Approach

The demand for sustainable and responsible investments has been growing rapidly over the last years, as has the demand for Environment, Social and Governance (ESG) ratings. ESG ratings describe how companies’ practices comply with sustainability standards such as the UN Global Compact or the Global Reporting Initiative (GRI). They allow for the management of risks and the selection of best-in-class companies. To assess companies’ ESG practices, Covalence uses a set of 50 criteria inspired by the GRI’s sustainability reporting guidelines.

Increasingly, investors want to know more than just companies’ practices and their level of compliance with ESG criteria. They also want to know about the impact companies have on society and the environment. For the Global Impact Investing Network (GIIN), impact investments are investments made with the intention of generating measurable social and environmental impact alongside a financial return.

While impact investing has historically focused on small, private companies, this approach is increasingly being adopted at the level of large, publicly listed companies. This is in part due to the influence of the Sustainable Development Goals, which call upon the private sector to develop products and services that contribute to solving the world’s major challenges. Mapping companies to the SDGs provides insightful material for impact analysis and supports thematic investment strategies.

An increasing number of companies publish ESG indicators. These indicators are communicated in absolute numbers (e.g. CO2 emissions in tons), in ratios (e.g. % of women on the Board) or in Boolean terms (e.g. existence of a Health and Safety policy in the supply chain: yes / no). Since 2016, companies have also started disclosing indicators relevant to the Sustainable Development Goals (SDGs) to reflect their positive impacts on society and the environment. Covalence acquires this data from external providers and integrates it into its proprietary ranking system.

Self-reported ESG and SDG indicators provide useful information on policies, processes and commitments. They respond to increasing demands for more transparency in the way a company conducts its business. However, the disclosed data is not sufficient to produce a balanced assessment. It is usually highly aggregated, mainly reporting global performance while providing little insight on local practices. It can also be positively biased, celebrating achievements and minimizing problems. There are data gaps. The use of additional data from third-party sources is therefore needed to document the perception of stakeholders and shed light on local situations.

Stakeholders such as NGOs, governments, trade unions and the media describe the role and activities of companies in positive and negative terms generating either endorsements or controversies. Since 2001, Covalence has specialized in the semi-automated analysis of such narrative content. This expertise materialized in the award-winning EthicalQuote reputation index.

We use data collection and classification tools relying on artificial intelligence techniques (machine learning, natural language processing) in order to analyse the narrative content. This process is reinforced by human interventions to classify the content in terms of polarity (positive/negative) and criteria. Our team of analysts thoroughly checks entries proposed by the software, thus ensuring high curation standards. Only sources that are publicly identified and available online are considered.

Today, the Covalence database includes more than one million documents from over 50’000 different sources on 6000 companies that have been classified and curated by more than 600 analysts in collaboration with over 30 universities.

The database leverages the use of machine learning techniques thanks to the expertise of our Scientific Advisor Prof. Patrick Ruch, field expert and professor at the University of Applied Sciences and Arts Western Switzerland. The use of classification algorithms allows us to fully automate the collection and pre-classification of information including complex information such as polarity – or sentiment – as well as multiple criteria.