Environmental, Social, and Corporate Governance (ESG) criteria allow to evaluate business's overall awareness and attention to social and environmental aspects. They are fundamental tools to direct investments towards sustainable projects and activities. However, each ESG rating agency has its own definitions, criteria and methodologies for the selection of the sustainable business portfolio, each not directly comparable with the other, which includes a set of indicators capable to measure sustainability. A shared understanding of what “sustainability” means is compelling. As a contribution in this direction, in the present paper we report ongoing work on the definition of a Knowledge Graph of some of the ESG indicators existing in the Literature. Links in the Knowledge Graph express semantically rich information including mathematical relations among indicators. We discuss the advantages for sharing, communication and comparison of this approach as well as open research issues.
Sustainability and resilience are central to the European political agenda. The Next Generation
EU financial instrument places issues such as equitable climate and digital transitions and the
ifght against climate change at the centre of the European recovery strategy. Among recent
research topics, that of Industry 5.0 is guided by the vision by which “in order to remain the
engine of prosperity, industry must lead the digital and green transitions” fostering “a vision of
industry that aims beyond eficiency and productivity as the sole goals, and reinforces the role
and the contribution of industry to society” in the belief that “Industries can play an active role in
providing solutions to challenges society including the preservation of resources, climate change
and social stability”1. Public funds cannot by themselves support an economic revolution that is
estimated in about 180 billion euros per year2, and the role of finance is crucial to direct economic
activities towards, and sustain the financial risk of, a green transition. However a shared notion
of what it means to be “sustainable” and how to measure “sustainability” is necessary. Quoting
from EU: “In order to meet the EU’s climate and energy targets for 2030 and reach the objectives
of the European green deal, it is vital that we direct investments towards sustainable projects and
activities. [...] To achieve this, a common language and a clear definition of what is sustainable
is needed. This is why the action plan on financing sustainable growth called for the creation
of a common classification system for sustainable economic activities, or an EU taxonomy3”.
The taxonomy was to come into efect in December 2021, but at present the process is still
ongoing, slowed down by the lack of agreement among the diferent stakeholders. In the quest
of shared criteria, Environmental, Social, and Corporate Governance (ESG) indicators have been
introduced by organizations like the German Investment Professional Association (DVFA)4 as
early as 2008, to quantitatively measure sustainability. They received the endorsement of the
European Federation of Financial Analysts Societies (EFFAS)5, thus gaining the status of an
oficial EFFAS Standard [
In the present paper, we report on an ongoing work contributing to a common understanding
of “sustainability” and integration of ESG indicators into a unique Knowledge Graph that can act
as a semantic substrate. Links in the Knowledge Graph express semantically rich information
including mathematical relations among indicators, that can be exploited to support their
sharing, communication and comparison. In Section 2 we report relevant related work on
semantic models for indicators and frameworks for ESG indicators. We describe the preliminary
survey and analysis of ESG standards [
Semantic modeling of indicators through shared vocabularies or ontologies emerged in recent years as an efective way to define measures through a precise, unambiguous representation.
A variety of approaches for semantic representations of indicators have been proposed in
the Literature, for documentation of indicators and their sharing, or with the goal to simplify
and support design and analysis of multidimensional data cubes or statistical datasets. In some
cases, indicators are directly used to design a monitoring framework like in [
3https://ec.europa.eu/info/business-economy-euro/banking-and-finance/sustainable-finance/eu-taxonomysustainable-activities_en 4https://www.dvfa.de/dvfa.html 5https://efas.com 6https://globalreporting.org 7https://www.sasb.org 8https://www.fsb-tcfd.org 9https://www.cdsb.net/
Most work focuses on definition of atomic indicators, with little or no representation of their
calculation formulas. Some exceptions include a formula representation with limited support to
manipulation, that is mostly performed through ad-hoc software modules [
Few contributions include a formal notion of dependency among indicators, e.g. in [
For what concerns semantic-based formats and standards for representation of indicators,
the Statistical Core Vocabulary (SCOVO) was one of the first proposals for the publication of
statistical datasets on the web as RDF. It included a minimal set of classes that however could
not fully support multidimensional modeling. Such a vocabulary was then superseded in 2013
by the Data Cube vocabulary (QB), a W3C Recommendation allowing to publish statistical data
following the Linked Data principles, and capable to model the schema of a cube as a set of
dimensions, attributes, and measures. QB enables the definition and the sharing of a statistical
dataset as a multidimensional cube, but a number of issues remained open, in that it does not
provide means to represent hierarchical relations between dimension levels and dimension
members, and lacks a controlled vocabulary for measures. Some proposals elaborated on this
standard to include further functionalities, e.g. to support OLAP operations on QB (e.g. [
We report also on recent proposals for semantic frameworks aimed to represent ESG
indicators. The SDG Interface Ontology (SDGIO) [
ESG has become synonym of sustainability, and criteria aimed at measuring and controlling
the sustainability of business’s activities can be decomposed along these three dimensions.
