Cryptocurrency Futures Portfolio Trading System Using Reinforcement Learning

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript titled "Cryptocurrency Futures Portfolio Trading System Using Reinforcement Learning" presents a timely and relevant contribution to the field of algorithmic trading and reinforcement learning applications in financial markets. The integration of A2C reinforcement learning with statistical techniques such as ANOVA for portfolio construction is well-motivated and clearly explained. The empirical application on cryptocurrency futures data (Binance) is particularly interesting given the current importance of this asset class.

-Although the authors claim novelty in combining A2C and ANOVA for portfolio optimization, further clarification is needed on how this combination improves over existing reinforcement learning-based portfolio systems. A more explicit comparison with state-of-the-art methods (DDPG, PPO, or ensemble RL) would strengthen the contribution.

-Please better differentiate your work from closely related studies such as Jiang & Liang (2017) or Liang et al. (2018), which also used deep reinforcement learning for portfolio optimization.

-The paper uses data from Binance and Glassnode, but the criteria for selecting the final 18 cryptocurrencies are not entirely clear. Were all low-liquidity or stablecoins excluded? Why?

-Please specify the source and nature of the on-chain data used. If only circulating supply is used, discuss its limitations and the rationale behind this choice.

-The choice of the A2C algorithm is justified, but no sensitivity or robustness tests are provided regarding the learning rate, discount factor, or other hyperparameters. Including an ablation study or sensitivity analysis would add value.

- Could the authors briefly explain how overfitting was avoided in the training process? This is important given the high risk of overfitting in financial time series.

-The ANOVA approach used to select assets based on timeframe performance is novel. However, this method assumes normality and homogeneity of variances. Please confirm if these assumptions were tested.

-The benchmark used (average interest rate of 2.389%) seems relatively low. Why not compare with a standard market index or a buy-and-hold strategy?

-The use of multiple tests (ANOVA, Dunn's post-hoc, t-tests) is appropriate, but corrections for multiple comparisons (e.g., Bonferroni or Holm) are not mentioned. Please clarify.

-It would be helpful to report confidence intervals (in addition to p-values) for key performance results.

-The manuscript would benefit from proofreading to improve English clarity and avoid redundancy (repetition in Sections 2.2 and 2.3).

-Figures and tables are informative but could be better integrated into the discussion. For instance, Figure 1 (CPTS framework) could be more detailed to highlight the data flow and agent-environment interactions.

Author Response

Comments 1: [Although the authors claim novelty in combining A2C and ANOVA for portfolio optimization, further clarification is needed on how this combination improves over existing reinforcement learning-based portfolio systems. A more explicit comparison with state-of-the-art methods (DDPG, PPO, or ensemble RL) would strengthen the contribution.]

Response 1: [Thank you for this valuable suggestion. We acknowledge that a comprehensive comparison with other RL algorithms would indeed strengthen our contribution. However, due to computational resource constraints and the scope of this initial study, we focused on demonstrating the effectiveness of the A2C-ANOVA combination as a proof-of-concept. We have revised Section 2.2 to better justify our choice of A2C and have added a discussion in the "Limitations and Future Work" section (Section 5, paragraph 2) acknowledging that future research should include comparative studies with DDPG, PPO, and other state-of-the-art methods.]

“[Subsequent research efforts should also focus on conducting comprehensive comparisons with other reinforcement learning algorithms such as DDPG, PPO, and ensemble methods, along with extended training and testing periods.]”

Comments 2: [Please better differentiate your work from closely related studies such as Jiang & Liang (2017) or Liang et al. (2018), which also used deep reinforcement learning for portfolio optimization.]

Response 2: [We have significantly enhanced the literature review in Section 2.3, paragraph 3, to better distinguish our work. The key differentiation lies in our explicit use of ANOVA to analyze timeframe-dependent performance and construct portfolios based on statistical significance of returns across different trading frequencies. Previous studies did not systematically investigate or statistically validate the impact of different timeframes on portfolio performance.]

“[While these studies demonstrate the potential of RL in cryptocurrency portfolio management, they do not explicitly consider the impact of different trading timeframes on portfolio performance. Our study differs from these works by employing ANOVA to statistically analyze the performance of cryptocurrencies across different timeframes and construct portfolios based on these findings.]”

Comments 3: [The paper uses data from Binance and Glassnode, but the criteria for selecting the final 18 cryptocurrencies are not entirely clear. Were all low-liquidity or stablecoins excluded? Why?]

Response 3: [We have clarified the selection criteria in Section 4.1, paragraph 1. We selected the top 20 cryptocurrencies by market capitalization on Binance futures and excluded two due to data availability issues (insufficient historical data for the full study period). Regarding stablecoins, none were present in the top 20 at the time of data collection (January 2022), which naturally excluded them from our analysis. This exclusion is also methodologically justified as stablecoins represent a fundamentally different asset class designed to maintain price stability rather than generate returns through price appreciation. As noted in recent literature, stablecoins are specifically engineered to minimize volatility through pegging mechanisms to fiat currencies or other stable assets (Bullmann et al., 2019; Lyons & Viswanath-Natraj, 2023). Since our study focuses on price prediction and return optimization, the inclusion of assets designed to maintain stable prices would be counterproductive to our research objectives. This distinction aligns with the growing academic consensus that stablecoins should be analyzed separately from traditional cryptocurrencies due to their unique economic characteristics and market functions (Au et al., 2024).]

