Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors,

Thank you for your interesting contribution.

This paper examines the predictive ability of binary sustainability and governance disclosures on the financial performance of firms in terms of Return on Equity (RoE) based on machine learning classification models (XGBoost, LightGBM, Random Forest) on 427 Turkish firms.

Although the models marginally outperform the random classification, they show moderate accuracy in general, and this shows that mere presence or absence of disclosures alone can not be a good predictor.

The study uses explainable AI methods (SHAP) to determine influential indicators of governance, including sustainability committees and particular reports, only to conclude that their impact is small and context-specific.

The results mention that more specific, transparent, and high-quality sustainability reporting is required to aid the decision-making of investors and support value creation, particularly in the emerging markets where the traditional regression method and ESG scores are not as efficient.

Some suggestions are:

• Elaborate and reinforce the theoretical framing by further specifying how the resource-based view is connected to the selection of binary governance and sustainability signals and why these trivial signals could reflect strategic value.
• Add more explanation as to why those particular machine learning models in particular (XGBoost, LightGBM, and Random Forest) have been chosen and explain why they are better suited to this study.
• Add to the accuracy description of model performance metrics the discussion of additional metrics of performance, including, but not limited to, accuracy, precision, F1-score, or AUC to provide a more comprehensive view of the effectiveness of classification.
• Explain possible factors that are going to lead to this meager predictive power, such as data constraints, quality of the indicators, or market-specific factors, and propose ways in which future research can help overcome such problems.
• Elaborate on the consequences of SHAP explainability findings and discuss how the identified key indicators might be used in making sustainability reporting practices and governance systems better.
• It is also possible to think about a more detailed comparison with the standard regression-based methods to show the value and the limitations of the classification and machine learning system.
• Make the presentation more readable and concise, especially in the results and discussion parts, to convey the important findings and their importance more effectively.
• Discuss the possibility of differences in data quality and the heterogeneity in reporting between firms on model results, and explain how these issues can be addressed.
• Provide recommendations on future research (including the addition of more granular or qualitative sustainability data, the introduction of other emerging markets, the application of other machine learning methods, etc.).
• Enhance the conclusion by making there a specific connection between findings and practical recommendations to policy makers, investors and corporate managers on sustainability disclosure and practices of governance.

I hope these comments and suggestions may help improve the quality of the manuscript.

Kindest regards

 

Comments on the Quality of English Language

The general level of English is sufficient for academic publication, although it might need a thorough proofread to eliminate some clumsy wording and enhance fluency.

For instance:

• Certain sentences are too long or complicated and this influences the readability of the sentence, a shorter and more comprehensible sentence would improve the understanding of the sentence.
• Technical explanations and terms are well-elaborated, but it would be better to have consistency in the terminologies (e.g., use of the term sustainability committees instead of corporate committees).
• Simple grammatical mistakes and inconsistencies in the punctuation occur in isolated cases and they need to be corrected to achieve high academic standards.
• Abstract and conclusion parts should be made in a shorter language to address the main findings and contributions more clearly.

To finalize the paper and make sure it is of international scholarly standards, it is worth considering hiring professional language editing services.

Author Response

 Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Thank you for the opportunity to review this interesting paper. It deals with an up-to-date topic in a relatively innovative approach.

Abstract explains well the content of the paper, charts the purpose, used methodology and main results.

The introduction chapter gives the framework of the studied phenomena; the literature review includes relevant theories and authors. The research gap in the existent literature is clearly explained. The studies on the relationship between sustainability reporting and financial performance are mainly conducted using regression-based relationship tests based on ESG scores, while there are quite rare models which analyse the presence or absence signal effect of corporate disclosures. Present research tests the discriminative capacity of binary indicators related to sustainability reporting in classifying the return on equity of the screened companies as high/low using machine learning methods.

The used methodological approach is valid and proven, suitable for this kind of study. The chosen sample of sustainability reports from 427 companies traded on the Istanbul Stock Exchange in 2024, is current, relevant and publicly available, so it permits the repetition of the study. Three tree-based models were classified according to the binary system.

Results demonstrate that tree-based models perform just a little better than the naive majority class rule and therefore have limited overall classification power. Important finding is that SHAP-based explainability analysis shows that the governance reports, integrated reports, and the existence of a sustainability committee stand out globally in classifying RoE compared to other variables.

Another important finding is the expressed need for reporting practices with deeper content, clearer evidence of governance quality, and stronger data integrity to better support investors' decision-making processes through sustainability and governance.

