"The Future of Finance" - EDHEC Speaker Series

[Music]
EDHEC Business School
EDHEC Speaker Series - The future of Finance
Estimaring corporate carbon emissions with machine learning

Raul Leote de Carvalho
Deputy head of the quantitative research group
BNP Paribas Asset Management

Tuesday, 4 April 2023
6 P.M - 7 P.M Paris time

Emmanuel Jurczenko
So Raul Leote de Carvalho is actually deputy head of the Quant Research Group at BNP Paribas Asset Management. His team is responsible for supporting research, development of systematic strategies using big data are just to try to support Alpha sources. Raul is very senior, got more than 22 years experience in quantitative investment experience and I discover in fact a few days before that, Raul in fact got another life before quantitative investment. So here the PhD in theoretical physics from University of Bristol. Get MDC in condensed matter physics based in chemistry. So you can see that there are different paths in fact to venture in quantitative finance. And we are very fortunate to have Raul today. Raul is based in Paris and he will talk today about how in fact to estimate non reported corporate carbon emission using machine learning. So, without further ado, let me turning to Raul. Raul, thank you again and the floor is.

Raul Leote de Carvalho
Today. I mean it's really great pleasure to talk to you about the work we have been doing on machine learning for predicting carbon emissions. So we can just go to the next slide where essentially we give the motivation for this work. So, I mean, we are an asset manager. We are a BNP Paribas Asset Management. So why do we spend time constructing machine learning models to estimate carbon emissions? So if we go to the slide, before essentially as an asset manager, fund managers do need to know about company greenhouse gas emissions. These days, I mean, asset managers acknowledge the need to accelerate the transition towards net zero. In fact, there is the net Zero Asset Managers initiative. So this is in slide two. The previous slide, there are 301 asset managers which sign to the Net Zero Asset Manager initiative which is actually with a total of $59 trillion US Dollars, and the assets under managed that will essentially have to converge to zero emissions by 2050, that's the commitment of the asset managers. There's also the Glasgow Financial Alliance for Net zero. It has 550 members from the financial industry from seven specific net zero alliances.

So yes, the financial industry is now committed towards reaching net zero by 2050. And the only way to do it well, you have to know the emissions of the companies because you finance companies, you invest in companies and you do need to know about how much of the emissions are they responsible for. So that's one of the reasons for essentially our strategy and to know how we're going to continue to invest and to finance companies. The second reason is because companies with higher carbon emissions face higher regulatory and legal risks. So I mean, just from a risk management point of view and when we decide to invest in companies, it is important to know how high the carbon emissions are. The other reason is also because increasingly our investors have not just financial utility, but they also have sustainability utility and they want to have a positive impact on environment through their investments. So it's actually our clients that also are asking us to take action when it comes to build portfolios. And in some cases they do want to have very specific constraints related on how they build their portfolios and constraints in particular related to carbon emissions and decarbonization.

So these are some of the main reasons why we need to have carbon emissions from corporates. Now, the issue is, and I will show that in a minute, is that there's not that much published data. In fact, a lot of the data that exists tends to be estimated. And what we found a few years back is that the models that they use by a number of data vendors, perhaps not of the quality we would like to have. So that's why we really embarked on this journey of calculating ourselves to have a tool that will allow us either to have numbers or to actually just check the quality of the data vendor so it can use different ways. And so far what we have done is produced a model for scope 1 and scope 2. So talk about that. So if we now indeed move to the okay, we can stay on this slide. Sorry, we can go back to the slide. So when you talk about carbon emissions, what are we really talking about? Well, first, the greenhouse gas protocol, which was essentially proposed in 2015 and these days is what is mostly used and sees essentially the emission of greenhouse gases as we have in this slide first is not just CO2.

I mean, what it's done is there's a way of converting emissions of other gases like methane and so on into a weight of equivalent CO2. And the way that is done is by well, it's known what is the warming power of different gases. So you can calculate an equivalent of how much weight of those gases in terms of emissions is to 1 ton of CO2. So you have a way of converting emissions of methane and other greenhouse gases into an equivalent weight of emissions of CO2. And that's essentially what is used. So when we talk about carbon emissions, usually you talk about equivalent CO2 emissions, which means all these gases that are here in this slide, they have the weights converted into equivalents of CO2 emissions. Then we have CO2 equivalent. That's what we really talk about. And then I already mentioned scope 1, scope 2, and there's something which is scope 3. So what are these? Essentially, according to this protocol, the carbon emissions and those of all the gases are divided into different scopes. And when we talk about scope one essentially we're talking about the direct emissions from a company that typically come from what is known as stationary combustion which can be burning oil, gas, coal and other things in boilers, furnaces but essentially by the company itself for its operations.

Can be also mobile combustion from the cars and vans and the trucks controlled by the firm. One by the firm and can be process emissions, for example, if it involves chemical production or the equipment that releases CO2 for example cement manufacturing. So this is also included and also includes fugitive emissions which can come from for example leaking in the case of methane, that's really a critical issue. But these are direct emissions from the company itself. Then you have scope 2 and scope 2 is indirect emissions. They typically associated with the electricity that is purchased by the company or steam or heat or cooling but essentially is what is bought indirectly by the company and this is what is scope 1, scope 2 and then there is scope 3. Scope 3 is essentially all the rest. It's also the more complete and more complicated it can also be the bigger, much bigger than scope 1 and scope 2 added together. And scope 3 involves what is known as upstream activities and downstream activities and essentially is everything that is in the supply chain of the company and that is not direct or indirect. So I mean, you can see for example employee commuting is part of scope 3.

You can see other things that are part of scope 3 includes the fuel from related activities, the transportation, distribution, bring materials for essentially the production purchase goods for example, that generate emissions and then downstream is essentially from the products, the emissions from the products or from the distribution of the products from the recycling of products. So all that and there's essentially 15 of those scope three depending on how you really put it together. So that's what we call carbon emissions. Now, if we go to the next slide in slide four, this is the picture on 2018 December. That's because of what we use for the calculations today. There's a few more companies publishing data but what we have in here is the number of companies that provide already-reported data on emissions.

Emmanuel
Raul, if I may, in term of the universe, when you mean broad global universe so what are you talking about?

Raul
Yes, so I mean, if you go through all our funds you find potentially these 15,726 companies at least you would back then when we started the analysis. And so if we want to report on every single company we can buy into our funds, that's our investment universe. So that's what we mean 17,000. Of course, I mean it includes a lot of micro companies and a lot of small companies. You can see there's the 3000 largest cap universe. That's for the largest cap funds, of course you go on a small universe. We also have what is available in Bloomberg where in terms of the universe they consider which is 11,700. So the issue is we potentially can invest in or you could it's probably more even now back then on investing 15,726 companies. And when we look at what we have reported we had only about 18% of scope 1 and about 17% of scope 2 reported and scope 3 even less, only 12%. But even if we go for the large caps of the companies that appear in our mainstream funds, flagship funds, which are invested typically in the largest capitalization so talking about the 3000 largest micro capitalization stocks, you can see that still it's less than, or it was at the time, less than half that were actually reporting.

And on scope 3 it was only 33%. So even for those funds, I mean if you faced with a decision do I want this company or that your criteria is carbon emissions, you can see you really have a problem because you don't have the data for half of your universe. So it's not fair. I mean, you can say well then I don't invest. But if you say I don't invest, you just forget about quite well. Most of 80% of our investment universe is really throwing it in the bin, which also is not something that we want to do. I mean, I'm sure there's good companies that do not have a lot of carbon emissions there. So I mean, this was what we were confronted with and that's why we embarked on this. The second reason we embarked on this is because when we ask some data vendors to extend, because they use models to estimate missing data so they have more coverage, but it's based on their own models. And what we realize is that most of the data vendors, when we talk to them, in particular back then, I mean, this also has been changing, but in particular back then, they had extremely simple models, typically linear models, which based on very few variables, typically size, number of employees and sometimes it just stopped there. And so, I mean, there was this other problem that when we started looking, we realized that the models that some data vendors had were really not very sophisticated and not that powerful and they were far from what we wanted to have because our objective is indeed to have leader on sustainability and sustainable investing. So that's why we embarked on constructing these models. And also back then, I mean, we talked to in particular some academics from the University of Otago at one of the Grassfield conferences and we realized that they were using machine learning. It's based on the discussion that we came up with, this way of modeling. So if we go to the next slide, I just mentioned what some academics were doing back then and in slide 5 we show what we have changed relative to the models that existed previously based on machine learning. When we looked at the paper they presented at the conference and that eventually they published, we realized that, well, we can go to the yes, exactly. So we realized that their effort was on developing essentially one model per industry. And sometimes if there was not enough data, they were kind of putting some industries together in terms of gigs, three gigs, two levels and so on.

