Energy Dependence, Environmental Quality and Banking Sector Capital: New Evidence from OECD Countries

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Please see the attachment.

Comments for author File: Comments.pdf

Author Response

Point to Point Answers to Reviewer 1

 

Q1. The manuscript makes an important distinction between physical risks and transition risks, but it could do more to explain how these risks actually affect bank capitalization. For example, how exactly do methane emissions put pressure on banks’ capital? Is it because borrowers become less creditworthy, regulators demand more capital or because of wider economic uncertainties?

A1. Firstly, we would like to thank the reviewer for their valuable comments that helped us shed light on an important component of the transmission channels from environmental risks to bank capital. With that in mind, we have modified Section 1: Introduction in order to provide a clearer description of both channels through which physical and transition risks influence banks' capital levels. To be more specific, physical risks, including air pollution, mainly work through macroeconomic effects and effects on borrowers, thus negatively impacting the economy and reducing borrowers' creditworthiness, resulting in higher default risks and lower bank capital. As far as transition risks are concerned, the paper elaborates on the meaning of indicators like methane emissions (CH4P) used in our empirical study as proxies for risks related to carbon-intensive industries facing tighter regulations. We discuss three main channels of risk transmission: (i) credit risk channel, due to the reduction in profit and default risk of firms engaged in carbon-heavy industries; (ii) regulatory channel due to potentially higher capital requirements and supervisory pressure; and (iii) uncertainty channel because of increased volatility and risk premia. We have also improved the interpretation of our empirical results by explicitly stating the relationship between methane emissions and banks' capitalization as related to the credit risk channel. The manuscript explicitly claims that high emissions indicate higher involvement of structurally exposed industries, resulting in lower creditworthiness of borrowers, and, hence, higher non-performing loans. As for the regulatory and uncertainty channels, they may play some role but cannot be identified due to the nature of our empirical specification. We are confident that such clarifications would significantly improve the relevance and contribution of the paper.

Q2. The finding that fossil fuel consumption seems to have a positive effect on capitalization is intriguing but not fully explained. It would be helpful to explore this further, perhaps by considering that banks might benefit in the short term from financing fossil fuels, even though this poses bigger risks as the world shifts to cleaner energy.

A2. We would like to thank the reviewer for providing an important observation, which greatly contributed to our understanding of the interpretation of the positive relationship found between fossil fuel consumption and bank capitalization. To address this point, we have added a further economic interpretation of our findings to Section 4. Namely, in our opinion, the positive FOSS coefficient should be interpreted using a short-run vs. long-run trade-off approach. On the one hand, in short- to medium-term runs, fossil fuel industries are known to have a well-established industry structure, stable cash flows, and high collateral value. Thus, banks' exposures to this industry may be considered less risky and even more profitable than those of other industries, which, in turn, could help them maintain a relatively high capital position. However, we would also like to note that the mentioned result demonstrates the trade-off between short-term financial stability and long-term goals. Indeed, while beneficial for banks in the short run, their involvement in financing fossil fuel activities becomes increasingly risky in the long run due to stricter regulation, technological advancements, etc. We hope that these additions clarify our findings considerably.

Q3. Using a fixed-effects model is a good choice, but the paper should also consider whether there might be reverse causality. Could it be that banks with higher capitalization are more likely to choose certain energy financing options?

A3. The reviewers are correct about the possible threat of reverse causality that the study may face. The authors agree with the reviewer's observations and have amended the manuscript to make this limitation clear. Namely, in section four, the authors now explicitly mention that while the fixed-effects model controls for unobserved country-specific heterogeneity, it cannot eliminate all endogeneity concerns, such as reverse causality. Thus, it is mentioned that more capitalized banking systems may be better capable of financing the corresponding sectors of the energy industry. Hence, the estimation can only reveal the conditional relations between the variables rather than their causal relationship. In addition, the authors would like to note that most of the environmental and energy indicators used in the regression analysis reflect the structural properties of national economies rather than the factors that can be directly influenced by the levels of bank capitalization. It further justifies the use of the research design presented in the paper. Finally, the limitation has been discussed in the conclusion of the manuscript as a possible idea for future research. In particular, dynamic panel models and instrumental variable methods may provide a solution to this problem.

