much? Okay? It's easy to say, okay, this regulation worked. Yes, we see a decrease of carbon emissions. But what is, I think, really crucial is to be able to say, by how much? By how much is regulation decreased carbon emission? By how much consumers were affected by the regulation? By how much prices have increased because of this regulation? And so I think what's important here is not to say yes or no, but to say how much. And this is what we can do because we are economists and also econometricians. And so what I'm going to tell you is that these questions can be addressed through a lot of methodologies. But in particular, we can address them using structural models, structural approach. And this is what I'm going to be talking about, structural models. So what is a structural model? It's a model with structure. Yes, of course. What is a structure? A structure is simply saying an assumption. So we're going to rely on assumptions. And assumptions will be basically on agent's behavior. So we're going to, instead of trying to link one outcome with one explanatory variable, we're going to try to specify agent's objective and their objective function and try to start from their objective function, their behavior, their optimal behavior, and derive the outcome. And the outcome is the outcome we observe. And we're going to try to match reality with theory. And this is going to imply you some parameter, some estimated parameters. I'll be more precise about this, of course. And then, yeah. So a structural model is also referred to as indirect difference. Why is that? Because as I said, it's not going to try to directly target, directly match, directly link an outcome of interest with regulation. But instead, you're going to specify a model where you have agents. These agents are behaving optimally and this is going to lead to outcome. And with this model, what is nice is that you can play with it. So play with it means you can perform counterfactual simulations. But basically it means you can play with it, meaning that you can think about other contexts, other situations. So if you think in the context of environmental regulation, you can think without regulation, with another regulation, with the most stringent regulation, what would happen to the market? And this is what we are able to do with a structural model. OK. And so why we want to use a structural model to evaluate regulation in environmental regulation on the car market? I can think about at least three reasons here. So if you think about, let's simply think about, we want to evaluate the effect of a carbon emission tax. Why not simply compare car sales before the tax and after the tax? And OK, this is the effect of the carbon tax. It's not a relevant, it's not that simple. That's my answer. First, the time is a confounder. Before the tax and after the tax, time has passed and things also evolve over time. If you think about car sales, the context of the car market, gas prices, for instance, evolve over time the response to if price of gas increased between before the tax and after the tax, you're going to not only capture the effect of the policy, but the effect of the policy and the effect of price increase for gasoline. Second, direct inference allows to say something about what's the effect of a tax on carbon emissions, but it cannot really say anything about the underlying mechanism. What was the, how these gas tax, these carbon tax actually changed carbon emissions is something we're interested in and we can say more using a structural model because we model agents behavior. So we know which agents, we will be able to say which agents are responsible for this change in carbon emissions. And reason number three is that we are interested in a lot of outcomes that we cannot directly observe. So for instance, if you think about utility of consumers, if they are now subject to a tax, the utility is going to be changed, but we cannot observe utilities of consumers. However, under a model, we can express utility as functions of parameter of the model and observable outcomes. So this is, I think, three reasons why a structural model is relevant and maybe sometimes more than a more direct approach to evaluate environmental regulation. So let me start with the technical stuff. An overview of a structural model, of a structural model to describe the car market equilibrium. So this is what we call the BLP model. It's quite, it's widely used and it not only has not only application for evaluating environmental regulation of the car market but it's widely used in empirical IO to describe a market where goods are differentiated. So when you think about building a structural model, you have to ask yourself, what are the important features of the market I want to put in my model so it's realistic enough to describe the reality? And the question you can ask yourself is who are the agents? So in the context of the car market, if I want to build a model to represent the car market, I think the agents are simply the buyers, the car buyers, you, I, and the car manufacturers, the one who produce and sell the cars. Then you have to ask a question of what is the character, what is the exchange goods? So here we exchange cars. We buy cars, manufacturers sell cars and what is the characteristics of the cars? And I think one important feature of the goods of cars is that they are highly differentiated. You think about a Clio and a Ferrari and these two goods are in the same market. So the differentiation of products is an important feature of the car market and you want to take it into account. And you also can think a little bit more and say, well, if cars are so differentiated, why? It's because maybe consumers are very heterogeneous and they have different needs and different preferences for cars. So about the agents, going back to the agents, what do they do? Consumers consume, we all know this from the first year in micro. Yes, so car purchasers purchase a car. Car manufacturers, what do they do? In the model, they're going to set prices. Of course, they do also other decisions, which is such as designing new car models. And they also invest in technology process to produce these cars. But in my model, unfortunately, I will abstract from these decisions of designing cars and