So they had a way of putting together industries based on availability of data and what they could do. But they're really aiming at one model per industry. That was their aim. And what we realized is that some industries, for example insurances, the predictions of the models they were putting together out of sample were really poor. And also they were using machine learning models using a methodology which essentially just throw everything that we can from machine learning algorithms and let's really try to pick up the best and construct something which for us was a problem because for us having some transparency of what is going on is also very important. Not just forecasting precise numbers, but knowing why we're precision forecast numbers because we have to explain why we're doing and what we're doing. When you rely on the data vendor, you can say well, rely on the data vendor, the model perhaps is not good, but you just say that if we're going to forecast ourselves, we have to say also why we're forecasting in a particular way and why we have carbon emissions in one way, which was a challenge. So what we opted, it was first we decided why not create a global model and just use sectors as an input.

And we realized that that works really quite well and solves the problems of out of sample predictability in the industries or sectors that previously were not good. Like insurances, we actually can get quite larger squares in the results in the validation in the test sample. So that's what we went for. And then also one of the things that we realized that was extremely important was to correct errors in the reported data that we were using to construct the model. So back when we started this, this was done iteratively I mean, every time we run a version of the model we also detected the companies that were really outside in the test sample, that were really outside where we wanted to be both in the test and validation samples. And we looked at why the data was really far out and we realized that there were errors in reporting and we could do that just by going actually to check what the companies had published. And finally, sometimes the numbers had not been updated, sometimes there were cop productions that had not been taken. And so by correcting those we actually managed also to improve the model.

And in the paper we describe still the fact that we did this by hand. Today we have an automated version for doing this. But that is very important. If you train a model on bad data, you will get not necessarily a good model. If you train the model on good data, of course chances you get a good model are really significantly higher. So that's something we've learned as we went on is that the reported data, as we get it, contains errors. And being able to correct those errors for cop productions, for misreported data and really going and looking at where the data has changed really significantly from one year to the other without any apparent reason, then we decided to just omit some of this data. If we could not find an explanation or correct the data or something. So the main two innovations is indeed by switching to a global model with sectors and industries as predictors and correcting the data, and I would say the simplification of just choosing the machine learning algorithms limiting to just in the case of the paper, two algorithms and in the case of real life now we actually finally focus just on one, I will talk about that.

So this is the first part of the model is that it's a global model and we do the deceptic cleaning and we pay a lot of attention to that. In terms of what goes into the model. We can go to slide 6 and we have essentially a number of predictors. A lot of them were inspired already in what academics were doing. We completed here and there and essentially this is the data set. So, as I mentioned, we have industry classifications, which is a categorical predictor. We have regions, as you have there. We have the revenue group, whether it's a high income or upper middle income, so on also as a categorical predictor, we have whether there's tax regulations or not. And then we go into the numerical data, whether we have the CO2 emissions, we have carbon intensity of energy mix and we have a number of other things related also to carbon emissions. Then we have something specific to company, but not just revenues and number of employees. We also added things like total assets, gross property, capex, age of assets, and finally also we added variables which depends on essentially to related to the energy production and energy consumption.

So this is what we tested. I mean, there were more predictors that were tested in the paper. We really only mentioned what was finally retained for the final version of the model. So this is what we work with. Although, I mean there was actually a few more things that have been tested, but they were not really retained by the model. So we decided just not to talk too much about those. And everything is done in log terms, as you will see, carbon emissions, they can increase quite fast. So we deal with log terms. So for the numerical variables, typically most of them are also in logs. So we regress logs on logs when we construct the models, or we model machine learning in terms of logs, because that gives us is as it was done and as it's done by most people. Because the data, if you look at it just as it is really not Gaussian distributed. But if you look at the logs, then you find something really close to Gaussian distributions. So that's why in these models you model logs.

So in the next slide, we talk a bit more about the machine learning approaches that we have used. So we chose to use a linear model particularly because when we started this, we really wanted to have something that would tell us a bit why the model is doing what it is. And when you have a linear component that becomes easier, at least you can look at what the linear component is doing. And we went for elastic necks. I mean, we could have done many of the others, but elastic necks essentially gives you the compromise between the rich regularization and the last regularization and it creates essentially the best compromise between using both. So that's a kind of standard choice for a linear model. And that's finally what we went on. Now, the predictions are quite easy to explain. The issue is not as accurate as nonlinear models. And for the nonlinear models from all the things that we could use. Finally, we settled for using what is known as Extremely Randomized Trees, which is essentially a version of Random Forests, with a number of simplifications in particular on the way that you do the split, which are random, as not really chose to be optimal. And that makes it more economical in terms of computing time.

So that's what we went for. And indeed, I mean, we find that it tends to be more accurate. But when we actually wrote this first paper, well, the problem is we could not figure out how to actually come up with an explanation of why we were getting the numbers we were getting. Now, in the meantime, we started using something which is known as SHAPley values. So in the meantime, we actually dropped the linear model. We just used a nonlinear regression model and we used SHAPley values to essentially find out why we get a certain prediction from the model. And I will talk that at the end because that's something that we've done just recently. But on this model as it was presented, I mean, also we decided to go for a conservative combination. We could have chose just picked the best model or the best result of the model, but mainly because the regulator also wants carbon estimations in case of uncertainty to be conservative. So actually we not went for the best of the best estimations, we went for a compromise. So essentially we just took what we call the maximum prediction combination, which is we just take the biggest value from the results that we get, which is not necessarily the one that gives going to get the best fit, but it's what we chose and this well can be changed.

And what we do now, we go for the best fit. But when we wrote the model, this paper, that's what we went for. So if we go to the next slide, I think it's interesting to spend a little bit of time on how these things are set up, or how we did set them up. So we first separate the universe into in sample and out of sample. In out of sample, we essentially put every while all the companies not reporting, all the companies for which we realized that the data had problems because it changed too much, or there's something we really couldn't understand. So everything that was excluded was put on the side. And what we did in we're going to show at the end is we compare our estimates for those companies that are excluded because they don't have emissions, or because the emissions, we don't think they are good in terms of what they're reporting. We can compare those predictions with the predictions from different providers. And what is interesting is that some of these companies have been now publishing. So what we're doing now, and we'll talk a bit later, we're now comparing the prediction of the model with actually what has been publishing.

So today the way we judge the model is actually based on the data that companies that didn't predict in the past are predicting now. When we constructed this first version of the model, we just used out of sample to really compare with other data providers. In terms of the construction of the model, we used log-transformed data, as I just mentioned, because it's what it is normally distributed and the predictor is also log-transformed. We used the technique called cross-validation. In the next slide, I will show a bit more what it is. But for cross-validation, what we did, we took the in sample data and we partition it in 80% for training, 10% for validation and 10% for testing. And I will show in the next slide when I get there how this works. But that's essentially what we have done. As I mentioned, we have these Iterative procedures. So once the model was developed, we didn't just stop there, we actually looked at the biggest outliers and checked if there was any reason for having them as outliers. And if we find that there was potentially an error in the data, we just removed them and we reconstructed the model and we found that that indeed increased the model.

So either we correct the data, in particular if we find that there's really something strange going on there, or else if we find that there's an error, then we remove. If we find there's no error, I mean, it's just an outlier, of course we keep the data. But this was investigated and quite often we found that it was the data that was wrong and needed to be corrected. So we corrected that by hand. And finally, in terms of evaluation of the model, I mean, at least for this paper, what we looked at mainly was the R2, just comparing essentially the data points we get with what was reported and also against what is the prediction. So on top, what you get in terms of prediction, and on the bottom, what you get in terms of reported. So by comparing essentially the two, we can calculate the R2 and you do this for the test set. So it's really as it's traditionally done for machine learning models. So in the next slide we just a bit on the cross-validation. So the way it was done, as I mentioned, the data was divided into 80% for creating the model, which is essentially every time you divide the data into five and you keep one data set as the test data and you fine-tune the hyperparameters of essentially your models.

For example, for the extra trees you have to fine-tune the depth of the tree, you have to fine-tune the number of trees and so on and so forth of all these parameters. Essentially you can fine-tune by running this a number of times and running it in this way and you use the test data essentially for that. And then at the end you come up with your model and you have essentially 10% of the data which was used for the final evaluation of the data with data that has not been used to construct the model. And that's where the R2 essentially come from. So that's a typical approach to put together machine learning models using cross-validation. Slide 10, we show what the model has done in sample using these, so what we have here on the left is scope 1 emissions. This chart, the way to see it is we start on the very left with the company with the highest emissions and here we have the reported emissions with looking at the quality of the model. So we have the blue dot right on top with the company with the highest emissions.