 

Q4. The abstract mentions that some variables might be missing, but it would strengthen the paper to specify which ones, such as GDP growth, inflation or regulatory capital requirements.

A4. We would like to thank the reviewer for making this constructive comment. As a result, the abstract was rewritten to explicitly state what type of omitted variables can affect the capitalization of banks. In particular, the revised abstract includes several macroeconomic and institutional determinants such as GDP growth rate, inflation rate, interest rate, and capital requirements. Additionally, the final part of the paper was revised to specify that such variables are not included in the regression model and therefore their influence on the results should be considered in this way. Furthermore, it might be useful for future research to include such variables as control variables in order to make the analysis more accurate.

Q5. The use of K-Means clustering is interesting, but the paper should explain why the chosen number of clusters was selected and include checks to ensure the results are robust, like silhouette scores or trying other clustering methods.

A5. The authors thank the reviewer for their insightful comments about this point. It has helped us significantly improve our approach to analyzing K-Means clusters. In the updated version of the paper, we provide a comprehensive rationale behind choosing the number of clusters. Namely, we test the K-Means solutions for a wide array of numbers of clusters (k = 2-15), applying three different criteria for validating their adequacy: silhouette score, Calinski-Harabasz index, and Bayesian Information Criterion (BIC). As it turned out, higher values of k (13-15) provide the best results concerning all of these measures. However, the resulting solution proves to be too fragmented and less economically interpretable. At the same time, lower values of k fail to catch the heterogeneity that exists within the sample. Thus, we chose k = 10 as a compromise between statistical fit and economic interpretability. To make our methodology even more robust, we perform sensitivity tests using several other clustering algorithms (density based, fuzzy, hierarchical, model-based, and Random Forest) and compare their performance against K-Means. To ensure comparability of the results obtained via different measures, we use several internal validation criteria: Dunn Index, Entropy, and the Calinski-Harabasz index. Based on the findings, K-Means can boast the highest balance between global variance explanation and information content. At last, we emphasize that the number of clusters was selected as a part of the robustness analysis, confirming consistency of the K-Means solution.

Q6. Applying Random Forest and KNN models adds value, but the paper should present their performance clearly, including metrics like RMSE, R² and which variables were most important.

A6. Thank you for your valuable suggestions and comments, which helped improve significantly the transparency of our ML approach and analysis. In the new manuscript version, the performance of all models used was clearly demonstrated through the metrics used, including RMSE, MAE, MSE, MAPE, and R². Importantly, we have included original and normalized metrics for all models to demonstrate their performance and comparability across models. Thus, the reader can clearly estimate the accuracy of the models used. Based on the results, it is clear that there is a difference in the performance of the best models. So, the K-Nearest Neighbors algorithm produces the smallest prediction errors in all error metrics (RMSE, MAE, and MSE). It indicates that the algorithm has the best predictive ability. In its turn, the Random Forest regression demonstrates the largest R² value, which shows that the algorithm can better explain the dependent variable. Furthermore, we have included an analysis of the importance of each variable for the prediction in our revised manuscript. We applied several importance variables, including mean decrease in accuracy, total increase in node purity, and mean dropout loss. According to all three measures, energy dependence (ENIM), renewable energy consumption (RENC), and air pollution (PM2.5) appear to be the most important variables affecting bank capitalization.

 

Q7. Since OECD countries vary widely, the paper should discuss whether the results are mainly influenced by large economies like the US and Germany or if smaller countries have a different impact.

A7. First of all, we would like to thank the reviewer for such an insightful comment concerning the impact of large OECD economies on the results obtained through empirical analyses. With respect to this issue, we performed an additional robustness test aimed at evaluating the effect that large economies might exert on the results we found using baseline data. Specifically, we excluded five largest economies in OECD—the USA, Germany, Japan, the UK, and France—to see whether the results would remain stable in terms of qualitative and quantitative characteristics. As for the findings of the test, the obtained results prove that they can be considered consistent. All the variables under discussion were characterized by the same signs and practically identical magnitudes in relation to the baseline regression. At the same time, it was revealed that the coefficient significance in regard to the main independent variables did not vary in the two regressions. The only distinction concerned the fossil fuel variable, which lost its significance in the sample excluding the most influential countries, although its sign remained positive. The findings presented above prove that there was no need to fear that results of the research are distorted by large economies, because they proved to be structurally consistent in both cases. The results obtained during additional testing were introduced to the final version of the manuscript, where the issue of cross-country heterogeneity was also discussed and a new robustness table was provided.

