Methodology
The Earth Capital Nexus team at the Grantham Research Institute on Climate Change and the Environment, London School of Economics and Political Science (LSE EarthCap), with support from Arc, has developed the Corporate Resilience Alignment Benchmarking (C-RAB) methodology.
At its core, C-RAB assesses corporate disclosures to derive company-level resilience signals that can complement existing asset-level geospatial risk modelling approaches.
The C-RAB methodology was developed with the recognition that, irrespective of future emissions trends, there is a need to build resilience to the impacts of climate change and nature loss already locked in. Financial institutions, investors, and regulators do not yet have sound oversight of what companies are doing to address these risks and how prepared they are. This oversight gap reduces the capacity of investors and supervisors to monitor portfolios and economies effectively.
The methodology has been developed with the financial sector in mind—particularly investors—but the elements it contains and the insights it produces are intended to be valuable to a wider array of actors, including regulators, supervisors, companies themselves, and civil society.
The full version of the C-RAB methodology can be accessed here.
4.Alignment of C-RAB with existing frameworks
The definition of resilience adopted in the C-RAB methodology aligns with the definition laid out in the IPCC (2022): “The capacity of interconnected social, economic and ecological systems to cope with a hazardous event, trend or disturbance, responding or reorganising in ways that maintain their essential function, identity and structure”. Conceptually, in assessing corporates, the methodology does not limit itself to how or whether a company can adapt to the physical impacts of climate change and environmental degradation. It also considers how that company affects the ability of others to adapt—a double materiality approach.
C-RAB builds on established principles of climate resilience-aligned investment, adopted widely in adaptation investment taxonomies, as well as other frameworks guiding corporate adaptation planning. A core design principle of C-RAB is to build on existing frameworks. To support interoperability and reduce additional reporting burden on companies, the indicators in C-RAB have been mapped to the indicators and disclosure elements of related frameworks, including:
- IFRS Sustainability Disclosure Standards (ISSB IFRS S1 and S2)
- Task Force on Climate-related Financial Disclosures (TCFD)
- Task Force on Nature-related Financial Disclosures (TNFD)
- ACT Adaptation methodology
- CDP Climate Change Questionnaire
- Corporate Sustainability Reporting Directive (CSRD) ESRS E1
A partial mapping is included in Table 2 below. The full mapping is available in the LSE EarthCap C-RAB methodology.
| C-RAB Tier-1 | C-RAB Tier-2 sub-metric | ISSB IFRS S1 & S2 | TCFD | TNFD | ACT Adaptation | CDP | EFRAG |
|---|---|---|---|---|---|---|---|
| Targets | Climate-related hazard targets | S2 para 33-37 | Metrics & Targets b) | 1.1 | E1-4 | ||
| Impact-based-driver targets | S1 para 51 | D.2; D.1; 14 core global disclosure metrics | 1.1 | 9.3 | E2-4, E3-4, E4-4, E5-5 | ||