And then you just add one company at a time and you go all the way to the company with the lowest emissions. So you have the reported emissions, scope one plotted one after the other for the companies ranked by scope 1 emissions. And the green dots is essentially what the model can achieve compared to the reported emissions. And on the right you have the same principle, the same graph for scope 2. And you can see that towards the less lower emission companies the quality of the fit is a bit less good. But in fact, that's also because we were being conservative and we were opting for the max estimation. Now do we remove that and we get a better fit in the models that are actually used in production, but for the paper and for what we are doing back then, that was the choice. So in fact they are squares also not as good as they could be because we had this conservative approach. Another way of visualizing this data is in slide 11, the next slide. We're just looking at the differences and we can see that the next slide in slide eleven, we can see that they're not exactly Gaussian, as I mentioned, we can see that indeed, we tend to overestimate carbon emissions, but that's also by construction here.

In fact, that has finally, for the very final version of the model, we decided to remove that. So we have distributions which are a lot more Gaussian. But this is how the errors of the model are distributed. And what is also interesting to see is look at the R2 of the model. It's in sample it's for the final evaluation of the model. We have R2, which indeed tend to be quite high. And what is important in particular is that they quite high for the most important sectors, which are the ones which have the highest emissions. Because as you can see also, when it comes to the emissions of sectors. Well, utilities for scope 1, which is the table on the left is the one that is the sector with the highest emissions. That comes materials and energy and transportation. And then you can see that quite fast the emissions fall quite significantly. So what is also more important is to have the best estimations for the companies that are going to be the biggest emitters, because that's what is more critical. And you can see that on scope 2 there's a bit less concentrated, the emissions, but still, materials, energy utilities have, in terms of scope 2, some of the biggest emissions.

You also have some of the highest R2 in those sectors. In any case, for scope 1, we do get above 80% for the most emissive industries. And for scope 2, we get more than 75% in terms of our square for 22 of 24 industries that we have in here.

Emmanuel
Emilie, I think you can just go to the slides because I don't think you're on the right one.

Raul
Yes, sorry. Yes, so I was talking about this slide. So you can see on the table on the left utilities, Energy, utilities, materials, energy, transportation. Really the red ones, they're the most emissions. Utilities makes 40% of scope 1 emissions, Materials makes 28% and Energy 19%. So you can see a lot of sectors. The emissions are quite small, relatively speaking, but we have quite high R2 for those sectors with the highest emissions. And you can see that in terms of scope 2, I mean, it's a bit less concentrated, but materials is really the highest in terms of emissions, the highest sector with 35%, and then comes energy and then starts going down in terms of emissions. But we have quite high R2 overall. And you can see for insurance, which was indeed one of the industries for which it was really other models really had difficulty, particularly the other machine models from previous literature which actually had negative R2. Because while we're talking about the validation so the test set that is used for validating the final results in. Terms of predictions. And you can see that here we have quite the extra trees achieve 85%.

So I mean the elastic nets less, but quite high R2. Now, what we also have done in the paper, we compared, so it's the next slide we compared with other estimates. So here, I mean the number of companies that we're showing is limited by what is available in terms of estimates from the two data providers we use, the Bloomberg and S&P TruCost. So on the left you have seven, just under 8000 companies, and on the right for scope 2, you have just the same under 8000 companies. And you can see there's differences between what we predict and what they predict. I mean, we've done some analysis. I mean, we do believe our numbers should be more accurate based on the analysis that we've done, but this is something also that has been done after we published the paper. But you can see again, that is in particular also at lowest emissions, that for this particular model the differences could be a bit bigger. But it's interesting that one of the other observations you can see that in particular as you go towards lower emissions, S&P Global TruCost models tend to be more conservative in terms of having lower figures.

Bloomberg on the other hand, tends to be more aggressive in terms of showing higher figures. But then as you go towards higher emissions, Bloomberg, that changes a bit less and becomes a bit more conservative, whereas S&P more or less converges a bit with what we are saying. So it's interesting to see that indeed reflects the fact that the approaches to modeling non-reported emissions are clearly not the and so that's one of the observations that we get in terms of comparing. So, I mean often it's also a related issue. People often say, well, I mean, ESG people don't agree on data. So I mean these carbon emissions, you can see the type of problems there are and people are competing to get the best numbers. And we kind of entered that race in here with our model. And you can see there's disagreement between the data, that is a fact anyway. Now just a word in slide 14 to the next slide on SHAPley values. So by collaborating with essentially what happened since we published the paper, we got to be known within the BNP Paribas group in terms of what we were doing. And we started collaborating with a number of other entities in the BNP Paribas group.

And in fact, the model was adopted a bit by the group and that's why there's a second iteration and the final iteration that went into production. And one of the things that finally came out from this work together with other people in the BNP Paribas Group was the use of SHAPley values essentially to have an idea of what predictors are contributing finally to the prediction. Now, the way that works, I mean, for us, which come from a financial world. We can see it as essentially the same way that when we calculate risk, we can calculate marginal contributions to risk which become additive because they take into account the interactions or the correlations with other variables. We can see SHAPley values a bit like that. I'm not going into how they are calculated. Also it's complicated. You need to use some approximations to calculate you don't really get the absolute exact shut value because that can be quite time consuming. But essentially what you get is something like this for each of the predictors. In fact, well, I chose one of the nicest ones, which is energy consumption. And what this tells you is that if the log of energy consumption is changing, as it show on the horizontal axis, so between minus 2 and 6, well, say between 0 and 4, between 0 and 4, the contribution of the log of the energy conception is essentially linear to the final result you're going to get in particular for scope 1.

And then as you go towards the extremes, that's really where the contribution is no longer necessarily linear. And the way this is done is just by you calculate and you just iterate through your data set all the possible values when you fix the value of energy consumption, that's how you calculate this. So what you have there is a calculation of the model for everything that was possible when in the data set you had the energy consumption at a given level. Now, for the energy consumption, also for scope 2, you can see it's a bit less linear as you increase energy consumption starts contributing less and less. So that's the type of feature do you capture with the nonlinear models, which is essentially what is interesting. And although the results here show that energy consumption, a linear model would not have been that bad, you do capture this bit of nonlinearity and some of the predictors, I mean, the nonlinear aspect can be more important than what I show here. So today for each of the predictors, we can indeed have this type of graphs which allow us to say, well, we have a machine learning model, indeed is extract is complicated.

Some people can say it's a black box, but this allows us to introduce transparency in the black box and say, well, but look, I mean, you have these figures because when you input your energy consumption find it is almost linear relationship except in the extremes where it may not change that much. So that's what we can show and that's quite important and that's why we dropped the less accurate elastic nets in the final model. So the next slide in terms of finally what we have today, without going into the details, because it's more or less based on what I've shown, is just some improvements in automation and just using extra trees. What we have is we actually calculating carbon emissions scope 1 and scope 2 for 32,000 companies for which we didn't have reported numbers today. I mean, we have bigger numbers in terms of what is reported. As you can see here, this is August 2022. We have numbers for which, even if we wanted, the data vendors would not give us what we would need. Now, this universe now is beyond asset management and what we use is really for the BNP Paribas Group.

So involves also knowing about every single company to which we lend, what are the carbon emissions. And anyway, if we were not to do our exercise, we could only really, for quite large number of companies, rely on estimates from Bloomberg, whether we liked it or not. Advantage of having the model is also you can tailor it more to the needs. And sometimes, for example, as we started, we started with elastic necks and extra teeth. Perhaps we want a more accurate model. Perhaps we want a more conservative model. By having our own model, we can also essentially handle all those and really tailor models. We can update it more frequently, less frequently. So that's the advantages of having our own production. So as conclusions, we have constructed. So in the next slide, we have constructed this model. We embarked on this project because we felt the need, we felt pressure and we felt the need. And so based on our contacts with academics, we found the opportunity and we now have this model in production at the level of the group it's at the moment for scope 1 and scope 2. We found that indeed, this model from our point of view, is more accurate than existing models.

The use of machine learning is clearly a big plus. Talk a lot about using machine learning in the financial industry. This is a case where we think it really allows us to produce more accurate data than non machine learning models. And so we believe indeed can be used to predict and report on greenhouse gas emissions with greater accuracy. We mentioned scope 3. So that's something we have been looking, there's been quite a lot of going on there. There's also data vendors that have starting producing. So we working on it, but it's more complicated and we questioning the extent to which we're going to continue working on it. So that's more or less where we are. But it's working process there what I've said, you can find pretty much everything in this paper, which if it's not yet open, it should be open for everyone to download very soon anyway.