 

Q8. The positive link between renewable energy and bank capitalization is encouraging, but it’s not clear if one causes the other. This might simply reflect that countries investing in renewables also have stronger economies overall.

A8. In this regard, we would like to thank the reviewer for this very valuable insight on the possible existence of endogeneity bias within the link between renewable energy consumption and bank capitalization. The specific aspect that we wish to highlight is that of macroeconomic strength having potentially resulted in our observed correlation. Consequently, the baseline specification has been expanded to include macroeconomic control variables such as GDP growth rate and unemployment levels, alongside an institutional measure of government effectiveness. The inclusion of these variables is motivated by their ability to account for general economic and institutional conditions, as well as labor market dynamics that can affect both the adoption of renewables and banks’ ability to remain stable. Specifically, results from the estimation using these control variables reveal that the sign of the parameter on renewable energy consumption remains unchanged from what was found under the baseline specification. The coefficient value is virtually unaffected as well and hence shows that our results are unlikely to have arisen as the consequence of macroeconomic effects alone. However, we would still like to stress the point made by the reviewer in the sense that we cannot strictly claim to have established any causality. As such, the revised version explicitly indicates that the findings only constitute conditionally significant relationships between the variables.

Q9. The machine learning findings should be connected with the econometric results to provide a more integrated analysis rather than treating them as separate.

A9. Many thanks for your valuable insight into our paper. We agree that there is a need for us to articulate a broader and more coherent perspective integrating our econometric and machine learning findings. As such, we have made explicit in the revision of the manuscript that our findings from the two analytical techniques are linked. Our econometric findings in section 4 present statistically significant conclusions about the relationships between the environmental variables, energy structure, and the capitalization of banks, revealing strong linear relationships in a fixed effect framework. Meanwhile, the machine learning analysis in section 6 is presented as an important complement to our findings since it identifies the non-linearities, interactions, and predictor variable ranking not possible to obtain from the econometric model. We make explicit that there is a relationship between the findings from the two techniques with respect to both consistency and interpretability. First, the variables considered statistically significant using the econometric technique, including renewable energy consumption, methane emissions, air pollution, and energy dependence, re-emerge as key predictors using the machine learning model. At the same time, while the panel data model presents information on conditional average relationships, the machine learning framework complements this by capturing non-linearities and regime variations consistent with our clustering analysis. Additionally, we have re-articulated our discussion of the econometric, clustering, and machine learning models as an overall framework instead of separate analytical tools in Section 7. This way, we can say that we use a triangulation approach where we apply econometrics to conduct inference, clustering to identify structural heterogeneities, and machine learning to capture non-linearity and predictability.

Q10. The paper rightly highlights the importance of macroprudential policies but the recommendations could be more specific. For example, suggesting how climate stress tests could be incorporated into Basel III/IV rules or how capital buffers might be adjusted based on a country’s energy profile would make the advice more actionable.

A10. Our thanks go to the reviewer for this perceptive comment and suggestions as to how one can make the policy implications more practical. In the revised version, the policy implications have received considerable improvement owing to the inclusion of a subsection that sets out explicit macroprudential policy recommendations based on the empirical findings. The specific ways to include climate risks in the Basel III/IV regulations are explained below. To begin with, climate-related stress testing should be included in supervisory procedures by expanding the traditional Basel stress tests to include the effects of carbon price shocks, regulatory tightening on high-emission industries, and negative physical shocks caused by environmental degradation. This suggestion is supported by our empirical findings, which show that methane emissions, air pollution, and energy dependence affect bank capital. Secondly, there can be climate-adjusted capital buffers. We recommend that the capital buffer requirements be determined depending on whether a country uses fossil fuels for energy generation and whether its economy relies heavily on energy imports. For instance, banking systems that function in countries whose economies have a high dependence on fossil fuel and imported energy might require higher capital buffers compared to systems in countries that have a high penetration rate of renewables. Thirdly, there could be climate-adjusted risk weights under Basel regulations. As is indicated in our empirical results, carbon-intensive industries should be assigned higher risk weights. On the other hand, environmentally friendly investments would attract lower capital charges compared to their carbon-intensive counterparts.