producing cars because I'm going to say these are medium, long-run decisions and what I'm going to be focusing here is short-term decisions of car manufacturers which basically can be seen as an immediate the effect of the reaction of car manufacturers to an unanticipating regulation. So I'm going to be abstracting from this. Of course, it's a limitation, but at least I'm honest on this. And so this is what we do in structural work. We make assumptions, but we are honest about it and we know that the analysis is valid within this scope and there are these assumptions. So what are the objective functions of the agents? Consumers, they're going to maximize their utility. We know this. And manufacturers, they're going to maximize profits. This is not very hard to figure out. And the last question you want to ask is how the regulation we want to analyze is going to affect agents. How is the regulation? How is the carbon tax? Is going to affect consumers and it's going to affect car manufacturers. So consumers, because now they are facing, for instance, let's think about carbon tax. Each car is going to be associated to a different level of tax. And so they're going to substitute when they were purchasing, for instance, a Ferrari before regulation. Under regulation, maybe they're going to buy now a Clio. So this is how you model the effect of environmental regulation on consumers. On sellers, on car manufacturers, of course these ones are not directly affected by the regulation because the tax, let's say, is the tax on consumers. But of course car manufacturers are not stupid and they anticipate that consumers are going to have to pay extra money to buy the Ferrari. And so how are they going to react to this? They're going to change the price of the Ferrari. They're going to change the prices to, because they anticipate consumers paying a tax. Now the equations. So this is how to represent the demand. So we have here a discrete choice model. So it's thinking about consumers making a decision of whether or not to buy a car, which is discrete, whether I buy or not. And if I buy, of course, I have to choose which one, whether I buy the Ferrari or the Clio. And so it's discrete because you make a decision of which one or zero. Okay. And so we, as I said, in structural work, we start from specifying agents' objective function, which is here utility. Consumers maximize utility. So what we're going to specify as econometrician is utility function. So here I say utility of consumer i associated to a car j is as follows. It depends on observable characteristics of the car. So for instance, these characteristics include fuel cost, weight, cylinder, whether the car is a coupé or not. So utility will also depend on how much you have to pay for this car, of course. So this is in pj here. And there is also a lot of characteristics that unfortunately, as econometrician, we cannot observe. So for instance, if you think about design, it's really hard to measure the design of a car, whether it's a good or a bad design. This cannot be measured. So all these characteristics that affecting consumers' choice, but that the econometrician does not observe, is going to be all grouped in this term, so xi j. And because everyone is different, and as I said before, consumers are very heterogeneous, there is an inducing quadratic term, which is the epsilon ij here. And this epsilon ij here is both individual and car specific, meaning that I may have a high taste for Ferrari, because I like Ferrari, but I have a very low taste for Clio. And I believe that I have a very bad taste for this car. But you have a very different inducing quadratic term. And this idiosyncrasy is, as econometrician, what things that we cannot explain, the residual, and we're going to make an assumption that these residuals are independent across individuals or across products, across cars. And this we're going to have a specified distribution for this error term. And in this case, we suppose it's extreme value, which implies after this logit model. You probably heard about this logit model. So this discrete choice model belongs to the class of logit models. And so these are the main variables of the utility. What are the parameters? So you have the beta i, which is basically the valuation of observable characteristics. So it represents how fuel costs transfers into utility. Alpha i is the price sensitivity, and it tells you how much you're sensitive to a price increase. Okay. And these parameters, because we think it's important to represent in this market the originality of consumers in this market are individual specific. But if you think about a large market with a lot of individuals, we don't want to estimate 6 million parameters. Of course we cannot do this. So instead what we do is we make distributional assumptions. We're going to assume that these parameters are all normally distributed in the population, which allows us to move from, let's say, a market of 6 million people, 6 million parameters, to two parameters. We just have to estimate a mean and a standard deviation of this distribution. So this is what we're going to do. Assume these normal distributions for these beta i and alpha i parameters. And these parameters is ultimately what we want to estimate. Okay. So how do we do this? Of course we don't, as I said, I go back. We don't observe utility. If we could observe utility of consumers, we could just estimate by linear regression. But we don't. So instead what we do is we're going to assume a quite mild assumption, which is the consumer is going to choose the product that brings him or her the highest utility level. So this is a quite mild assumption, okay? Assumptions that consumers maximize utility is not so strong, I think. And given this assumption and given the assumption we made on the distribution of the idiosyncratic term, what we can express as function of parameters and variables is the probability that one consumer, i, is going to choose the car j. So basically, and this has the following expression here, the exponential here, x beta divided by one plus sum of exponential, blah, blah, blah. So basically what it means is that if you have high x beta or low alpha p, then you're going to have a higher market share. But also what's a lower, sorry, lower