Emmanuel
Okay, so thank you very much. Thank you Raul, for the great presentation. So, before I pass the floor to Jeremy, I just wonder, so the result you presented were on 2018 carbon emissions. So now you must have 2020 or 2021. So I was just to know how those results have changed or not. So, a couple of questions from my side. First, in terms of the landscape that means, do you see an improvement and in what size, what scale regarding the reported versus non reported in your universe? First question, and the second question is in term of your relative performance. If you compare to third party providers, is your model still highly competitive? Do you see third-party providers now improving their performance? Maybe that they are using something similar to what you are using? So, just in a nutshell, is there any changes with respect to your original paper and what are the evolutions with respect to the landscape of carbon data?

Raul
Yes, so very good question. So, as I mentioned, I was there at the beginning when this model was put together. Then once this started being taken by the BNP Paribas Group, I was much less involved or hardly involved. So what I know it was done, the model evolved into having all the extra trees. And in terms of what we now use to validate the quality of the model, it's actually the data that is being published by companies that were not reporting before, but for which we had predictions based on the more sophisticated version of the model. And in terms of the results, looking at those companies that have started reporting, because there's been an increase in the number of companies that report these days. And in fact, we found that indeed, we were reporting the numbers quite accurately on average, and we had the most accurate predictions. In particular, when we compared the numbers that other data vendors were forecasting, we didn't look at our scares, we look at are MSCIs and we find that we do have the lowest errors in the model. So we essentially quite happy in terms of the quality of the predictions that we were producing and the models that we have created.

And in the paper, the second paper that was published, those results are published and indeed the one that seems to be closest is MSCI for scope 1, but still they have a higher error. And for scope 2, I mean, we clearly have the lowest errors. So we're quite happy in terms of that. The differences, well, they can be big for some data providers. Thank you.

Emmanuel
So Jeremy, if you want to ask some questions, especially knowing that you're working on how to estimate scope 3 carbon emissions, I'm sure that you have a lot of interesting questions to ask to our comments.

Jeremy Gallean
I hope so. First off, thank you very much for your insight. My first question relies more on what do you do with the information you get rather than how you get it? When reading your paper, you mentioned that having high GHD emissions transcribes a transition risk which can be defined in negative cash flow from fines or operating risk from migrating to a greener competitor. My question is, has your model, which has been rolled out to the entire BNP Group, is it actionable yet in the sense that do you still rely on exclusion lists for your instruments, or has there been a threshold established, meaning that instruments with GHG emissions higher than a certain threshold will be excluded from your action?

Raul
Yeah, well, there's different types of applications. First, we do not tend to work with carbon emissions in absolute terms because that's also not fair. If you run a big business compared to a micro company, our emissions are obviously much larger. So we work with intensities because it's really about the size you have to normalize by the size of a company in some way because some companies are just bigger than other. So you don't want now all of a sudden just say, I'm going to solve the problem by investing in micro caps because they have really low emissions. They may have low emissions, but if they really exposed to electricity produced by coal in terms of intensity, it can be a disaster. So we tend to use data, at least for asset management at a portfolio level, normalized by intensity. In terms of applications, I mean, we do decarbonize our portfolios. So the intensities are indeed used to make sure that our portfolios have lower emissions than the benchmarks market, capitalization-weighted benchmarks. And then there's different targets. We do ourselves have also a net Zero policy. So we want to decarbonize essentially what we manage for our clients.

So we have targets and this is also part of what we use in order to measure and to see if we are on the right path. And there's different approaches and we're also going to change the way we are doing because indeed one of the issues with the carbon emissions and what we estimate in here is that really what we observe typically in the past, once the data is published, is based on reported data that was calculated for previous year. So, I mean, we want to be also more reactive. So we also in our approach, this is part of it, but we want to introduce a more forward-looking component. In fact, we're working on approach which is also going to see whether the companies are actually committed and doing what they say and going on the right path. So are they aligned or are they aligning or they just simply don't care. So we're going to take something awesome of what looking. But this is all part of the toolkit and the data sets that we need to get to that. So that's the type of usage and this is in asset management. Now, I don't really want to talk on behalf of other entities, but they also have similar needs in terms of reporting, understanding of the exposure of the businesses, in terms of reducing carbon emissions on their side, in terms of who they finance and how they finance and detecting, if they are on a good trajectory or not, to decide finally if they're going to continue to finance or not.

So all that is used, and I mean, what we have here is our attempt to just do it the best we can because unfortunately we cannot do this without data. And data, as you could see, is car. So we're trying to really do this in the best way we can. Also by having found out that data from data vendors was not of the quality we thought it could be.

Jeremy Gallean
Well, about that, apparently there is some issues with the data reported by companies less with scope 1 and 2 than scope 3, which has a lot of, since it's voluntary disclosures, there might be some of the 15 categories that are emitted or forgotten about. And about that, I have a quick question about your model, which has extraordinary results and with the increase of mandatory policies and frameworks forcing companies to disclose their emissions, namely the CSRD or the CC 2024 in the US. Do you think that the model can become sufficiently precise every time that it can beat companies which account for carbon emissions using a bottom up approach with lifecycle analysis and EIO, I think is what you mentioned in your paper?

Raul
Yeah, well, it's a good question. We would prefer that we just get reported numbers and that they are of good quality and don't need to do that because then we can do our job, which is manage funds rather than estimate carbon emissions. So, I mean, yeah, fingers crossed that everyone starts reporting and the data is audited and it's of great quality and we don't need this, but in the meantime we need it. So the data is not there yet and anyway, there's errors in data. Maybe at some stage the model will be used. I mean, the model on our side, the way it's used is in particular used to challenge what we get from the data vendors and well, that's already what we find is quite a good use for the model but to really make sure that we understand the data that we have. But if everyone starts reporting even better. Now, what you say on scope 3 is true and scope 3 is not necessarily going to be reported in the same way as scope 1 and scope 2. Indeed, one of the issues also scope 3 is that it can be significantly larger than scope 1 and scope 2, but also can have double counting.

So there's quite a lot of issues with scope 3 but it's clearly something that we going to start using and we're going to start using in this year, later in the year. So there's quite a lot of work on scope 3, we did start by constructing some models but we did not find results as good as for scope 1 and scope 2. And also there's a question how we do it because, well, you have all these components or do you estimate all the components? Which is not really good because in some cases also you don't have data, as you say, when you say scope 3, you don't really know how many of the components have been used, even when it's reported. So, I mean, those are some of the difficulties. So we had come up with some approaches like just reducing the number of components and trying to work with that. But all that kind of is a bit well, I'm not working on that. And from what I know, also there's some new data sets that have been produced by some data vendors which are being analyzed also quite carefully. So there's work going on and there's still a possibility that we have a model which will help us understand scope 3 and report and use scope 3 but it's work in progress and it is indeed much more complicated. And one of the things also that we've been trying is to use supply chain data. But again, supply chain data, the quality is not at the level we really wanted it to be. So it's an avenue that also we started does help a bit, but then, I mean, it's a bit incomplete. So we're also working on supply chain as a way maybe to get and solve the problem. Help solving the problem.

Jeremy
Thank you very much for your answer. I see there are some questions in the QA.

Emmanuel
Maybe you can propose effectively that we so there are two questions. I will pick the first and Jeremy will pick the second, if you're okay, sure. I think it's more just to clarify with respect to your paper and your presentation. So the first question is wonder, as your model is used to forecast unreported data, how can you assess its performance out of sample? So just to explain, in fact, how you compute the out of sample error or performance, if you can just develop on it?

Raul
Yes, the way it was used. And the figures that are reported in the paper are essentially based on typical cross-validation approaches where you just reserve a final bit of data for the evaluation at the very end. That's really how the R2 were calculated. And as I mentioned, what we switched from since the paper, we started doing it differently because we've been doing this for a while and we had indeed predictions which have been put together already over well, the paper just appeared in 2022, but is based on work done 2018 for emissions then. So, I mean, we could use the model to generate predictions for companies that in the meantime started reporting. And so in the second paper, that's really what we're looking at. And we believe that's really a lot more powerful because indeed, we had a model and now we can really look okay, predicted some figures the company published. So how good was our prediction? And that's, as I say, where we see that our model is generating the lowest errors in terms of comparing what was the prediction in the past and what we getting now for that particular date for a company that in the meantime started reporting. So we're putting a lot more focus on looking at those numbers than even at the numbers that we published in the paper that indeed show the quality is good, but it's based on a small sample that we keep 10% of the data set that we keep on the side just after doing all the construction of the model. Really just do the predictions for those 10% of companies and estimate R2. So we moved away from that, and we're looking really at errors based on reported data that was reported exposed.

Emmanuel
Thank you.

Jeremy
I think we missed a question from the gentleman, but it's in two parts. So the second question is, is there some types of assets or sectors which are more challenging to capture from data vendors? For instance, private equity or infrastructure? Also, how do you deal with missing data imputation before modeling?