 

Q11. The apparent contradiction where fossil fuel financing seems to support resilience should be linked to regulatory challenges. Banks might look stable now but could be building up risks related to the energy transition that will surface later.

A11. We appreciate the reviewer's constructive point that helped us to understand the results more precisely and connect them with the regulatory issues. In the revised version of our paper, we focus on the contradiction associated with the finding of increased bank capitalization resulting from greater fossil fuel exposure. First, we formulate it as an "apparent contradiction" and discuss its meaning in the context of financial regulation. Specifically, we argue that the connection between higher fossil fuel use and increased capitalization of banks is likely caused by their relatively good profitability, maturity, and value of collateral that allows banks to have good short- and medium-term prospects of profitability and safety. However, we also note that such positive effects can mask the long-term risks of transitioning to cleaner energy sources, asset revaluation, and creation of stranded assets that can arise due to climate policies and technological innovations. More importantly, in the revised version of our paper, we discuss the implications of this contradiction for regulatory practice. In particular, we highlight that banks are likely to be sufficiently capitalized according to traditional approaches and frameworks while having high exposures to carbon-intensive sectors. This suggests the presence of a regulatory blind spot and the need to incorporate transition risks into prudential supervision.

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript addresses a relevant and timely topic linking environmental factors, energy structure and banking sector resilience. However, the empirical strategy requires substantial strengthening. The current analysis relies on correlations without adequately addressing endogeneity or omitted variable bias, which limits causal interpretation. The combination of econometric, clustering, and machine learning approaches is not fully integrated and would benefit from clearer methodological justification. The theoretical framework should be sharpened with explicit hypotheses and clearer economic mechanisms. Additionally, robustness checks (such as alternative specifications, additional controls and sensitivity analyses) are needed. The interpretation of results should be more cautious and less normative. Improving clarity and conciseness of the exposition would also enhance the manuscript.

Author Response

Point to Point Answers to Reviewer 2

Q1. The manuscript addresses a relevant and timely topic linking environmental factors, energy structure and banking sector resilience.

A1. Thanks dear reviewer.  

Q2. However, the empirical strategy requires substantial strengthening.

A2. We thank the reviewer for raising this essential point about the overall manuscript’s empirical strategy. To address the reviewer's comment, we have made improvements to the manuscript's empirical approach from several angles. First, we have enhanced our empirical approach to account for endogeneity and omitted variable bias by estimating the lagged versions of regressors (lagged to avoid reverse causality) and including additional controls such as macroeconomic controls (e.g., GDP growth rate) and institutional ones (e.g., government effectiveness, unemployment). By incorporating these aspects, we make the interpretation of the findings more robust, considering that the results should be viewed as conditional relationships, not causal effects. Second, we added new robustness checks, which include various model specifications (alternative regressions), the inclusion of additional controls (such as those mentioned above), estimation using lagged regressors, and excluding the largest OECD countries from the sample. Overall, the results prove to be quite stable across different specifications. Third, we have enriched our methodological framework and enhanced its cohesiveness by providing an explanation and discussion of how our three approaches (panel econometrics, clustering analysis, and machine learning) fit into one another. We included a subsection where we explain and discuss the role played by each of the approaches. More precisely, panel econometrics offers statistical inference, clustering analysis identifies structural heterogeneity across countries, while machine learning adds some value in terms of variable importance and other insights. Fourth, we have enhanced the theoretical framework by formulating four hypotheses (H1–H4) and explaining the economic mechanisms underpinning the relationship between environment/energy-related variables and the stability of the banking system. This allows us to strengthen the link between the theory and the empirical findings. Fifth, we have revised the interpretation of the results section of our manuscript by making the presentation of the results more cautious and clear (and avoiding making normative statements).

Q3. The current analysis relies on correlations without adequately addressing endogeneity or omitted variable bias, which limits causal interpretation.

