BNP Paribas AM turns to machine learning for carbon emissions

BNP Paribas Asset Management is using machine learning to estimate carbon emissions for companies that do not report their carbon footprint.

Raul Leote de Carvalho, deputy head of the quant research group at BNP Paribas Asset Management, says its modeling of carbon emissions will provide estimates for some 10,000 companies. The model’s approach was inspired by a paper authored by researchers at the University of Otago in New Zealand, detailing how machine learning can be used to improve the prediction of corporate carbon emissions for risk analysis by investors.

“We have been consulting with them and discussing this approach for a couple of years during our research phase,” Leote de Carvalho says. 

To model carbon emissions, the fund is using different types of machine learning, such as elastic net, XGBoost, and random forests.  

“The model is built by finding the factors that can predict the emissions of companies that [have] already published data—this is our training dataset—which are then used to come up with carbon emissions estimates for the companies that currently do not publish their carbon emissions,” he says.

The models use factors such as the scale of operations of the company, which can predict emissions. For example, if the factor was company size and there were two oil companies with one producing significantly more oil, that company is likely to emit more carbon, even if it isn’t reported. 

The gaps in the data, as well as the lack of a standard approach to reporting carbon emissions data, make some in the industry cautious about the potential of machine learning at this stage. 

Axel Pierron, associate director of client relations at Sustainalytics, says the full benefit of machine learning and artificial intelligence (AI) will be more important once there is a full level of exposure of carbon emissions by companies.

He says a machine learning approach is highly dependent on the quality of the underlying data. “That is where I am still a bit hesitant. I think it’s a very useful tool—I already see its usage—but I think we still have that issue of data quality,” he says.

He stresses the need for a human to qualitatively assess the data. “That is why there is so much demand for people [with] any ESG expertise and competency, because there is that need to re-work on the data,” he says. 

Indirect carbon emissions 

Some types of data are harder for machines to assess than others. BNP Paribas’ model will first be used to estimate Scope 1 and 2 emissions, with work also in the pipeline to make estimates on Scope 3.  

Scope 1 emissions cover direct emissions by a company from owned or controlled sources, such as company facilities or vehicles. Scope 2 relates to indirect emissions from the generation of purchased electricity, steam, heating, and cooling consumed by the company.

Scope 3 emissions cover all other indirect emissions in a company’s value chain, including upstream and downstream activities. Leote de Carvalho says Scope 3 can even be the most important scope for some sectors. One example is the auto industry, as a sizable proportion of the emissions from makers of internal combustion engine vehicles come from consumer use of their products.

“Scope 3 is the most complex and difficult to estimate because of the interdependencies it implies. We are unaware of anyone in the industry currently accounting for Scope 3 because it is difficult to estimate, and good estimations are not yet available—as far as we know,” says Leote de Carvalho. 

But he says BNP Paribas is working on it. “We plan to have a different version of the methodology for the estimation of Scope 3 emissions, which will probably also rely on supply chain data at least in a second stage of the modeling,” he says. 

He says the value chain nature of Scope 3 means BNP Paribas has a greater chance of creating a better model for Scope 3 for a given company by taking into account its supply chain, at least for some industries.

Mixing in supply chain data 

Its supply chain data about companies comes from two sources: Exiobase, for estimating emissions and resource extractions by industry, and Bloomberg’s supply chain data. It is focusing mainly on ESG and sustainable investing, with natural resources being one of the first areas of focus.  

“We are using these databases to see not just the companies in which we invest, but through the supply chain, what our current position really is. And we are working on this project at the moment, and we plan to have something in 2021. That is something we are working on, and something that most likely will be used in the fund we plan to launch this year based on ecosystems,” says Leote de Carvalho.  

He says BNP Paribas plans to launch a fund based on natural resources, with an aim to invest in companies that are doing the best in terms of reducing resource consumption and minimizing waste. “We plan to use the data we are currently calculating to better estimate the exposures of the companies to water consumption and forest consumption, and the way they use it,” he says.

Graph databases are one way BNP Paribas analyzes its supply-chain data, as they can provide a deep dive into a company’s different relationships. “Scope 3 might actually be the first example where we combine machine learning and the use of graph databases,” Leote de Carvalho says, adding that the machine learning methods that will be used for Scope 3 will probably be the same as for Scopes 1 and 2. The difference is in predicting the variables to be used. For Scope 3, some variables will be related to the emissions from companies in a supply chain, which will also likely be industry-dependent.

“There is the potential for double counting,” he says. “In the case of automakers, Scope 3 emissions of a maker of internal combustion engine vehicles will also be counted as Scope 1 for the company that bought or leased those cars for their own business. This ensures that companies feel responsible for their emissions across their entire supply chain.”

Pierron from Sustainalytics says when dealing with estimating Scope 3 emissions, the important consideration is consistency so there can be a fair comparison between participants. He warns that when trying to measure Scope 3, companies could have different reporting parameters within the same industry at different times. For example, one company might not report its carbon footprint outside of its home market, while another company does.

“We often need to have a quality assurance process, where you actually have analysts who really know the industry and they look at the data by geography to see if the data is relevant. Because sometimes what we find is that you will have companies that are under-reporting, or not setting the right parameter of reporting,” he says.

Sustainalytics also provides Scope 3 data, and, Pierron says, there are also other vendors in this space. However, he says understanding emissions for Scope 3 is a “very challenging exercise,” and a number of corporations now understand that ESG data is becoming more important within their investment strategy. “Therefore, some of them may tend to minimize their level of emissions,” he says.