One recommendation which is suggested for a more fluent reading of the paper, even if the reader is not acknowledged with the narrow field would be, to explain the acronyms, when they first appear in the text (e.g. SHAP).

Author Response

 Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Thank you for the opportunity to review this manuscript, which presents a relevant and timely contribution to the field of sustainability reporting and classification methodologies. The authors investigate innovative approaches to classifying publicly available sustainability- and governance-related committee reports. In doing so, they identify a meaningful research gap: although econometric modelling is widely employed in the field, existing studies often lack systematic approaches and rigorous quantitative analyses. The authors address this gap using a dataset comprising 427 companies listed on the Istanbul Stock Exchange in 2024.

While the manuscript offers valuable insights, several areas could be further strengthened. Notably, the paper would benefit from including key standard sections typically expected in empirical research, such as recommendations, limitations, implications, and avenues for future research. Explicit recommendations tailored to different stakeholder groups – policy makers, corporations, investors, civil society, and academic audiences – would enhance the practical relevance and support a more holistic presentation of its contributions and serve as a solid background for the section on implications.

The research methodology and results presented in the manuscript have limitations that are not discussed. A dedicated limitations section, coupled with guidance for future research, would provide transparency and support other scholars' research. Additionally, the Discussion section could be reconsidered. Although positioned after the Results, the section would benefit from a more extensive engagement with existing literature. Given the dynamic nature of the research field, critically comparing the authors’ findings with those of other scholars would enrich the analysis and offer readers a better understanding of the implications.

The comments provided do not question the relevance of the research presented. The publication could be recommended for publication after some minor improvements.

Author Response

 Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

• The manuscript is overly long and repetitive, especially in the literature review and introduction. Several sentences are unclear or grammatically inconsistent; professional language editing is needed.
• The paper claims to fill a “major methodological gap”, but the gap is overstated. The theoretical discussion is extensive but not tightly linked to the research question. The introduction over-explains background theories without synthesising why binary disclosures theoretically should classify ROE. (SUGGESTION: needs more justification)
• Use of Q1 2025 ROE as the sole measure of financial performance is problematic. Quarterly ROE is volatile and subject to seasonal and sector-specific effects. (SUGGESTION: provide robustness checks (e.g., annual ROE, alternative profitability measures, or sector-adjusted ROE). The choice of median split to define high/low ROE needs more justification. Why not quantiles, industry-adjusted medians, or continuous modelling? There is no sensitivity analysis to confirm that results do not change materially across alternative thresholds.
• The paper itself shows that binary indicators offer limited signal value. However, there is no deeper investigation as to whether richer textual content, length of reports, committee activity level, or presence of KPIs would improve predictive power. (SUGGESTION acknowledge that binary tagging may be too coarse and suggest integrating richer ESG textual or numeric data)
• Model performance is extremely low, barely above majority-class accuracy, yet the paper still interprets weak SHAP signals as meaningful. SUGGESTION: Add confidence intervals, alternative evaluation metrics (AUC, balanced accuracy), and an expanded discussion of why models perform poorly.
• No discussion is given on tuning procedures, hyperparameter choices, or cross-validation. (SUGGESTION: provide details on tuning grid, cross-validation method, and prevention of overfitting.)
• SHAP values are correctly computed, but their interpretation is overstated. For instance, mean SHAP values around 0.02–0.08 indicate extremely weak signals. SUGGESTION: The paper should explicitly acknowledge that SHAP importance ranking does not imply practical materiality.
• Table 3’s p-values appear inconsistent with standard chi-square distribution expectations (e.g., χ² = 27.401 corresponds to p < 0.001, not 0.0979). SUGGESTION: Check for reporting errors in all statistical parts.
• Some references lack full publication details or correct formatting. SUGGESTION: Ensure MDPI sustainability referencing guidelines are followed strictly.
• Please, expand the discussion of weak predictive performance instead of trying to extract strong conclusions from weak signals. In addition, add limitations acknowledging the coarseness of binary data and possible measurement error in sustainability reporting.

Author Response

 Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 4 Report

Comments and Suggestions for Authors

Thank you for your quick response to my comments on your respective scientific article.

Over 95% of my comments were addressed correctly, especially those concerning the length of the article and the need to review and revise the statistical section (p-value). I believe the research, in its current state, is highly suitable for publication unless the other reviewers have a different opinion.

Author Response

 

Dear Reviewer,

Thank you very much for your positive evaluation and supportive comments in the second review round. We appreciate your recognition of our revisions and your conclusion that the manuscript is suitable for publication in its current form.

Sincerely.