Raul
Yes. So on the first question, indeed, on the universe that a lot of the companies that were missing, indeed, they corresponded to private assets in terms of missing data in the paper, we kept it relatively simple. Whenever there was a data point missing, look at the sets and the type of data sets. Also in the paper, we have the stats, but essentially we just pushed previous numbers that had already been reported, and that essentially we managed to complete the entire data sets. So in terms of what was missing, it was not really that much. And we managed to complete the data sets just by pushing four numbers that had been published in the past. It's an approach. There's potentially other approaches. When we did the paper, in fact, on the production model, I'm not sure if they did something more sophisticated, but when we constructed the paper, the model that is in the paper I'm talking about, that's essentially what we did.

Emmanuel
Okay, you pick the last one Jeremy?

Jeremy
Ok. Is carbon offsetting complicating the estimates or taking into account to some extent?

Raul
Carbon offsetting, no, as far as I know. I don't see why we to be honest, I'm not sure if we used it, tried it out as a variable I don't know if we have enough data to actually use it as a variable. We didn't I mean, would we get better data by inputting carbon offsetting? To be honest, no. I don't have an answer to that question now. Would it make a difference? Don't really know. I don't see why it would make a big difference. But what we can make is a company that is prepared to pay and buy and offset carbon in one way or the other and continue to emit, still have higher emissions for us, still be reported. I mean, here we're not saying that the company has these emissions and then is buying these carbon offsets. In fact, we don't take carbon offsetting in the carbonization efforts in the funds. So if the company is doing okay, it's excellent. But the number is still the number. So that's it.

Emmanuel
I have a question. Did you look so you already give an answer the private public. But beside that, did you look to the characteristics of the company that report versus the one that do not report? Are they different? Are they similar specific pigeons?

Raul
Well, the one feature that is really just obvious and comes out from the data very easily is that large cap companies publish more. I mean now I don't remember any other features really being common and that large cap companies published is also understandable. It's also general bias on ESG because they have means, they have more means to hire people and to have people focusing on this. As you become smaller and smaller, the costs of having people just calculating these things or doing this get higher. So for as long as you can avoid, if you don't think that that is going to have an immediate benefit on your business, you try to avoid. So that's the one bias and you can see it in the numbers that we showed right at the beginning. When you go to the 3000 larger companies, you already have almost 50% reporting and if you look at the entire universe, you can see that it's most smaller companies that not report.

Emmanuel
Okay, so I think we are close to the end. So just want to thank first Jeremy, thank you very much. Looking forward to your result Jeremy, that you will send to Raul. And obviously I want to thank Raul for sharing his expertise. So great talk. Thank you very much Raul. It's a pleasure. So I just want to take a moment to announce that we will have a next speaker series. This will be this month too. On the 18 of April we will continue our Journey on the Future of Finance and we will be happy to host Alex Edmans from LBS that will talk about ESG and more precisely do company, do corporation need a purpose? So thank you very much. Thank you Raul.

Raul
Welcome. Pleasure. See you. Bye.

Emmanuel
Thanks Raul. Thanks Jeremy.

Jeremy
Thanks.

Make an impact
EDHEC BUSINESS SCHOOL
[Music]

[Music]
EDHEC Business School
EDHEC Speaker Series - The future of Finance
Making sense of Green Finance
How a smarter financial sector can contribute to the net zero transition

Jean Boissinot
Deputy Director (Financial stability)
Banque de France

Tuesday, 4 February 2023
6 P.M - 7 P.M Paris time

Emmanuel Jurczenko
Okay, so let's go ahead. Hello everyone, and welcome to the second season of the EDHEC Virtual Speaker Series on the Future of Finance. I'm Emmanuel Jurczenko, director of the Graduate Finance Program. And again, it's my pleasure to welcome you today. So, for this today's session, first I want to introduce our two commentators. So we got David Zerbib, that is professor of Finance at EDHEC, one of our experts in sustainable investing. We have also the pleasure to welcome Alexanne Heurtier. Alexanne is a student in the MSC in Sustainable Finance and Climate Change that is conducted in partnership with Ecole des Mines. And it's my pleasure to introduce our guest speaker for today's session. So we are very fortunate to have Jean Boissinot. So Jean is deputy director at Banque de France. He's the head of the Secretary of the NGFS, so Network for Greening, the Finance System and before joining the Banque de France, Jean head various position at the direction Général du Trésor. He's an alumni of Ecole Polytechnique and the Paris School of Economics. And last but not least, Jean has published recently a book that I recommend titled Finance Verte, published from Dunod. And here is the book.

So, I read yesterday and really learned a lot of stuff. So I really recommend all of you, if you did not already read it, please. And we will have kind of overview of some element that he described in his book. So we are very happy to have him speak today and to help us to better understand what is about green finance, why finance matters for the zero carbon transition and does climate change matter for finance. So, without further ado, I'm turning to Jean. Welcome Jean, and the virtual floor is yours.

Jean Boissinot
Thank you so much, Emmanuel, for the very kind words. So let me pull up my slides. All right. Is it working? Perfect. Thank you very much. So, again, thank you for the invitation to speak about green finance. Indeed, my keynote of today is, I would say, loosely inspired by the book. There's a few more things in the book, but I wanted to have a discussion with you today about making sense of green finance because it is something that seems of use the idea of green finance that everybody is speaking about. But in fact, there is a lot of confusion, I think, around it. And just to illustrate the confusion or the diversity of the perspective that you can have about green finance, I played with a few logos of things that you can loosely relate to green finance. And I think you would have given a slide twice as giga. I could have filled up with some more things. So it's a bit of a disorganized field, I would say. And part of the specifics of green finance and the problem of green finance, I would argue, is that everyone is calling green finance something that he sees or she sees as green finance.

And there is very little consensus around the idea, which in the end means that some would argue that green finance is the solution and some would disagree and very much argue that it's pure illusion. And I think that this keeps the discussion in a place where we cannot see exactly what finance has to do with the transition. And I would blame part of this confusion, I think on the advocates for green finance because you can find good, bad and very ugly reasons to promote green finance. The good reason to me, and I will hope that you would be convinced at the end of the keynote that this is how we should be thinking about green finance. The good reason is that we need to finance the transition and to deliver a climate consistent capital allocation. The bad reason perhaps is an argument that relates to the fact that public finances have been stretched to the limits over the past ten to 15 years and especially in the past five years. So you find people arguing that because of this limit that we have reached with regard to public finance, maybe there is a role for private finance to pick up the slack.

And to me, what's a bit wrong with this idea is that in fact, as we are looking at a complete transformation of the economy and the financing of that transformation, maybe we can look at precedent. And in no places or in no previous past episodes, I could see public finance being the main driver of an industrial revolution. Maybe the ugly reason to me of promoting green finance is the idea that financial policies will make it up for the lack of ambition of climate policies. It is hard to put a price on carbon, it is hard to have proper climate policies and maybe we can let the financial sector do the heavy lifting. It has to do with the idea of advancing a value based agenda also. And I think that's something that is raising a lot of questions and raising pushback from many fronts. And overall, when we argue for green finance on the basis of not very good arguments, we run the risk of over promising basically what finance can do and under delivering in the end what finance is able to do. So before I jump to trying to substantiate...

Emmanuel
Jean, just a question is it possible just to increase the sound?

Jean
I'll try to keep the mic further. Sorry, apologies for that. So I'll speak less with my hands, which you didn't see. Sorry. So before jumping to the core of my demonstration, just a bit of an historical detour into how I get involved into green finance and I think what really happened around the COP 21 in Paris, which I think was fundamental for this development. So in 2011, 2012, we had a big discussion in France around the lack of progress in addressing climate change. It was just before the Grenelle conferences and the Grenelle laws and so on. And in fact, we see little progress in reality and finance was identified as the possible culprit for all that lack of progress. And the rational behind that was twofold, was mainly the fact that finance was seen as being in bed with the no transition team, the all majors and the like. And that's true that the financial system, the financial sector has a lot of connection with these type of companies because they've got a lot of friends and this is clearly a type of client that would deserve special attention by the financial actors. But I think that beyond that which we should not overestimate, there was also the idea that finance goes with the flow and we are only financing what the economy is doing.