Environmental criteria focus on the impact of the business on environment. Social criteria
assess the impact on the community with which a business interacts, like people, employees,
other companies in the value chain. Finally, criteria related to governance aim at assessing
how much ethics and good practices drive the management of business, considering equity
or transparency of decisions among others. In this work we started the analysis of two of the
existing standards, the DFVA-EFFAS standard [
One of the first attempts to define a guideline for the inclusion of ESG criteria in company
evaluation is the work done by DVFA, subsequently adopted by EFFAS and described in [
Environment, Social and Governance are considered as diferent segments of the general sustainability problem. In addition, a fourth area called LongTerm Viability (identified by the letter V) is defined, that represents the ability of a company to produce long-term profits without sacrificing resources or skills in the short term. Indicators used to quantify and report on companies sustainability are organized according to an ESGV taxonomy that organizes KPIs in General (i.e. applicable to any industrial sector) and sector-specific ESGs topics for each area. Table 1 shows an excerpt of General indicators.
For each indicator a specification is given, shown in Table 2 for indicators ESG 2-1 and ESG 2-2. The specification remains at a very abstract and informal level. Note that indicator ESG 2-2 is in fact a set of specific indicators difering in practice for the normalization factor. Sometimes the normalization procedure is reported, e.g. for the industrial specific ESG 11 NO, SO Emissions, the Unit/Calculation is specified as “Total NO,SO Emissions/total passenger-km” but in general calculation rules are not reported, not even in a descriptive way. Furthermore, it is unclear the relation existing between ESG 2-2 and ESG 11 (i.e., are NO and SO greenhouse gases?)
The standard has undergone through several revisions up to 3.0 [
The GRI Standard is a system of interrelated documents organized into three series: GRI Universal Standards, GRI Sector Standards, and GRI Topic Standards for a total of 37 documents (2021 edition). The first two contain guidelines on how to report information about the impact of a company on the economy, environment, and people, including human rights. The Topic S G V Area ESG E ESG 1 Energy eficiency
ESG 2 GHG emissions ESG 3 Staf Turnover ESG 4 Training & qualification ESG 5 Maturity of workforce ESG 6 Absenteeism rate ESG 7 Litigation risks ESG 8 Corruption ESG 9 Revenues from new products
Indicator ESG 1-1 Energy conpsumption, total ESG 1-2 Energy consumption, specific (intensity) ESG 2-1 GHG emissions, total ESG 2-2 GHG emissions, specific ESG 3-1 Percentage of employees leaving p.a./total employees ESG 4-1 Percentage of trained employees p.a./total employees ESG 4-2 Average expenses on training per employee p.a ESG 5-1 Age structure/distribution (number of employees per age group, 10 year intervals) ESG 5-2 Percentage of workforce to retire in next 5 years ESG 6-1 Number of mandays lost per employee p.a.
ESG 7-1 Expenses and fines on filings, law suits related to anti-competitive behavior, anti-trust and monopoly practices ESG 7-2 Reserves on preventive measurements against anti-competitive behaviour, anti-trust and monopoly practices ESG 7-3 (other) litigation payments, total ESG 7-4 (other) litigation payments, reserves ESG 8-1 Percentage of revenues in regions with TI corruption index below 6.0 ESG 9-1 Percentage of revenues from products at end of life-cycle ESG 9-2 Percentage of new products or modified products introduced less than 12 months ago standards are those of interest for the purpose of this work, containing each a set of indicators for a specific so-called material topic, ranging from economic and financial performance, to the usage and production of materials and resources. An explicit taxonomy does not exists, although the subdivision in documents reflect some form of organization of indicators, as well as reference to general ESG areas. For instance, the standard GRI 305 - Emissions provides 7 indicators: Direct GHG emissions; Energy indirect GHG emissions; Other indirect GHG emissions; GHG emissions intensity; Reduction of GHG emissions; Emissions of Ozone-Depleting Substances (ODS); Nitrogen oxides (NOx), sulfur oxides (SOx), and other significant air emissions. For each indicator a guideline for the calculus is provided by a mathematical expression or, more often, as a textual description. For instance, in the case of GHG emissions intensity: Calculate the ratio by dividing the absolute GHG emissions (the numerator) by the organization-specific metric (the denominator).
Gases included in the calculation; whether 2, 4, 2, , , 6, 3, or all.
The list of organization-specific metrics include: units of product; size, number of full-time employees, monetary units.
The standard also provides other recommendations for the calculation, like the breakdown of the indicator by business, country or type of activity, or the unit of measure.
