The impact of corporate governance on narrative disclosure tone: a machine learning approach

Abstract

Purpose This study examines the influence of corporate governance indicators (CGIs) on the textual tone of nonfinancial firms in a developing economy. Design/methodology/approach The data from 1,250 annual reports of listed nonfinancial firms in Pakistan are collected for 10 years. The narrative disclosure tone (NDT) is derived using the sentiment analysis of annual reports, resulting in six distinct NDT scores. The CGIs data are also extracted from the annual reports. The fixed effects model is used as the primary analytical tool, supplemented by machine learning-based linear regression. System GMM and two-stage least squares regressions are employed for robustness checks. Findings The findings reveal that most CGIs significantly influence all six NDTs. These results align with the existing theoretical literature, except those related to audit committee independence and gender diversity. Research limitations/implications The study is limited to the use of annual reports as a source of narrative disclosures. Future research might employ other sources, such as earning press releases and social media. Practical implications Within the unique regulatory environment of Pakistan, the study offers insights for regulators to enhance the efficacy of independent directors, discourage concentrated ownership and promote the inclusion of women in board subcommittees to establish the authenticity of textual disclosures. Originality/value The study adds to the limited literature on the determinants of NDT. It underscores the importance of understanding textual tone for informed investor decision-making and restoring investor confidence. Moreover, it contributes by focusing on six NDTs and exploring the interplay between CGIs and textual tone.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 18 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.