So if the economy is not transitioning, well, it's a bit difficult for finance to transition ahead of the economy. And I think that's a more reasonable argument. But the real question behind that is not whether finance is responsible, but is or whether it's a good scapegoat for the lack of progress. But is it really the case that finance is holding back the transition? I mean, we all know that for the transition to happen we need typically climate policies being implemented and investment decisions by firms and by households and so on. So is really finance holding back the transition? And we came out of that debate in my team with a better understanding of maybe we have to ask ourselves is there something there that we should be looking at more closely beyond the kind of easy argument to put all the blame on Finance. And soon after that, in fact, we were asked to work on the COP 21. And the one key steer that we got from the Finance Minister for whom I was working at the time was please don't let finance get in the way of an agreement in Paris. And for that to happen, we had to go past the idea that finance matter for the transition which is obvious, you don't have a transition without investment and you don't have investment without finance. And that's true both in general in the economy, but also in the context of the COPs because of the commitment that advanced economies have taken toward less advanced economies. The 100 billion question typically. So there is no matter disputing that finance matter for the transition. But the real question is the other way around basically. What if climate change matter for Finance? Because if climate change matters for finance, then it's a whole new set of things that things that we can play with basically too to make some progress. And in answering that questions, basically you can identify three main drivers. The first one is related to the no transition team arguments and it's about financial institution protecting their reputation and it's a communication battle. I'm doing some nice things, but you're not doing anything else than wind-washing, et cetera, et cetera. And it can eat up quite a lot it makes for good headlines but very little progress on the ground. But that's one driver to get the attention of financial institution on the transition.

The second driver is more related to client demands and preference. And that's the idea that and some financial institution would argue about that at that time and even today, if my clients want green products, green saving products to finance green investment, I'm very happy to follow them. And if they want something that is less green, then so be it. And that's in relation, I think, to the idea that finance goes with the flow. The problem with that is that because we are not speaking of a set of preferences that are completely independent from one another and disconnected from reality. What you are doing when you are engaging into green finance only on the basis of your clients demand? You're basically building up a balance sheet that Alpha Fit will thrive in the transition and the other one will fail. And if there is a lack of transition, Alpha Fit will fail and the other one will strive. And I think it's not a very good management, good risk and business management to basically build up your activities and your company in such a way that in any state of the world you're losing half of what you have done. So I think the third and main driver for looking at what climate change means for finance is really about focusing on the economics of climate change and the net zero transition.

And that's what I would try to get around with you in the coming slides, having in mind that I will be a bit quicker than with an uneducated audience. So this is, I think, very common ground for all of you, the fact that we are witnessing a rapidly rising global temperature and that these anomalies we cannot explain them without factoring in human causes in the models that we have. So this has something to do with the GHG emission that we are producing. What is very important is also to keep in mind that we are witnessing the very first impact of climate change. But it is going to be a lot worse if we let it go and we don't address it. And in that respect, 1.5 degrees is not so good, but not as bad as two degrees and certainly not as worse, so to speak, as an increase of four degrees. So there is really something to be done to minimize the level of climate change we are confronted with. And because that level is entirely dependent on the concentration of CO2 emissions, that really means that we need to get to a net zero situation.
And that brings me to my next slide about carbon budget, which is what people don't completely realize. Net zero is not an option. The only question that we have to ask ourselves is when do we reach net zero because we have to get to net zero if we want to stabilize the climate, whether it is to 1.5, 1.7, 2 degrees or even four or five. And that gives us basically the time frame for the transition. Having in mind also that nothing in that is completely there well. Nothing is deterministic at all. It's really a shift in distributions of distribution. So we are basically making a bet on the future and trying to find the safest way to remain in a place that is comfortable enough. So some would argue that we are clearly not where we should be, and I would not disagree. But I would also like to highlight the fact that we are making some progress and that things are maybe starting to be a bit less underwhelming that they used to be. Not completely overwhelming, but less underwhelming, I would argue. So basically we have made a lot of progress if we compare to where we were before the COP 21 and if you look into and what matters is the graph at the bottom right of the slides.

Basically where we stand right now, we have with the current policies, a 7% chance to remain below two degrees. And that's not a very comfortable place, but if we manage to implement the NDCs, these seven goes to 16. And if we follow through with all the net-zero targets that have been adopted by countries over the past five years since Paris, basically, we've got more than 80% chance of staying below two degrees, and we've got a fairly good chance of staying below of 20% chance of staying below 1.5 degree, which is really what we should be aiming at. So clearly we are not where we want to be, but we are really making a lot of progress if we compare to where we were less than ten years ago. And part of this progress is, I would say, predicated on the fact that the economics of the transition have been changing a lot. We are witnessing an industrial revolution that is happening and it is happening not because of a lot of innovation being rolled out in the economy, but because a lot of things are happening due to a constraint basically. And this is somehow an industrial revolution in reverse motion or with a different driver.

But what is quite clear is that now we have the capacity to produce electricity from renewable at cost and soon at scale. And that's covering about 40% of all the emissions. And that has a lot to do with the increasing the scale of what we have been doing with regard to installed capacity and what we have learned along the way. And when you look at the latest IPCC report, basically you see that we have technical solutions for almost all problems related to GHG emission and we know what to do. I mean, maybe it's hard to implement these policies, but we know what to do. Moving from the microeconomics to the macroeconomics of the net zero transition. It's also very important to understand that beyond making sure that all these investments happen what really matter is not just the investment but also what we will do, but also what we will not do. And if you look at it from a macro perspective basically the additional investment that we need is in the vicinity of a few hundreds of billions of dollars every year. So something that the financial sector knows how to deliver. And basically if you look at the transition from this macro perspective the transition is a capital allocation question.

It's not about finding a lot of new resources and investing in difficult stuff. It's really about making sure that the capital allocation is delivered as it should. So what about green finance in that context? In that change of perspective basically I would argue that we should frame climate change as a macro development. That means two things basically bring nature back into the picture make sure that we really account for the material impact of climate change because it will have very real macro consequences and understand that there is no alternative to net zero. And the second aspect to it is really to understand the scale and the dynamics of the industrial revolution that we are facing. And again, if you combine the question of no alternative to net zero on the very limited time frame that we have and the fact that it's technically feasible and economically reasonable to transition our economies, basically we are in a place where it will be a macro question, not a micro question. So it's not something that will be a gradual adjustment. It's something that will, I hope, most probably happen in a limited time frame with big consequences on the macro side.

And that's the reason why a climate change is meaningful for finance. It's meaningful for finance because we need to grasp the full scale of the climate opportunity but also because finance is in the business of building bridges to help the economy develop. And it's bridges that goes from now into the future and we need to reassess basically the way we were thinking about this future and really factor in climate change and the transition into all our financial decisions. Which brings me to what I would say is a better way to look at green finance by distinguishing and to bring some order to the very diverse picture of the beginning is to distinguish between the visible phase or the visible side of green finance, financing, green investment. And it's a question of, well understanding the risk return of green projects. It's about managing technical risk, it's about technological risk, it's about taking into account the longer term perspective of these investments. It's about taking into account the system dimension of this investment and the fact that there is a lot of system externalities into these decisions. It's about recognizing that we are facing takeoff risk that we are dependent on policies when we are taking this investment decision and financing this investment and it has brought a lot of financial innovation think about green bonds, think about a few others, blended finance and whatever and it's about finding the right way to pair up the policy support and the financing basically. It's also honestly a question of how to make sure that we are providing as much green asset as the investor side wants. Because we are in a position of not having enough projects and assets to feed in the demand side of the market and if we are not able to do so, then we will have a greenwashing problem. And greenwashing is just basically the volume question, the volume version of a price bubble. It's an unsociable demand chasing after too few assets. But beyond this very visible face of green finance there is also I would say the hidden face of green finance which is really where finance matters. It's not about financing wind farms and PVs. I mean that's important but that's not the main contribution of finance to the transition. Our main contribution to the transition is to make sure that the capital allocation we deliver or the financial system is delivering is consistent with the climate change to keep. And this is something that really goes, I mean the first side of green finance is maybe a few hundreds of billions or a few thousands of billions of dollars every year. The dark side is really about every single investment decision to be reassessed against the perspective of the change of climate on one end and the transition on the other end. And basically you can rearrange everything that you can. Think of in the field of green finance under these two items and you can find that it is something that would speak to every bit of the financial system. So I think it's a better framework basically to think about what we are doing as financier in that respect rather than arguing that it's about thematic funds, it's about ESG investing or it's about green finance and blended finance et cetera, et cetera. I think it's good that we have a collective understanding of how each bits and pieces is fitting with one another and how this is contributing to the bigger transition. And maybe to conclude, just let me share two big questions that I think are important at this stage.

The first one is related to the way climate risks are priced in or not. In fact, I would argue that most of the policy decision that we are taking now whether it is a microprudential regulation, macro prudential policy, monetary policy operation and so on in fact we are making an implicit assumption around the fact that the climate risk are priced or not in the market. And whether it is the case or not we should be considering very different options. So it's a very important questionsn on the policy side I would say. On the private side, it's also quite clear that a lot of private sector practices are in fact also related to that assumption. And if we look carefully, and I wish there were more studies on that and more research on that, what we can see is that climate change is indeed increasingly priced in, especially over the past five to seven years. But it's unclear to which extent these developments are priced in. It's not quite clear that they are all priced in out of the right baseline or that we could basically infer some kind of market implied baseline with regard to the transition. So one big question is really to try to understand what the market is telling us and is pricing and not pricing because that's where a lot of good or bad decision or the basis on which a lot of good or bad decision will be made.