From this brief analysis it should be clear the dificulty to precisely make sense of these informal specifications and translate them in an actual procedure for the calculation of the indicators. As a result diferent companies, or the same company in diferent times, may easily interpret and implement them diferently, with outcomes hardly comparable.
The Knowledge Graph has been defined by referring to the KPIOnto ontology [ 16], which provides terminology for definition of an indicator and its properties, including a description (KPIDescription), the unit of measurement (unitOfMeasure), its objective (hasBusObj), the aggregation function (aggrType), the compatible dimensions of analysis (hasDimension), and its computation formula if available (hasFormula).10 The ontology has been extended with a further set of domain-oriented object properties, classes and instances, to specify the link between an indicator and the ESG class it refers to (forESGClass), the standard in which it is defined (inESGStandard) and the oficial name used in such a standard ( hasESGCodeName).
To make an example, the following triples define the indicator “ESG 2-2 GHG emissions, total”. It is an Environmental KPI, measured in million of tons, that can be aggregated through sum and can be analysed along the Time dimension. The prefix kpi refers to KPIOnto, while prefix esgo refers to the ESG extension and esg stands is used for definition of indicators. esg:GHG_Emiss_Tot a kpi:Indicator; kpi:KPIDescription "GHG Emissions, total"@en; esgo:hasESGDescription [ esgo:forESGClass esgo:GHG_Emissions; esgo:inESGStandard esgo:DVFA_EFFAS; esgo:hasESGCodename "ESG 2-1" ], [ esgo:forESGClass esgo:GHG_Emissions; esgo:inESGStandard esgo:GRI; esgo:hasESGCodename "GRI 305-1" ]; kpi:unitOfMeasure esgo:MLNtons; kpi:hasBusObj esgo:Environmental; kpi:aggrType kpi:Sum; kpi:hasDimension esgo:Time; kpi:hasFormula [ kpi:hasFunction om:plus; kpi:hasArgument [ kpi:hasArgumentPosition "1"^^xsd:int ; kpi:hasArgumentName "addend" ; kpi:hasArgumentValue esg:Tot_CO2 ], .........].
10Ontology specification is available at https://kdmg.dii.univpm.it/kpionto/specification/
Whenever a computation formula is available, it is represented as a mathematical expression including arguments and operator(s). The formula can be derived from the GRI Standard as the summation of several pollutants, and is only partially reported due to the lack of space. The following triples define the complete formula for indicator “ESG 2-2 GHG emission per unit of production volume”, that can be calculated by dividing the total GHG emissions by the total volume of production. The fragment of the Knowledge Graph representing the above-defined formulas is shown in Figure 1. esg:GHG_Emiss_VolProd kpi:hasFormula [ kpi:hasFunction om:divide; kpi:hasArgument [ kpi:hasArgumentPosition "1"^^xsd:int ; kpi:hasArgumentName "dividend" ; kpi:hasArgumentValue esg:GHG_Emiss_Tot], [ kpi:hasArgumentPosition "2"^^xsd:int ; kpi:hasArgumentName "divisor" ; kpi:hasArgumentValue esg:Tot_VolProd ] ].
The Knowledge Graph contains the definition of 150 indicators, including all 125 indicators from DVSA_EFFRA standard, along with a selected set of indicators from GRI. A number of 57 indicators are compound and their calculation formulas have been defined. It was implemented as a RDF graph in GraphDB, which provides standard SPARQL-based access mechanisms. A fragment of the resulting Knowledge Graph is shown in Figure 2.
The Knowledge Graph representation brings a number of advantages: first of all it provides a formal description of the informal specifications contained in the standards. This often requires to make the meaning of those specifications explicit, and thus should be checked against the original intended meaning of author’s standards, but again the formal representation can simplify the browsing and verification of definitions. Browsing can support also final users when they need to get acquainted with standards, e.g. by supporting to analyse formula-based relations among indicators, as well as by applying formula drill-down, since simple unfolding mechanisms allow to decompose a KPI into its components [17]. It can be adopted as the enterprise reference vocabulary in a supply chain to deal with interoperability of their ESG data and reporting. Ultimately, if the Knowledge Graph is certified by standardization bodies, it can become an invaluable tool for companies and financial institutions to certificate and compare ESG disclosures.
Besides advantages simply related to the existence of a reference model, reasoning mechanisms can be developed on the Knowledge Graph to define flexible and powerful tools for interoperable ESG management. They rely on a set of logic-based functionalities, which also includes math-aware services in Logic Programming to provide non-standard reasoning capabilities[16]: • Automated calculation of Performance Indicators is provided by reasoning services.