A3. We agree with the reviewer in highlighting the issue of endogeneity and omitted variable bias. While the baseline model is useful in understanding conditional relationships, there might be problems related to reverse causality and omitted variables. In order to strengthen our identification strategy, we conduct some auxiliary tests in the updated version of the paper. In the first place, we estimate lagged versions of our main regressors to allow environmental and energy variables to occur before bank capitalization. This helps alleviate potential issues associated with reverse causality and makes the relationship more credible from this point of view. The second auxiliary test involves the use of additional macroeconomic and institutional control variables such as GDP growth rate, government effectiveness, and unemployment rate. They reflect general economic situation and institutional conditions which are likely to affect estimated coefficients. Notwithstanding this addition, the estimates stay extremely stable across specifications and the coefficient on renewable energy consumption is positive and statistically significant. Finally, we clearly state at every opportunity that the results should be considered as robust conditional relationships and are not necessarily causal relationships. In sum, while we do not make claims about causal identification, our auxiliary analyses help address the issues of endogeneity and omitted variable bias raised by the reviewer.

Q4. The combination of econometric, clustering, and machine learning approaches is not fully integrated and would benefit from clearer methodological justification.

A4. We are deeply grateful for the reviewer's insightful observation regarding the incorporation of various methodological perspectives. Indeed, in the revision of the paper, the authors made efforts to further enhance their justification for applying a combination of methods and make the integration of econometrics, clustering, and machine learning more consistent. First of all, we included an additional subsection into the methodological section of the paper, where the combination of methods was clearly defined and described. This subsection emphasizes the complementary role of each approach in relation to others and highlights the particular features of each method: while panel econometrics is applied to estimate conditional relationships and control for heterogeneity, clustering helps to capture cross-country differences and identify structural regimes, and machine learning allows us to get a new perspective on the results and check the importance of variables and possible non-linearities in the relationships under consideration. Furthermore, it was decided to make the links between different results more explicit. Thus, we now explain how the countries' regimes identified by clustering are related to econometric results: in our interpretation, the results obtained in each country regime show the same relationship between energy consumption and financial markets, but its nature may differ significantly depending on country characteristics and economic environment. Also, the results obtained in the machine learning part of the analysis are used as a validation step of the main determinants discovered by econometrics. Finally, we would like to underline that these three methods were not applied separately from one another. Instead, the combination of econometric, clustering, and machine learning methods can be considered as a triangulation, which takes account of statistical inference, structural heterogeneity, and predictive performance dimensions of the problem.

Q5. The theoretical framework should be sharpened with explicit hypotheses and clearer economic mechanisms.

A5. We acknowledge and appreciate the insightful comments of the reviewer in this respect. In our revised version of the manuscript, we have significantly improved the theoretical framework by formulating our hypotheses and elaborating on the economic mechanisms underpinning the relationship between environmental and energy issues and bank resilience. In particular, we added a clear list of hypotheses to be tested (H1–H4) at the end of the Introduction section. Hypotheses clarify the nature of the links among such environmental variables as methane gas emissions and air pollution, energy dependence as captured by net energy imports, energy transition dynamics reflected in renewable energy consumption, dependence on fossil fuels, and how each of them affects bank capitalization. Second, we have considerably expanded our discussion of the economic mechanisms by which environmental and energy-related variables can affect banking sector resilience. There are three major channels that have been discussed in detail: credit risk channel, whereby the adverse impact of the environment on borrower performance leads to increased risk of default; macroeconomic channel, whereby the negative consequences of energy dependence and environmental damage result in increased volatility and uncertainty in the economy; and transition risk channel, resulting from tighter regulation and technological innovations that lead to a revaluation of carbon-based assets. Third, we have enhanced the consistency between the theoretical part and empirical analysis, where the nature of the econometric model and its implications have been clarified.

Q6. Additionally, robustness checks (such as alternative specifications, additional controls and sensitivity analyses) are needed.