And finally there is one question that keeps us, that is haunting us basically once we have started to work on climate is there something else beyond climate? And I will argue that on the basis of the work that we have done in the NGFS, we are actually currently doing a lot of reading and research. I would argue that indeed it's not just climate change that matters. There is a lot of the more general problem is nature blind economic decision. It's the fact that we are completely forgetting or taking for granted a number of things that we use and that we should maybe question the continued availability or the maintenance of the quality of them in the future. And that's clearly something that we need to reassess basically. One line of arguments is to well question whether this is somehow all be summed up into climate and purely climate change is something that has a number of indirect impact on, but it's not just climate. And for example, if you think about freshwater, clearly climate change will have an impact on freshwater availability but a lot of the main problems are not related to climate.

They are related to the way we manage the current resources that we have. Something that is a bit tricky is that the risks that we are starting to look at, they're in fact a lot more local and granular than climate change. Climate change is in fact a global problem and something that we can rather easily understand. But what we are looking at is can be global in a number of cases but is also very often a lot more local and a lot more granular. And when I'm saying that, it does not mean that it is less material from a macro perspective, just that it is a lot more complex to understand and identify. And this complexity is in fact the problem that we are facing and we collectively need I think attractable framework to account for this more local and more granular phenomenon. I think and I would argue that the current way of thinking about physical and transition risk is the right way and can be generalized to other nature-related developments. Now, the question is everything that was easy to grasp and had to do with the ton of greenhouse gases emitted into the atmosphere, basically that summed up the problem, or what's the equivalent of that, or how to think about nature in general.

And what I'm showing you on the right-hand side of the slide is a personal, at this age, a personal thought around how to generalize the progress we have made with regard to climate change. Basically connecting the planetary boundaries, which are the things that we will have to collectively address and that would matter for our societies. To ecosystem services, which is the thing that we use without thinking about it and our economy is dependent on. And if you think about it and try to generalize what we have done on climate change, basically our dependencies on ecosystem services that are unpriced and overlooked so far can be related to physical risks. On one end and the impact of our economic activities onto the planetary boundaries basically has to do with the transition risk. I mean, that's because we will try to manage the system to bring them back within a manageable within manageable limits that we will raise a number of transition risk. And then the question if we want to generalize a bit or to use that framework for expanding what we have done on climate to other nature related risk. The two questions that we need to address is to know whether going from the planetary bondaries to the ecosystem services that matters is enough or if there are other development that we should take into account and also to understand which ecosystem services we are using that we are critically dependent on.

And I would argue that from a macro perspective, they are not all born equal, I would say. But the question to have this track table framework, the question is which are the ones that we should for sure include into our city? And that is about it. So, very happy to get your reactions, comments, questions and remarks.

Emmanuel
Thank you very much, Jean. So right on time. So, thanks. So now I let David and Alexanne ask some questions or reactions.

David Zerbib
Yeah, well, thank you so much, Jean, for this highly interesting and insightful presentation. You've clearly explained what green finance is today, what we should expect from green finance and what we should not expect today. I'd like to ask you a kind of slightly provocative question to go straight to the point. So we are in a world where the carbon tax is probably not ambitious enough, especially in some areas in the world. We have evidence that the preferences for investors to purchase green assets is not that huge. And at the same time, we know that we want to re embed finance within a nature ecosystem. If I may say within nature or constraints. So my question is to which extent green finance can be coercive? What in your opinion is acceptable and what is the limit of acceptability? Can we constrain investors not to invest in some asset or compel them to invest at least x percent of the balance sheet into green assets? Or from a central bank point of view, there are different avenues in terms of yeah. So I leave the floor open to you for the answer.

Jean
Thanks a lot, David. So perhaps it's the time that I should stress a disclaimer that was on the first slide. Everything that I'm speaking about or any views that I'm expressing tonight are clearly personal ones. It's a personal answer that I would give to your question. I fully agree that we don't have the climate policies and the environmental policies that we need at this stage. And that's a problem. That's very much a problem because in fact if we want to manage the transition collectively we need all actors to play their part. We need the policymakers to set up and implement the right policies. Right policies means a lot of things. It means ambitious enough, it means sustainable policies, policies that can be sustained over time. It means policies that are politically acceptable. So it means also just policies and so on. So I think it's important that we have these policies put in place. It's also important because without these policies basically none of the decision that are taken by the economic actors will be fully aligned with the objective of re-embedding economic activity into nature. So we need these policies, then we need the investment of firms and households to adjust to this new set of policies and then we need the financial sector to understand the implication of the policies so that basically it is able to discriminate between the good project and the bad project early on, but also to be there to accompany the transition.

In fact, if I wanted to summarize my view on the transition is that it's a tango that needs three bits to be danced, basically. So going to your question about coercivity basically no. Before going to that coercivity question your comment on the preference for green assets that you assess not to be very material or not to be important enough. I think there is a lot of appetite for green assets and the problem is perhaps more, I would argue more on the supply side of producing these green assets than on the demand side. But the problem is that what finance is able to do is always limited by the risk-return of the project basically. You cannot ask a normal financial institution to finance a project that is not bankable especially if you're looking at financial intermediaries who have to respond to the two sides of their balance sheets basically. It's impossible for them to be making that kind of arbitrage of, you know, giving precedence to the asset side. Or to the liability side. So they need to find the right place. And that's what the policies are there for, basically reconciling the two parts of the equation and making sure that the project that should be bankable are bankable and that the project that should not be bankable are not bankable.

So that's where I see the issue. And then to your coercivity question, to me it's really a question of what we think is the social value of the financial sector. Is it to bring fund to one place to another or is it to allocate capital? If we think that the main value of the financial sector is just mechanically to move some resources from one end to the other in a very automatic way, I would say, maybe we can go for some coercivity. But I mean, that would require a lot of planning, a lot of information that I'm not sure anyone have in the system. If we think that the social value of the financial system is really to deliver capital allocation that is informed by a lot of individual decisions, a lot of tattooment, a lot of bets and so on. But that in the end goes into the right direction, then I would fear to interfere too much with that process by being coercive and by saying, okay, now what you need to finance is x percent of your balance sheet should be devoted to Green. It's a question that I'm faced with regularly by investors saying if someone can tell me what is green, I'm ready to finance.

And my answer to that is well, no one will tell you what is meaningful to finance. What really matters is for you to play your role as an investor and make sure that you are putting your money where it should be put in light of basically the future as you sit and as an informed view of the future as you can.

David
Thank you very much Jean

Emmanuel
Perhaps just to jump to that the challenge with respect to this capital allocation is obviously how you assess the performance and the risk, right? And we are more in an uncertainty setting than in the risk setting. So I just wonder what we need to do in terms of progress toward the direction because you need to have some information, you need to have some tools in term of asset allocation. So what is your vision about where we stand today and what we need to progress toward this very important mission, which is the capital allocation to have the right one with respect to the climate risk?

Jean
In a nutshell, I think we have started to move in the right direction and we are clearly not where we should be yet and we should not have a sequence approach to the problem because we don't have the time to have this sequence approach to the problem. I mean, develop the perfect data and develop the perfect models and so on and when we have perfect everything, then act upon those. We are born to start experimenting and to innovate along the way. But I'm really mindful about the urgency of what needs to happen in the grand scheme of things. I'm also quite impressed by the strength of the momentum in the financial sector when things are done in the right way. Typically in the NGFS we are a network of central bank supervisors. We have joined forces to develop the tools that we needed to do what we think we have to do with regard to climate change. So improve supervision, make sure that we are able to monitor climate-related risks at the macro level, take into account climate change in monetary policy and so on and so forth. So it's really central banks trying to develop the tools that they need because they have, as everyone realized, that they know way too little about what to do, but we need to go fast.

I'm really impressed by the fact that five years ago it was some kind of visionary but completely out of the monet by eight institutions who said together well, we need to collaborate to find the solutions. And we are now 140 institutions. We are collectively supervising all of the global systemically important banks, for example. So this is a group that is not marginal by any way you look at it, from advanced economies, from emerging markets, from developing economies and the work is progressing steadily and really faster than everyone expected. I mean, five years ago you would have told me that in five years we would have climate stress testing for real. I would have be very happy. But I would have perhaps be a bit cautious about our collective ability to deliver. And that's what has happened. Also in the financial system this is happening. You may remember that a bit more than two years ago the Single Supervisory Mechanism, the supervisory arm of the ECB, has set up supervisory expectation with regard to banks for what they have to do or for what they should be able to do on climate.

