Investigating the Transition Dimension in Climate Investing with Natural Language Processing¶

Financing the Transition: the Hidden Dimension in Paris-Aligned Benchmarks¶

Investors mainly focus on portfolio decarbonization when speaking about climate investing

Net zero investment portfolio more complex, because two goals:

1. Decarbonizing the portfolio
2. Financing the transition

PAB popular methodology in climate investing: falls short in incentivising the transition to a green economy:

1. Under-weights high impact sectors (Energy, Materials, Utilites)
2. Overweights Information Technology, Financials, Real Estate

Clear decoupling between the real economy and PABs

A portfolio contributing to transition will overweight green technologies

Our Approach: a Text-Based Transition Investing Methodology¶

We develop a firm-level exposure to green activities measure with text mining

We consider the construction of a transition portfolio based on benchmark optimization

A Green Taxonomy¶

The EU Taxonomy¶

EU Commission has introduced the EU Taxonomy for sustainable economic activities

Activities are defined as green if they provide a contribution to a least one environmental objective

The main drawback of this initiative is that underlying data simply doesn't exist at the issuer level yet

Green, Brown or Mixed Taxonomy?¶

Differentiating exposure to green, brown and neutral activities can be helpful for investors

Green taxonomy: identify more striclty green activities

Brown taxonomy: promote exclusion strategies from brown activities

Mixed taxonomy would combined both green and brown taxonomies, with neutral activities identified as the remaining ones

Our Approach: A Mixed Taxonomy¶

Green activities from the EU taxonomy (we combine both green and enabling technologies, but exclude transitional activities)

Our approach stops at the first stage: the eligibility

We use the activites in oil, gas and coal as a stringent definition of brown activities

Examples of green technologies:

• Electricity generation using solar photovoltaic technology
• Production of heat/cool from bioenergy
• Underground permanent geological storage of CO2

Examples of brown technologies:

• Oil & Gas Exploration & Production
• Oil & Gas Storage & Transportation

Using Business Description for a Firm-Level Estimate of Greeness¶

Embeddings with a Sentence Transformer Model¶

We first need to transform the business description and activities descriptiom from our taxonomy from unstructured data (text) into a numerical representation

We use a Sentence Transformer model to create numerical vector representation of the meanings of the business description

Cosine Similarity¶

We now have numerical vector representations of the business description and activities description

We can apply principles from semantic search by determining the closeness of our two vectors with cosine similarity

The closer the cosine similarity to 1 is, the more related the descriptions are

Greeness Measure¶

Finally, we adopt the following rule to attribute a greeness score to each issuer:

1. If cosine similarity less or equal to 0.5, a 0 score is attributed to the issuer.

2. If the activity for which the cosine similarity with the business description was the highest is among the brown activities a negative score is attributed. Otherwise the score is the cosine similarity.

Transition Portfolio¶

Transition Investing Objectives¶

1. Minimizing the tracking error volatility risk and
2. Improving the greeness of the portfolio version compared to the benchmark

Conclusion¶

Net zero investing involves two objectives: portfolio's decarbonization and transition to green activities

Paris-Aligned Benchmarks: falls short in addressing the second objective

Firm-level measure of greeness with a Natural Language Processing approach in order to build a transition portfolio, following a mixed taxonomy

Transition dimension can be addressed with the use of text-mining

Transition dimension can be integrated into the construction of net zero portfolio, targeting both the decarbonization and the transition dimensions