All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
Both reviewers have agreed that the authors have addressed their comments.
[# PeerJ Staff Note - this decision was reviewed and approved by Claudio Ardagna, a PeerJ Section Editor covering this Section #]
The authors have addressed all the concerns. The paper can be accepted in its current state.
NA
NA
NA
The article is in acceptable form after the update
The reviews regarding experiment section are well answered
The results and updated article can be accepted
The reviewers find that the manuscript does not clearly establish the contribution of applying SVR to green finance beyond prior forecasting studies. They point to methodological weaknesses (fixed RBF kernel, default parameters, risk of information leakage), limited dataset description and AHP validation, and the absence of comparisons with modern ML baselines. Results are reported only numerically without linking to finance practice. Reviewers recommend clearer justification of contribution, systematic parameter tuning, adaptive kernels, detailed data reporting, and stronger benchmarking.
**PeerJ Staff Note:** Please ensure that all review, editorial, and staff comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
The current framework uses ESG-weighted scores as features and expert-normalized “investment probability” as the label, both derived from the same expert pool. This design risks information leakage and circular validation.
Revision Suggestion: Separate the construction of features and labels. For instance, Group A experts could provide ESG dimension ratings (later aggregated via AHP), while Group B experts or ex-post performance outcomes (e.g., default/return after 1–2 years) should define the labels. Alternatively, retain the same experts but ensure a temporal separation—features assessed at time T and labels validated at T+1.
The manuscript only mentions applying AHP weighting without reporting Consistency Ratio (CR) or robustness checks. Without this, the reliability of the weight assignment is uncertain.
Revision Suggestion: Report the full AHP judgment matrix, compute CR values (ensuring CR < 0.1), and present the hierarchical structure including secondary-level indicators. Conduct sensitivity analysis by perturbing weights ±10% to evaluate stability of the SVR results, and visualize the performance fluctuation (e.g., RMSE vs. weight perturbation curve).
The SVR model is reported with “default parameters,” yet no systematic search for C, ε, and γ is described. This weakens the credibility of the reported performance.
Revision Suggestion: Employ nested cross-validation (outer 5-fold, inner Bayesian optimization or grid search) to jointly optimize C ∈ {10^-3…10^3}, ε ∈ {10^-3…10^-1}, and γ ∈ {10^-3…10^1}. Provide contour plots or heat maps to illustrate the validation error response surface with respect to (C, γ), and explicitly report the final optimal parameters with confidence intervals.
The study only considers the Gaussian kernel. However, ESG-related features may exhibit heterogeneous distributions that could benefit from hybrid kernels.
Revision Suggestion: Explore multi-kernel SVR (e.g., linear + RBF, polynomial + RBF) and compare against single-kernel performance. This would enhance robustness and generalizability, especially across heterogeneous green finance datasets.
The dataset description lacks clarity on sample size, time span, and industry coverage. Without this, reproducibility is limited.
Revision Suggestion: Provide a detailed data profile table (sectors, sample counts, observation periods) and, if confidentiality is an issue, include summary statistics or synthetic examples.
Current baselines are limited. The strength of SVR cannot be fully assessed without comparisons to alternative machine learning models.
Revision Suggestion: Benchmark against Random Forests, Gradient Boosting (XGBoost/LightGBM), and shallow Neural Networks. Present statistical tests (e.g., paired t-test or Wilcoxon signed-rank test) to demonstrate whether performance differences are significant.
Several equations are incompletely formatted, and figures lack adequate axis labels and units.
Revision Suggestion: Revise mathematical notations for consistency, add comprehensive captions to figures, and ensure all variables are explicitly defined in the text.
The study presents SVR as the core method but does not sufficiently explain why SVR is particularly suited for green finance risk assessment beyond generic regression. Existing works already apply SVR to financial forecasting.
- Clearly articulate what is novel—e.g., does the integration with ESG scoring introduce a fundamentally new angle, or could any regression model achieve similar results? Without this clarification, the contribution is overstated.
- A fixed RBF kernel is used across all samples, yet the ESG space may be non-uniform across firms or sectors. There is no attempt to learn or adapt the kernel function based on data characteristics.
- Explore adaptive kernel learning or integrate locally weighted SVR to improve model fit in heterogeneous ESG datasets.
- The SVR parameter optimization (C, ε, γ) is mentioned but not explained. Provide a table of candidate parameter ranges, optimization method (grid search, Bayesian optimization), and report final tuned values.
- Add a comprehensive dataset description table, including sector breakdown, observation years, sample size, and missing data treatment.
- Experiments only benchmark SVR against itself or basic variants. Without comparing to modern alternatives (e.g., Random Forest, XGBoost, LightGBM, or simple linear regression as a baseline), the claim that SVR is superior remains unconvincing.
- Several equations are poorly formatted or missing variable definitions (e.g., in the SVR formulation, ε-insensitive loss is not clearly defined).
- Rewrite all mathematical expressions, define every parameter upon first use, and ensure consistency in notation.
- The discussion of results is purely numerical (RMSE, MAE) without connecting back to what these results imply for green finance practice (e.g., investment screening, portfolio risk management).
- Expand the discussion to explicitly explain how the model’s predictions would assist financial institutions, policymakers, or ESG investors in decision-making.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.