How Corporate ESG Narratives Evolved around a Disclosure Mandate: Textual Evidence from a Frontier Market

This paper examines whether mandatory ESG disclosure in a frontier market produces substantively differentiated reporting or merely symbolic, boilerplate language, addressing a question central to sustainable development: how to assess disclosure content where third-party ESG ratings are sparse or absent. Using Kenya's 2021 Nairobi Securities Exchange mandate as the setting, the study applies natural language processing, combining a fine-tuned transformer-based classifier with BERTopic topic modelling, to a hand-collected corpus of over 100,000 sentences from 59 listed firms across 2010 to 2024. Disclosure intensity is traced through firm fixed-effects regressions and its substance through the cross-sectional distribution of topics across sectors. Within firms, disclosure rose after the mandate by about eleven percentage points, concentrated in the environmental and social pillars. More decisively, disclosure topics are strongly conditioned on sector, each industry foregrounding its own material risks, a pattern inconsistent with symbolic convergence and indicative of substantive reporting. For regulators in emerging markets, the findings suggest that structured, market-level disclosure mandates are associated with more extensive and more sector-specific reporting than voluntary initiatives alone, while the text-based measure offers a transparent, low-cost tool to monitor disclosure where commercial ratings are unavailable. By showing that mandated disclosure can elicit substantive rather than symbolic reporting, the study supports efforts to improve corporate accountability in Sub-Saharan Africa, contributing to the United Nations Sustainable Development Goals on climate action, decent work, and responsible institutions. It offers the first textual analysis of ESG disclosure in a Sub-Saharan frontier market, introducing sector-specific materiality as an internal validation strategy for text-based disclosure measures in settings that lack external ratings.