[Everyone’s AI] Can we evaluate a company’s ESG? ESG AI & Rating

In this article, we will talk about “ESG” which is an acronym that stands for Environmental, Social and Governance(in finance and or business context) and has recently become a hot keyword. We will also take a look at the MSCI ESG Rating, which calculates the ESG rating of companies, as well as the Ainize demo that calculates the ESG score of companies using Node2Vec posted by Hannahawalsh.

If you would like to check out the project right away, please refer to the link below!

Recently, the term “ESG” has been appearing everywhere. Many companies invest a lot of money for ESG management, and the amount of investment in stocks and funds related to ESG has increased rapidly. According to Forbes, ESG investing is “a strategy for investing in companies that are trying to make the world a better place.” So what does ESG have to do with making the world a better place?

Source : ESG investing gaining traction in Korea

What is ESG?

Like we mentioned in lines above, ESG stands for Environmental, Social and Governance. But, why do companies need to put these three elements together and associate them with management? Let’s take a look at how Forbes defined each element in ESG.

First, Environment refers to"How a business affects the environment". This includes the carbon footprint of manufacturing, toxic chemicals, amongst other environmental implications.

Then, Social refers to “How a company can contribute to society.” This includes race and gender equality, amongst other social implications.

Finally, Governance refers to “How a company’s management drives positive change.” This include executive salaries, leadership diversity, compensation to shareholders, amongst other company governance implications.

To summarize the above, ESG is what a company needs in order to achieve “sustainable” growth.

So, do you know how companies can actively engage in ESG activities? If you can calculate the ESG scores of companies, you can tell how much ESG management activities they are doing.

MSCI, a subsidiary of Morgan Stanley, one of the world’s largest investment firms, conducts its own ESG ratings of companies and funds. Let’s take a look at the methods used by MSCI to evaluate ESG ratings.

MSCI ESG Ratings

MSCI first decided on major subjects related to each field in ESG (E, S, and G) and then set up Key Issues for each subject. For example, the main subject of E (Environment) is climate change, and the Key Issues belonging to this main subject includes Carbon Emissions and Product Carbon Footprint.

When a total of 37 Key Issues are selected in this way, a score for each Key Issue is calculated. In order to be able to do this, the score of Risk and Opportunity present in each Key Issue is calculated.

First, the Risk score is calculated by synthesizing the degree to which the company is exposed to risk and the management score for the corresponding risk. In other words, a high risk score can be obtained only when appropriate management is performed according to the degree of exposure to risk. The picture below shows the graph where the Risk score is calculated for KI (Key Issue).


Now, the Opportunity score, just like the Risk score, is calculated by combining both exposure and management scores. In other words, when an opportunity is exposed, a high opportunity score can be obtained by managing it appropriately to take advantage of it and make a profit. The picture below shows the graph where the opportunity score is calculated.


Finally, the ESG score of a company or fund is obtained as a weighted average of the Risk and Opportunity score for each key issue, then, based on the obtained score, one of seven grades (CCC, B, BB, BBB, A, AA, or AAA) is obtained. In MSCI ESG Ratings, Natural Language Processing is used for data collection, analysis and evaluation of predictive results, not for direct ESG rating evaluation.

Not only MSCI, but also global investment companies and asset managers such as Softbank and Goldman Sachs are calculating their own ESG ratings. However, since the standards and calculation methods are still different for each company that calculates the ESG rating, the ESG rating of a company may be calculated differently. In fact, the correlation between the widely used MSCI and Sustainalytics’ (another rating company) ESG rating is only 0.53. Therefore, direct human ESG calculation may be inaccurate or biased, so AI, especially NLP, is actively trying to calculate ESG score. One remarkable example is TrueValue Labs, which was acquired by FactSet, an American financial software company.

Demo with Ainize

This time, let’s take a look at the ESG analysis project deployed in Ainize. hannahawalsh created a program to analyze ESG scores and distributions of companies using Node2Vec.

Node2Vec is a method of vectorizing nodes on a graph, similar to Word2Vec, which finds the embedding vector of a word. Therefore, in Word2Vec, learning was performed using sequences of words, while in Node2Vec, learning was carried out using sequences between nodes in graphs. A sequence of nodes is created using the DFS or BFS method. In the picture below, the first figure shows the label created when using DFS and the second figure is using BFS. With DFS, homophily is reflected, and with BFS, structural equivalence is reflected.

Source : node2vec: Scalable Feature Learning for Networks

Therefore, ESG AI uses Node2Vec to vectorize the relationship with the companies mentioned in the article and express it in coordinates. You can try the demo in this link.

So far, we’ve covered ESG and how you can measure a company’s ESG activity. As time goes by, it is expected that the size of companies investing in ESG management and the size of individuals making ESG investments will increase. If AI can calculate ESG ratings, it seems likely that it will be able to calculate ESG ratings for more companies in the nearby future. If that happens, I think more companies will invest in ESG management. What are your opinions? Share with others in the comments!


1 Like