probability of buying. But also what we can see is the probability of buying car j not only depends on what you can get as utility from car j, but also what are the other products on the market. This you can see in the denominator. You have the sum over j prime, sum over all the other cars in the market utility you can get from them, okay? So this is the probability that a consumer i buys car j. But this again we don't observe. I cannot observe my probability of buying a car. So we don't observe this. But instead what we observe is the market share of car j on the market. And the market share is what? It's just the aggregation of individual probability. The sum of individual probability. And so sum means integration. So basically I can express the market share of my car j. As my Clio as a function of parameters of the model. It's just simply the integral over all the distribution of alpha and the distribution of beta of the individual probability of buying. And so market share we observe. So basically we're going to try to match this theoretical expression to market share we observe in reality. And this to do what? To estimate the beta and the alpha parameters. Unfortunately what's going on is that we have the xi. We talk about the xi is what the econometrician doesn't know. So we need to deal with this. And what's going on is that the xi is going to be the error term and is what we're going to use to construct moments condition. So one contribution of the paper by Barry Levinson-Pakers is to prove that this equation here has a unique solution in xi. Because you can see that the xi j is here but also here. And that s j depend not only on xi j but on all the xi j prime. So basically there are some very good researchers that prove that this equation has a unique solution in xi. And so we can solve this equation numerically to recover xi 1 to xi j as function of parameters of the model and observable market shares and characteristics of the good, prices of the goods. And this with the xi we're going to construct what we call moments condition with instruments typically. And this moments condition we're going to use them. We're going to use the generalized method of moments to estimate the parameters of the model. So I remind you the parameters of the model are what? All the valuation for characteristics. But because we assume a distribution of individual preferences we only have average to estimate and standard deviation. Okay. So now think about what's going on under an environmental regulation. To be more interesting I think about the fee bait here. What's happening on the demand when you introduce a fee bait? So now instead of having to pay pj, I have to pay pj plus lambda j, lambda j is a fee bait. So what's going on is that it's going to affect the probability of buying. Okay. So yeah, fee bait changes the demand condition. On the supply side what we are modeling is car manufacturers that are selling several cars, car models, several brands and set prices to maximize profit. So we express price as the sum over all the car I'm producing, a manufacturer produces p minus c, the margin basically price minus cost times the sales. The sales are market shares time market. Okay. So you maximize this. You get a first order condition associated to price optimality. And what we can see is that the price depends on the cost of the goods, of course, the demand elasticity. So how the market changes when price changes. But also the diversion to other cars the manufacturer produces. Okay. So under a fee bait what's going on? So the profit function under fee bait is the price that they set minus the cost times the market share. But this time the market share depends on the price they set but also on the fee bait. So basically car manufacturers anticipate that demand is shifted by the fee bait. And so what we can show is that the fee bait has the same effect of a cost reduction for a rebate or a cost increase for a fee. Okay. So I think I'm running quite out of time but I will tell you a bit of my research on fee baits for France. And I hope I will raise your curiosity to read my paper. Of course I have to make some advertisement for my work. So in France what happened in the year 2000 is two regulations. In 2006 there is an introduction of an energy label, the classes of CO2 emissions. And in 2008 the fee bait I talk about is basically linking these classes of CO2 emissions to either a rebate or either a fee. Okay. And so the first paper I wrote is to try to understand what is going on on the emissions of cars between 2003 to 2008 and in particular to measure the contribution of each policy to the decrease we observe in carbon emissions. And so what we do is develop a structural model of demand and supply for the car market and try to measure to simulate we don't try, we do. We simulate what would have happened on the market if there was no regulation. If in particular we found evidence that this regulation changed consumers preferences towards carbon emissions and so we simulate what would have happened if consumers didn't change their preferences. We also look at the effect of gas price increase because between 2003 to 2008 there was a significant price increase in fuels. And so what we can say at the end is that changing preferences accounted for 40% of the decrease in CO2 emissions of the period. Fuel prices were responsible of 13%. The monetary effect of the fee bait, the fact that you have to pay more for high carbon emission cars is responsible for 14%. And so residual, which we interpret as technological progress is responsible for 33% of this decrease. So this is what we can do, an example of what we can do with structural models and counterfactual simulations. Then this is a recent TSE working paper I've been working on recently. Is distributional aspects associated to the fee bait because by definition fee bait imposes a tax on subsidy so basically implies winners and losers. And so what we ask here is whether this policy achieves some redistribution or not and who are the consumers that are better off and who are the consumers that were worse off because of the fee bait in 2008. And what I think about is that it actually depends on how we finance the fee bait. The French pollution values because actually it costs 228 million