A6. The authors thank the reviewer for highlighting the importance of improving the empirical methodology by conducting more robustness tests. We include in our updated manuscript a series of supplementary robustness checks to account for possible issues related to sample selection, omitted variable bias, and endogeneity. First, we assess whether the results are affected by the presence of a few dominant economies. In particular, we estimate the model while excluding the top five economies among the OECD countries (US, Germany, Japan, UK, and France). The results show that the findings hold with respect to the signs and significance levels of the estimated parameters, suggesting that the results are not due to influential observations but rather represent general trends within the OECD group. Second, to examine the possibility of omitted variable bias, we extend the baseline model by adding some macroeconomic and institutional indicators, such as GDP growth, government effectiveness, and unemployment rates. As expected, the three independent variables of interest (RENC, ENIM, and PM2.5) prove to be resilient to inclusion of new covariates, showing their robustness both in magnitude and statistical significance. Finally, to address the concerns about reverse causality and simultaneous equations, we consider a lagged specification, where regressors are lagged relative to the response variable, bank capitalization. As expected, the primary relationships hold, with lagged regressors demonstrating stable and statistically significant associations with bank capitalization. Overall, these alternative specifications show considerable consistency, providing strong support for the findings' robustness to different modeling techniques and identification strategies. These robustness tests add substantial value to the empirical approach proposed in our study and directly respond to the reviewer's recommendations.

Q7. The interpretation of results should be more cautious and less normative. Improving clarity and conciseness of the exposition would also enhance the manuscript.

A7. We thank the reviewer for his very valuable and constructive comments. As a result of that, we carried out an extensive revision of the manuscript by adopting a more prudent and less normative interpretation of our results while improving overall clarity and conciseness. First, we have changed our interpretations of the empirical results. Specifically, phrases like ‘leads to,’ ‘implies,’ and other that would suggest any kind of strong causal connection were changed to softer versions like ‘is associated with,’ ‘may reflect’ and ‘is consistent with,’ etc. We have also explicitly mentioned in the manuscript that all our results should be considered in terms of conditional association due to possible existence of endogeneity issues, omitted variable problems and/or reverse causality. Second, this approach was adopted consistently within different parts of our paper, including introduction, empirical findings, robustness tests, cluster analysis, and machine learning part. Particularly, when discussing clustering results we do not consider clusters as different structural regimes but as groups of countries having some descriptive features. At the same time, we provided additional warnings regarding overstatement of the results of the analysis of clusters and regime switching. Third, we improved readability of the paper by making it more concise and clear without reducing our main message and arguments.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors,

Thank you for your thorough and thoughtful revisions. I appreciate the way you have addressed each of the comments with clarity and rigor.

The manuscript now provides stronger explanations of risk transmission channels, clearer interpretations of fossil fuel and renewable energy findings, robust methodological checks and more actionable policy recommendations. The integration of econometric and machine learning results is particularly effective in strengthening the overall framework.

I am satisfied that my concerns have been adequately addressed and I believe the paper has been significantly improved.

Best Wishes!

Reviewer

Author Response

Point to Point Answer to Reviewer 1

 

Q1. Dear Authors, Thank you for your thorough and thoughtful revisions. I appreciate the way you have addressed each of the comments with clarity and rigor. The manuscript now provides stronger explanations of risk transmission channels, clearer interpretations of fossil fuel and renewable energy findings, robust methodological checks and more actionable policy recommendations. The integration of econometric and machine learning results is particularly effective in strengthening the overall framework. I am satisfied that my concerns have been adequately addressed and I believe the paper has been significantly improved. Best Wishes!

 

A1. Thanks dear reviewer.

 

Reviewer 2 Report

Comments and Suggestions for Authors

The revised paper shows a substantial improvement compared to the previous version. The authors strengthened the methodological framework, by adding robustness checks, macroeconomic controls and lagged specifications, while also adopting a more cautious interpretation of the empirical findings. The integration of panel econometrics, clustering analysis and machine learning provides an interesting multidimensional perspective on the climate, energy, finance nexus. The empirical results are generally consistent across specifications and the discussion of limitations and endogeneity concerns is appreciated. The paper contributes to the growing literature on environmental risks, energy transition and banking sector resilience. However, some issues still deserve attention before publication. First, the manuscript remains excessively long and repetitive in some sections, particularly in the interpretation of robustness checks. Second, the English language and stylistic quality could be further improved for clarity and readability. Third, the clustering and machine-learning sections should be more explicitly connected to the core econometric findings in order to strengthen the overall narrative coherence. Finally, the authors could consistently avoid causal terminology such as “impact” or “effect” unless causal identification is formally supported. Overall, the manuscript is interesting and publishable after minor revisions.