| Implementation | Products and offering | S2 para 14; S2 para 22a | Strategy b) | TNFD 14 core global disclosure metrics | 6.3; 8.1 | Q5.4 / Q5.4.1 / Q5.4.2 / Q5.4.3; Q3.6 / Q3.6.1 / Q3.6.2; Module 9 Water module | E1-3, E2-3, E2-4, E2-5, E3-2, E3-4, E5-2, E5-5 |
| Engagement | S2 para 14; S2 para 22(a)(iii) | B.1 | 2.1, 2.2, 4.2 | Q4.10; Q4.11 / Q4.11.1 / Q4.11.2 ; Q5.11 series | E1-3; ESRS 2 SBM-2; ESRS 2 GOV-3 / E1 GOV-3, E2-2 | ||
| Investment | S2 para 22; S2 para 29 | Strategy b); Metrics & Targets c) | B.4; D.1 | 7.1; 7.2 | Q5.3 / Q5.3.1 / Q5.3.2; Q5.4 / Q5.4.1 / Q5.4.2 / Q5.4.3; Q5.6 / Q5.7 / Q5.7.1; Q5.9 | E1-1; E1-3; E1-9 | |
| Processes | Risk and impact assessment | S2 para 25; S2 para 22(b); S2 para 30; S1 para 43-44 | Risk Management a); Risk Management b); Strategy c) | C.1; C.2. LEAP approach - Locate, Evaluate, Assess, Prepare | Module 4; Module 5 | Q2.1; Q2.2 / Q2.2.1 / Q2.2.2; Q2.2.7; Q2.3; Q2.4; Q5.1 / Q5.1.1 / Q5.1.2 | E1 IRO-1; E1 SBM-3; E1- |
| Alignment across strategy | S2 para 14; S2 para 22(a)(ii); S1 para 28 | Strategy b) | B.2 | 1.1; 3.1 | Q5.2; Q4.6 / Q4.6.1 | E1-1; ESRS 2 SBM-1 | |
| Alignment across business processes | Risk Management c); Strategy b) | C.2; C.3 | 6.1; 6.2, 7.1 | Q5.3 / Q5.3.1 / Q5.3.2; Q4.6.1 | E1-2; E1-3; ESRS 2 IRO-1 | ||
| Governance | Board composition | S2 para 6 | Governance a) | A.1.ii | 1.2, 9.1, 9.2 | Q4.1 / Q4.1.1; Q4.1.2; Q4.2; Q4.4 | ESRS 2 GOV-1 |
| Oversight mechanisms | S2 para 6 | Governance a); Governance b) | A.1; A.2 | 1.2, 6.1 | Q4.1.2; Q4.3 / Q4.3.1; Q5.1 / Q5.1.1 | ESRS 2 GOV-1; GOV-2; GOV-5; E1 GOV-3 | |
| Management incentives | S2 para 6 | Governance b) | A.2 | 1.2; 9.1 | Q4.5; Q4.5.1 | E1 GOV-3; ESRS 2 GOV-3 | |
| Reporting | 3.1 | Q4.12 / Q4.12.1; Q13.1 | ESRS 2 BP-1, BP-2 |
4.C-RAB metrics, sub-metrics, and indicators
Assessing corporates directly on resilience requires granular, contextual knowledge of whether a given action has produced resilience benefits. Doing so at scale, and in a manner that supports comparison across companies, is beyond the realms of feasibility. C-RAB therefore assesses the degree to which a company's practices align with the principles that define resilience.
This is an important distinction and merits emphasis. The results of a C-RAB assessment do not tell users that Company A is more resilient than Company B. They tell users which company has more closely aligned its business practices with principles that define resilience. This is an intentional methodological distinction that shapes what C-RAB can and cannot be used to claim.
The C-RAB framework is a hierarchical combination of indicators organised across four tiers. Indicators at the highest level of aggregation (i.e. Tier 1) are referred to as Metrics, and are as follows:
- Processes: How risk and impact assessments are conducted and integrated into business strategy and operations.
- Implementation: The actions a company is taking, including via its products, engagement, and investment.
- Governance: Board composition, oversight mechanisms, management incentives, and reporting.
- Targets: Whether a company has set time-bound and specific targets to address the climate-related hazards and direct impact drivers most significant to its sector.
Under each metric lie multiple layers of sub-metrics and indicators, organised in a tiered hierarchy (see Table 4 below). Full explanatory notes for each indicator are provided in the LSE EarthCap C-RAB methodology.