And they took an initial picture of the practices in the banking sector and basically we had 25% of the banks at the time had at least a basic understanding of the risk. So clearly we were starting from nowhere. Two years later they took the same picture and this 25 is now 85. So a lot of progress in the past two years. What the ECB has also assessed is that basically all of the banks have material gaps. So not everything is perfect a long way to that situation. But still they have been able to identify best practices for almost all of their supervisory expectations. So it's possible to go where the supervisors think you should be operating to be able to deliver this climate consistent capital allocation. Which means, basically having the right data. Even if it's not perfect, you start measuring them. Having the right credit decision process, having the right sectorial policies, having the right management of your continuous management of your risk and so on and so forth. So it's a work in progress that we have to accelerate. But clearly, when we decide to go into that direction, it's something that we can deliver.

Emmanuel
Thank you, Jean. So, Alexanne.

Alexanne Heurtier
Thank you again for your presentation and for your book. Very insightful. I may have a more macroeconomic question for you regarding a subject you mentioned. So you're talking about the heterogeneity of the economies, but also of responsibilities. You also talked about the commitment of $1 billion for the developing economies, but the targets is still not met. What do you think about the loss and damage fund that was announced at the COP 27? And what maybe a second question. What could be the role of private institutions to maybe be complementary to the public targets they have?

Jean
So maybe two ways of looking at your question. I think it's important that we collectively keep in mind this common but differentiated responsibility that we collectively have for climate change. And also keep in mind that it's a bit more complex than just advanced and emerging and developing economies as of now. So we need to also update the way we were thinking about this $100 billion pledge. But still, it's indeed a very important question because the transition is not something that will happen if advanced economies only look after what has to happen in advanced economies and so on and so forth. I mean, the transition is more than a collection of domestic transition. There is a global dimension to that, and there is a lot of capital flows to be scaled up to finance investments in places that lack the resources to fund this investment. So it remains a very important question how to scale up these capital flows. And to be completely honest, that's a question we are currently in the NGFS, but a few others among the multilateral development banks and others are looking at how to make it real that these capital flows happen, not because we have pledged they should happen, but because it's a necessity that they happen for the transition to happen.

And it's both feasible and manageable. So on the loss and damages fund that was set up in Charm el-Cheikh, I think it was a longer virtue I mean, that's again, a personal comment. It was a longer virtue decision. But I fear as someone with a finance mind, that we are missing half of the decision basically. It's as if we have provided an insurance against the worst of climate change, but we have not kept the risk that climate change will keep on worsening and worsening and worsening. So I think it's something that is begging for a strengthening of the commitment of everyone to transition away from fossil fuels, basically. And if we can have these two bits, then I think we can start to think a lot more realistically about what needs to happen and what can happen with regard to cushioning the poorest economies and the less developed economies from this risk. And I mean, the case in mind is what has happened in Pakistan over the last year with a dramatic heat wave in spring and early summer moving into a catastrophic "mousson" in September where literally 12% of the Pakistani's land that was underwater.

The shock to the Pakistani's economy is 10% of Pakistani's GDP. So it's something that is really I mean, we need to find ways to make sure that we know how to usher and rebuild, build back after these type of shocks. But it's meaningless to try to build back after these type of shocks. If we don't do two things first, limit the risk, which means that transition accelerate the transition and second, build up resilience and adaptation when we rebuild after the shock. And in fact, we don't need to wait for the shock to embed adaptation. Which I think brings me to trying to answer to your last bit, which is what's role of the private sector for loss and energies? And I would here to say that to me loss and damages is not a private finance question, it's a public finance question because I don't see the business model on which you can build up inflows and outflows early mobilization of resources and then payment back to the investor. In that context, maybe I'm wrong, but I think it's something that is a bit tricky but something that I think is very important is embedding adaptation into the mindset of financial institution and especially enhance the way insurance is contributing to solving the problem.

Basically, if we could reveal the value of ex-on-play adaptation to limit the cost and well, the risk and the cost of climate change, I think there we could find some kind of business model which the private sector could play a role in but when it is to only indemnify for the losses and the loss and damages I fear that it's really difficult for anyone who is not ready to lose the capital that he or she has invested to be killed in that field.

Alexanne
Thank you very much. Maybe Mr. Zerbib, you have another question?

David
I was wondering whether we could leave the floor to the audience because we have only six minutes remaining.

Emmanuel
Yeah so, might be able to pick several questions. So I will take one coming from Shaia. Do you anticipate debt and equity capital to play fundamentally different roles in green finance?

Jean
I would argue that in fact, so typically green bonds to me, if you understand it well, is a great instrument because it's an instrument that is bringing some practices of the equity market into the bond market. Basically, to me it's an engagement device for debt investors. So I think obviously we will not mix or we should not be confused about the role of debt and equity on both sides. But I think the two of them have something to play and the two of them may be learning from the book of the other in that respect.

Emmanuel
Thank you. Another question which is related to, let's say mine. Considering the lack of historical data and the assumption that markets currently don't correctly price climate risk, how can models that seek to price climate risk be validated? Basically how we can be confident that these models are working. So, question of model risk?

Jean
Yeah, that's a very good one. So I think that first it's good to make sure that when we say model, we can mean very different things. And when you are looking at economic models, you can play with a lot of exogenous that can tell you stories about the impact of climate change. So in fact, what you need is to bring together the type of models that you are using more in the central world with the type of models that comes more from the academic side of the conversation. And in fact, that's part of what we are doing in the NGFS, in building the NGFS climate scenarios. But yes, it's important to have in mind that some models can speak about something that we have not seen in the past because they are built in such a way that you can play with the main drivers that you see for the future. And we need to have a better dialogue between these models, I think, and the financial models which very often are built on the idea that well, the past is some form of equilibrium that will keep on repeating itself, which we know is completely untrue.

Maybe back to the question I think we also need t,o no, I won't be answering right now. I think something else we need to do is also stop thinking in terms of in a too probabilistic way or trying to make prediction. We need to basically stop optimizing and thinking more in terms of obviousness, basically. So maybe we have models that are not very good, but because we know that they are not very good, we should think about using them not to optimize or to maximize something, to maximize a return, but very much with regard to minimizing the risk that we are making the wrong decision. So we need to perhaps change a bit the way we think about models and the way we use them. That being said, I think it's not completely well, it's not too much to ask to basically tweak a discounted cash flow model, for example, to start, see whether the model that we see, the price that we see are in line with what we know about the rise or the future damages of climate change, for example, or the rise in carbon prices and carbon prices and so on.

So it's not something that we can say, well, we know for sure that it's price or not price. It's more like to what extent can infer something about these risks that we know for sure to be rising are or not reflective in the mean? I hope it was clear enough. I'm sure it wasn't, but I hope it was.

Emmanuel
Okay, thank you. Perhaps last question for today from Laurent. What are your views on the fact that the political momentum and rules seem to diverge between Europe, US and China? Kind of geopolitical question?

Jean
Yeah, I think it's a good question. I hope that in fact, the economics of the transition are starting to be too powerful to be ignored. And I think we can have reached that stage so that cooperation between the big blocks can in fact be reestablished because we have a common economic interest into moving toward the transition and we have also, to be completely honest, a common interest in avoiding the worst of climate change. So the current geopolitical situation is not very conducive of an accelerated transition, but it might still be the case that we can manage it and we can manage the adverse consequences of the Splack of cooperation between the block.

Emmanuel
Okay, thank you very much. I don't know if David or Alexanne want to add something before we close this session.

David
Just to say thank you again for this lighting presentation.

Emmanuel
Okay, so thanks to all. Thank you again, Jean. Very great presentation. For the one that wants to go a bit further, read the book before we close down, just take a moment to announce that there will be a next speaker series that will be on the Tuesday 20 March, will be a bit related. We'll have Alex Edmans from LBS that will talk about do corporation get a purpose. So, thanks. And Jean, if you want to conclude.

Jean
Yes, maybe a final word. What I am very convinced of is that in fact, every bit of the financial sector has to play its part to the transition. And the confusion that I was speaking of at the very beginning is in fact something that we should be leveraging off because we need people in banks, we need people in big asset managers, in private equity firms, in institutional investors, pension funds, insurance companies and the like. We need all of these players to in fact contribute to the transition. So there is not one single solution. But what we really all need to see happening is this kind of change from the very bottom to start to realign what we are doing with our common future.

Emmanuel
Okay, thank you very much, Jean. And from our side, we are preparing the future generation of finance. So happy to contribute at our level. Thank you very much.

David
Thanks a lot.

Emmanuel
Thanks again, Jean, for taking the time and looking for one.

Make an impact
EDHEC BUSINESS SCHOOL
[Music]