Author Response

Point to Point Answers to Reviewer 2

Q1. The revised paper shows a substantial improvement compared to the previous version. The authors strengthened the methodological framework, by adding robustness checks, macroeconomic controls and lagged specifications, while also adopting a more cautious interpretation of the empirical findings. The integration of panel econometrics, clustering analysis and machine learning provides an interesting multidimensional perspective on the climate, energy, finance nexus. The empirical results are generally consistent across specifications and the discussion of limitations and endogeneity concerns is appreciated. The paper contributes to the growing literature on environmental risks, energy transition and banking sector resilience.

A1. Thanks dear reviewer.

Q2. However, some issues still deserve attention before publication. First, the manuscript remains excessively long and repetitive in some sections, particularly in the interpretation of robustness checks.

A2. We appreciate the reviewer’s insightful comments. As a result, we have extensively edited our paper to eliminate any unnecessary duplication and improve readability, with specific focus on the robustness checks’ interpretation. In terms of content, Sections 4.1, 4.2, and 4.3 have been significantly reduced and reformatted into a more streamlined format. Any redundant information about methodologies and coefficient stability was either deleted or summarized. The current section highlights the reason for each robustness check, significant variations in the coefficients, and conclusions drawn from the results. Moreover, the concluding part of the robustness tests was reduced to prevent duplicative presentation of the results discussed in each of the previous sections. All in all, this paper underwent major changes to eliminate any redundancy and strengthen the paper’s readability while maintaining its key methodological and empirical contributions.

Q3. Second, the English language and stylistic quality could be further improved for clarity and readability.

 

A3.  This is a very important remark, which we are thankful for, as our manuscript was edited and improved regarding usage of English and stylistics. In particular, the Introduction section and the explanation of the results obtained empirically were substantially improved through:

• reduction of too long and complicated sentences;
• minimization of repetitions and unnecessary transitional elements;
• improvement of the flow of the paragraphs to make them readable;
• clarification of imprecise sentences and improvement of their wording;
• making the terminology consistent throughout the manuscript;
• simplification of some methodological and interpretative parts.

The part about the robustness check was also made shorter and clearer, without repeating the explanation of what has already been said. Therefore, our manuscript has been edited for clarity and readability.

Q4. Third, the clustering and machine-learning sections should be more explicitly connected to the core econometric findings in order to strengthen the overall narrative coherence.

 

A4.  The authors are grateful for this valuable comment. The manuscript has been re-written to increase narrative coherence between the three analytical parts (the econometrics, clustering analysis, and machine learning). Firstly, the authors emphasize in the text that the panel regression analysis serves as the central tool in estimating the major relationships among environmental factors, energy structure, and bank capitalization. Secondly, the connection between the clustering analysis and the results obtained using econometrics is specified, emphasizing that the former serves as a means of assessing heterogeneity across countries and establishing country types based on the relationships. Moreover, the section concerning machine learning has been improved to highlight its contribution to the empirical framework. The authors emphasize that machine learning models allow evaluating the importance of variables identified using econometrics and searching for non-linear effects missed by panel regressions. In order to make the narrative more coherent, transition paragraphs were added at the beginning of Sections 4, 5, 6, and 7, and the integrated discussion section was improved.

Q5. Finally, the authors could consistently avoid causal terminology such as “impact” or “effect” unless causal identification is formally supported.

A5. Many thanks for this important comment. It has led us to revise the manuscript so that causal language is not used unless a causal claim can be firmly made on the basis of the empirical framework. In particular, causal language such as “impact,” “effect,” “determine,” “influence,” “support,” and similar terms have been changed consistently through the entire manuscript, replacing them with associative language such as “association,” “relationship,” “linked to,” “related to,” and “predictive relevance.” This has been done consistently for all parts of the manuscript: Introduction, Literature Review, Econometric Results, Robustness Tests, Clustering Analysis, Machine Learning section, and summary tables. Furthermore, we have clearly stated in both methodological and robustness sections that panel regressions establish only a conditional relationship and not a causal effect, as causal identification cannot be established from the given empirical framework.

Q6. Overall, the manuscript is interesting and publishable after minor revisions.

A6. Thanks dear reviewer.