Federated learning could make ESG reporting scalable and privacy-preserving, but progress is blocked largely by absent data standards and poor data quality, according to an expert Delphi study; policymakers and firms should prioritise interoperable taxonomies and data-quality assurance before large-scale FL rollout.

Main Finding

The paper finds that federated learning (FL) can improve the scalability and credibility of ESG reporting, but adoption is substantially hindered by multiple challenges. Using expert validation and fuzzy multi-criteria ranking, the authors identify and prioritize 18 anticipated challenges; the top two obstacles are lack of ESG data standardization and poor ESG data quality.

Key Points

• Growing demand for ESG reporting motivates exploration of FL as a privacy-preserving way to aggregate cross-firm information without centralizing raw data.
• Adoption of FL for ESG reporting is complex; the paper systematically identifies and ranks the obstacles.
• Study workflow: literature search (Scopus, Web of Science) → initial pool of 15 challenges from literature → expert panel added 4, 1 was rejected in Delphi consensus → final set of 18 validated challenges.
• Validation used the Pythagorean Delphi technique; prioritization used Pythagorean fuzzy Analytic Hierarchy Process (AHP).
• Top-ranked barriers: (1) lack of ESG data standardization, (2) ESG data quality issues.
• Theoretical contribution: maps the role and friction points of FL in ESG reporting literature. Practical contribution: prioritized, actionable insight for stakeholders to target interventions.

Data & Methods

• Data sources: scholarly literature identified via Scopus and Web of Science (details of search strings, time range, and counts not reported in abstract).
• Challenge identification: initial list of 15 challenges from the literature; expert panel review produced +4 candidate challenges; one candidate was rejected via Delphi consensus, yielding 18 validated challenges.
• Validation: Pythagorean Delphi technique — an extension of classical Delphi that uses Pythagorean fuzzy sets to capture experts’ uncertainty and consensus more flexibly.
• Prioritization: Pythagorean fuzzy AHP — a multi-criteria decision-making method where pairwise comparisons use Pythagorean fuzzy numbers to reflect imprecision in expert judgments; used to rank the 18 challenges by importance.
• Output: a prioritized list of 18 challenges with the top two identified above. (The abstract does not report the number of experts, weights, or detailed ranks beyond the top two.)

Implications for AI Economics

• Market frictions & public goods: ESG data standardization is effectively a public-good problem — lack of interoperable standards raises coordination costs, reduces comparability, and inhibits network effects that FL needs to scale. Economics research should model standardization as an equilibrium outcome of firm incentives and coordination mechanisms.
• Information asymmetries and market efficiency: poor ESG data quality weakens the informational benefits of FL, limiting effects on capital allocation, risk pricing, and corporate governance. Economists should quantify how data-quality improvements via FL could change cost of capital and investment flows.
• Incentives for data sharing: FL reduces privacy and proprietary-data costs, but firms still face strategic incentives not to share noisy or unfavorable ESG signals. Mechanism-design and contract-theory approaches can help design incentives, reputation systems, or compensation schemes that induce truthful participation.
• Regulatory and policy roles: prioritized interventions should target standards and data-quality assurance (e.g., mandated reporting schemas, third-party verification, interoperable taxonomies). Subsidies or coordination platforms may be needed to overcome collective-action problems and finance upfront costs of data governance and FL integration.
• Competition and market structure: FL may alter competitive dynamics by enabling joint model benefits without raw data pooling; however, heterogeneous adoption and data quality can create winner–loser dynamics. Empirical and theoretical work should explore adoption thresholds, externalities, and welfare implications.
• Research directions: cost–benefit analyses of FL adoption in ESG reporting; dynamic adoption models with learning and network effects; empirical measurement of how FL-enabled ESG reporting affects firm valuation, lending spreads, and investor behavior; policy experiments on standardization and audit mechanisms.
• Practical recommendations for stakeholders: prioritize establishing common ESG taxonomies and minimum data-quality standards; invest in data governance and cleaning; pilot interoperable FL platforms with clear governance and audit trails; involve regulators and standard-setters early to reduce regulatory uncertainty.

Summary takeaway: Federated learning promises to enhance ESG reporting while preserving firm privacy, but its economic value depends first on solving collective data problems—standardization and quality—so policymakers and firms should target those areas to unlock FL’s benefits.