| Tier 1 (Metric) | Tier 2 (Sub-metric) | Tier 3 (Indicator) |
|---|---|---|
| Processes | Risk and impact assessment | Risk assessment |
| Scenario analysis | ||
| Impact assessment | ||
| Alignment of business strategy | Integrating adaptation | |
| Integrating do no significant harm (DNSH) | ||
| Alignment of business processes | Mainstreaming A&R and DNSH into business processes | |
| Acting on environmental risk and impact assessments | ||
| Acting on social risk and impact assessments | ||
| Implementation | Product and offering | Share of revenue from activities that build resilience |
| Products and services that do no significant harm | ||
| Engagement | Value chain engagement | |
| Government engagement | ||
| Investment | Share of investment that builds resilience | |
| Adaptation plan | ||
| Governance | Board composition | Board member expertise |
| Board member training | ||
| Oversight mechanisms | Responsibility over risk and adaptation | |
| Using risk assessment for decision-making | ||
| Signing off on scenario analysis | ||
| Board internal performance review | ||
| Risk committee mandate | ||
| Management incentives | Types of entitlement | |
| Reporting | Safeguarding | |
| Accountability | ||
| Targets | Climate-related-hazard targets | Each company's material risk |
| Impact-driver-related targets | Each company's industry-based impact driver |
The framework has been thoroughly reviewed by a range of expert reviewers—please refer to the C-RAB methodology for further details.
4.Producing C-RAB data using a large language model (LLM)
C-RAB has been designed to be adaptable and applicable across different contexts. Users of the methodology may choose to run their own assessments in-house using public disclosures, privately disclosed information (for example, during due diligence) or other sources. Assessments may be conducted manually by trained human assessors or with the support of large language models.
For ResilienceArc, the C-RAB methodology was combined with a state-of-the-art large language model (LLM), co-developed by LSE EarthCap and the University of Zurich. The C-RAB indicators were reformulated into a series of questions that the LLM answered using a dataset of publicly available disclosures for each company.
Information was scraped from public websites. On average, 25 files were compiled for each company, comprising annual reports, sustainability reports, regulatory filings, corporate policies, and webpages. Where possible, only the latest version of each document was used.
Supplementary information was provided for each question to guide the assessment and enhance reliability of the data, in line with previous approaches. This supplementary guidance information was drafted iteratively and tested on multiple occasions to ensure accuracy of LLM response.
Beyond the comprehensive testing and validation exercises conducted for this model, following the procedure of Schimanski et al., LSE EarthCap also compared the results of the tool with equivalent human responses by hand. For the Boolean indicators, an expert human analyst evaluated 455 questions. The expert manually answered a question based on the reports and compared the answer with the LLM’s output.
Through this procedure, LSE EarthCap identified that 83% of the LLM responses were matched. Discrepancies between the LLM and the human evaluator arose primarily because the human evaluator inferred additional information from other publicly available sources (e.g., a specific human rights policy), whereas the LLM relied more strictly on the information provided in the sustainability report.
A similar procedure was carried out for number-retrieval-style indicators. Number retrieval indicators ask the LLM to find a specific number or ratio in a company’s disclosures. Here, two expert analysts evaluated 100 questions, with each human analyst evaluating them once. The results between the two human analysts matched 75% of the time, while with the LLM they matched 65-67% of the time, depending on the reviewer.
The discrepancies among the human evaluators reflect the at-times contradictory statements made by corporates in their disclosures. The fewer matches between human and LLM results are also due in part to human evaluators being able to infer ratios, whereas the LLM required a ratio to be given. For example, human evaluators can calculate ratios from absolute numbers, whereas the LLM will only reproduce exact ratios that are given.
The same procedure was applied to category-style indicators. Category indicators ask the LLM to find information and classify a company into a set of (usually five) predefined categories based on that information. Here, two human analysts each manually evaluated 210 indicators across 7 questions (i.e. 30 per question). The results of the two human analysts, on average, matched 60% of the time, with a range of 43-93% depending on the indicator. The results between the LLM and the analysts matched 46-57% of the time on average. This discrepancy may seem high. However, when allowing for error within a category, concordance was 77% between the two human analysts and 64-73% between the LLM and each human analyst. The lower matched rates, both between the human analysts and the LLM, compared to the Boolean or number retrieval indicators, reflect the difficulty even humans have in capturing the nuance that category questions require. The LLM generally classified responses more favourably than human analysts.