euros in 2008 and this has to be somehow supported by consumers. And depending on whether you impose a uniform tax meaning that everyone paying the same or a proportional to income tax, the redistributive aspects are different. And this is what I found here is we have a bell shaped curve when we impose a lump sum tax. So basically what it means is that the fee bait policy favoured middle income class consumers at the expense of low income and high income consumers. However as soon as we introduce a proportional to income tax to subsidize the fee bait, we have some redistribution. We can see that the effects on consumers is almost zero until 25,000 euros for an income below 25,000 euros and then consumers are worse off, the richer the consumers, the more welfare loss they experience. And in this paper also I investigate the effects, the after effects on air quality. So as I said because fee bait only targets carbon emissions, what happens to the emissions of local pollutants? And actually what we find is that the fee bait is responsible for an increase in all the emissions of local pollutants but by not so high levels. So the increases are relatively small but still it increased emissions of local pollutants and where it increases the most is actually maybe a bit problematic because the nitrogen oxide and the particulate matter increase the most in rich and dense municipalities which is exactly where the pollution is the most problematic. The last, I don't know how much I can spend. 10 minutes? Let's see. The third piece of work I've been doing with, I was doing, it's done now, it's forthcoming with my co-host Mario Samano, is about comparing regulations. As I said, we have different instruments. So it's important to say something if to know whether one instrument is better than another. And so what we do in this paper is to compare fee baits to fuel economy standards. So fuel economy standards, I told you before at the European level it works the same so you have a standard that each car manufacturer must meet and if the standard is not met, then you have to pay a penalty for non-compliance. And so of course comparing two regulations is not so easy. And so what we propose here is a smart way to do this comparison because we are able to construct equivalent policies and compare them in terms of welfare effects on consumers, manufacturers and in terms of carbon emissions. And so what we mean by equivalent policies, it means that we are able to, for each standard we want to impose in a country, we are able to say what would be the equivalent fee bait and equivalent means it would reach the same fuel efficiency and the same tax revenue. And so by keeping this outcome constant across the two policies we can focus our comparison on effects on consumers, effects on manufacturers and total welfare effects. And so what we find, so here it's not so pretty as graph but it shows the variation of welfare due to different standards. So this is zero standard and this is high standard for France in blue and for US. So we do the analysis for both France and US. And what we find is for any level of standard regulation, any stringency of policy, the fee base dominates cafe standards for both countries, for both UN and France. So it seems that fee baits are better in terms of welfare effects and for any kind of policy stringency. But however if we look a little bit closer in terms of manufacturers' profits, we can see that not all manufacturers are better off under fee bait. If an aggregate profit is higher under fee bait, some manufacturers would be better off under a cafe standard. And these manufacturers are this here, Fort Toyota, Renault Fiat. And why these manufacturers are better off under a standard is because they take advantage of these manufacturers being very hurt by the regulation. While under the fee bait these guys are not losing too much. They lose but not too much. So they cannot gain so much under fee bait. So this is the results we have on comparing instruments. And the last piece of work, I'd like to say two words on this because it's a work in progress. So as soon as I'm not asked to do an opening lecture, I will work on this. And it's about externalities across countries. So as I said, it's important to think about whether a national regulation affects other countries. And the question is, is the fee bait going to affect German consumers? The answer is no if the two car markets are independent. And another answer is no if the market is perfectly integrated. Because if you have a perfectly integrated, only one market, it's unlikely that car manufacturers would react to a policy that is only affecting a small share of the market. But actually the reality is in between. We don't have perfectly integrated markets, but markets are not independent either. Because we have an open border, we can totally buy a car in Germany and import it to France. This is what we call parallel trade. And so because of this existence of parallel trade, a national regulation is going to affect another country. And the question is by how much? And this depends of course on the importance of parallel trade, whether markets or more, how much parallel trade is there. And so this is something we're working on and it's not easy because we need a model of car markets with imperfectly integrated markets. And this to my knowledge has not been built. So the first part of this project, before analyzing the effects of environmental regulation, is to build a new structural model. So which leads to my conclusion, I hope I convince you it's important to quantify effects of environmental regulation, and this can be done using structural model. And performing counterfactual simulations. Of course I said it before, but I provided here an extract of some results on policy evaluation. This is not at all what's only there. There are a lot of work on environmental regulation. And I just show you my work because I think I'm able to talk a lot about it. So it's better. But I also think there is also more to be done. And in particular in quantifying new effects, investigating new types of regulation. But for this I think it's not easy because we would have to develop new models, I think new structural models to investigate new questions. Thanks for your attention.