Composition of indicators in formulas, based on the semantics of mathematical operators, in fact enables their automatic manipulation through symbolic resolution of equations. Such services are capable to derive all alternative formulas for a given indicator (get_formulas), to evaluate the common set of dependencies among a set of indicators (get_common_dependencies), or to derive what indicators can be computed starting from those already available (get_computable_indicators). • On top of the reasoning services, advanced functionalities can be defined with the purpose to support monitoring of indicators across multiple organizations. This may be particularly useful for cooperating organizations to assess the status of shared Business Processes. The set of KPIs produced by the organizations is compared in order to derive, through the exploitation of get_common_dependencies and get_computable_indicators, common indicators among them. They include not only indicators that are mapped to the same ontological concepts, but also indicators that can be calculated from those available. • Management of the graph, including the creation of new indicators or the maintenance of their definitions rely on consistency check mechanisms. They are capable to verify that a new indicator is neither equivalent to nor incoherent with already defined ones. While equivalence_check may help to perform de-duplication of indicators to minimize redundancy, the guarantee given by coherence_check is critical to keep the graph consistent by avoiding contradictory definitions [16].
From the analysis of standards it emerges a profound heterogeneity in the format, structure, terminology adopted in the definition of indicators, and in the approach to their categorization. This is widely recognized as a limit to the development of actions towards a green transition, as discussed in the Introduction. In this respect, semantic technologies can and should provide support to eforts towards homogenization and harmonization. The present paper provides a contribution in this direction, starting from the construction of a Knowledge Graph for the modeling of ESG indicators. This sometimes required a hard work of interpretation of underspecified definitions, that could be not correct or shared by standardization bodies. We believe that the formal representation produced would ease a possible verification and dialogue with stakeholders.
The inherent modularity and flexibility of a graph structure, combined with the ability to represent and reason on the compound nature of indicators, has demonstrated to be suitable to tackle the challenge in an incremental way, checking time by time the coherence of the overall graph.
Still, several open issues exist that are not covered in this preliminary proposal and represent interesting research directions: • while the expressivity of the model has demonstrated itself suficient to represent the ESG indicators structure (formula), some information contained in the standards still remains uncovered. It is mainly related to the semantics of atomic indicators that, despite being atomic, often have an underline calculation procedure not falling in the formula category. As an example, “processed orders” can be defined as a subclass of orders, but besides requiring to resort to domain ontologies, which is certainly possible, it is purely terminological and does not solve the ultimate challenge of providing the interpretation. A step towards the implementation of efective tools for the calculation and certification of indicators would require a data integration system and GAV/LAV mappings as a means to directly refer to data in business’s information systems that correspond to an atomic indicator. • There is evidence of specifications at diferent abstraction levels in the standards. For sure the standard cannot cover the lowest enterprise level. This suggest that the model should be extended to explicitly deal with and support abstraction management. • Localization issues: there is evidence of diferent implementations of the standards, for instance the GRI standard is provided in several languages, with diferent names for indicators. This issue can be addressed by referring to best practices for multilingual ontologies, ranging from simple design patterns such as referring to textual labels for the indicator name/description with an attached language tag, to more complex patterns such as language content negotiation. However, other more tricky localization issues exist, related to the definition of the indicator’s formula, for instance due to diferences in country’s constraints and regulations (e.g. the definition of employee and its categories, the components of a salary) • Standards are “living bodies” subject to revision. Hence the model should be endowed with the capability to support changes and versioning. • The graph model simplifies the integration of diferent standards, however instead of adopting this view, an alternative approach towards interoperability of standards can be pursued. This implies the modeling of each standard as a distinct Knowledge Graph with its own categorizations and definitions, building relations between graphs’ concepts. Interoperability would correspond to more flexibility and simplicity in the management of standards than a design-time integration of the diferent views provided by the standard. Organizations may disclose their ESG indicators following the preferred standards, and suitable reasoning over the Knowledge Graphs would still provide proper translations and correspondences. In this direction, the model easily supports the derivation of the equivalence of compound indicators provided that “same-as“relations are established among atomic indicators in diferent Knowledge Graphs. This reduces the cost of a manual alignment or of generating large shared vocabularies of indicators terms, at the same time improving the quality with respect to simple terminological similarity techniques.
This work was partially funded by Veeco Srl. Authors would like to thank Miriam D’Aloia for her contribution in the development of the Knowledge Graph. and Corporate Valuation. Version 3.0, https://efas.com/wp-content/uploads/2021/09/KPIs_ for_ESG_3_0_Final.pdf, 2010. Accessed 25-02-2022. [16] C. Diamantini, D. Potena, E. Storti, Sempi: A semantic framework for the collaborative construction and maintenance of a shared dictionary of performance indicators, Future Generation Computer Systems 54 (2016) 352–365. [17] C. Diamantini, D. Potena, E. Storti, H. Zhang, An ontology-based data exploration tool for key performance indicators, in: OTM Confederated International Conferences" On the Move to Meaningful Internet Systems", Springer, 2014, pp. 727–744.