["We selected the top 20 cryptocurrencies by market capitalization on Binance futures. After excluding two cryptocurrencies with insufficient historical data, we were left with a final set of 18 cryptocurrencies for our experiment. The selected cryptocurrencies are all actively traded on Binance futures and have sufficient liquidity for our analysis. No stablecoins were present in the top 20 at the time of data collection. Additionally, stablecoins are methodologically excluded from price prediction studies as they are specifically designed to maintain price stability rather than generate returns through price appreciation, making them unsuitable for our research objectives."]

Comments 4: [Please specify the source and nature of the on-chain data used. If only circulating supply is used, discuss its limitations and the rationale behind this choice.]

Response 4: [We have clarified in Section 4.1, paragraph 2, that our on-chain data consists solely of USDT circulating supply from Glassnode, which serves as a proxy for market liquidity. The rationale for this macroscopic approach is that different cryptocurrency protocols have fundamentally different on-chain characteristics, making protocol-specific metrics incomparable across our diverse asset portfolio. By focusing on USDT circulating supply, we employ a universal liquidity indicator that affects all cryptocurrencies equally, regardless of their underlying technology.]

["The circulating supply of USDT from Glassnode, which we use as a macroscopic proxy for market liquidity that affects all cryptocurrencies uniformly. Since each cryptocurrency protocol has different characteristics and on-chain metrics that are not directly comparable (e.g., Bitcoin's hash rate has no equivalent in Ethereum or other networks), we adopted this universal approach to ensure methodological consistency across our diverse asset portfolio. However, the utilization of on-chain data is limited. This study considers only the circulating supply of USDT and does not consider various other on-chain indicators, such as active addresses, transaction volumes, and protocol-specific metrics."]

Comments 5: [The choice of the A2C algorithm is justified, but no sensitivity or robustness tests are provided regarding the learning rate, discount factor, or other hyperparameters. Including an ablation study or sensitivity analysis would add value.]

Response 5: [We acknowledge the importance of hyperparameter sensitivity analysis. Due to computational constraints, we conducted limited hyperparameter exploration focusing on the most critical parameters. In Section 3.2, paragraph 1, we have documented our hyperparameter choices and added a discussion in the limitations section about the need for more comprehensive hyperparameter optimization in future work.]

["The hyperparameters used for training are as follows: learning rate = 0.001, discount factor (gamma) = 0.99, and entropy coefficient = 0.01. Future research should include more extensive hyperparameter search for the A2C algorithm to optimize performance further."]

Comments 6: [Could the authors briefly explain how overfitting was avoided in the training process? This is important given the high risk of overfitting in financial time series.]

Response 6: [We have added a detailed explanation in Section 3.2, paragraph 3, describing our overfitting prevention measures: (1) standard training/testing data split, (2) inclusion of volatility and maximum drawdown penalties in the reward function to prevent overly aggressive strategies, and (3) use of LSTM networks to capture temporal dependencies without overfitting to short-term patterns.]

["To mitigate the risk of overfitting, we employed a standard training/testing split of the data. The model is trained on the training set, and its performance is evaluated on the unseen test set. The inclusion of penalties for volatility and maximum drawdown in the reward function also helps to prevent the agent from learning overly aggressive and potentially overfitted strategies."]

Comments 7: [The ANOVA approach used to select assets based on timeframe performance is novel. However, this method assumes normality and homogeneity of variances. Please confirm if these assumptions were tested.]

Response 7: [We have added clarification in Section 4.2, paragraph 3, confirming that we tested the assumptions of normality using the Shapiro-Wilk test and homogeneity of variances using Levene's test. When these assumptions were violated, we employed non-parametric alternatives (Friedman test and Dunn's post-hoc test).]

["Before conducting the ANOVA, we verified the assumptions of normality and homogeneity of variances using the Shapiro-Wilk and Levene's tests, respectively. As some of the data did not meet these assumptions, we also employed the non-parametric Friedman test and Dunn's post-hoc test."]

Comments 8: [The benchmark used (average interest rate of 2.389%) seems relatively low. Why not compare with a standard market index or a buy-and-hold strategy?]

Response 8: [The 2.389% benchmark is based on the Bank of Korea's base rate at the time of the study period, representing the risk-free rate of return in the Korean market context. We acknowledge that comparing with buy-and-hold strategies and market indices would provide more meaningful benchmarks for evaluating trading system performance. However, the limitation of having only aggregated return data (RoR) from our ANOVA experiments, without access to the underlying time-series data needed to reconstruct individual trading histories, prevented us from implementing these more comprehensive comparisons.]

["In addition, exceeding the average benchmark interest rate of 2.389% (based on the Bank of Korea's base rate) during the training period was considered a criterion for portfolio construction. We use the average interest rate as a benchmark as it represents the risk-free rate of return in our market context. We acknowledge that comparisons with buy-and-hold strategies and cryptocurrency market indices would provide more comprehensive performance evaluation but were unable to conduct these comparisons due to data limitations in our current study design."]

Comments 9: [The use of multiple tests (ANOVA, Dunn's post-hoc, t-tests) is appropriate, but corrections for multiple comparisons (e.g., Bonferroni or Holm) are not mentioned. Please clarify.]

Response 9: [We have clarified in Section 4.2, paragraph 3, and Table 4 that we applied the Bonferroni correction for multiple comparisons in our post-hoc analysis to control for Type I error inflation. Dunn's post-hoc test with Bonferroni correction for multiple comparisons.]

Comments 10: [It would be helpful to report confidence intervals (in addition to p-values) for key performance results.]

Response 10: [We have added 95% confidence intervals for the mean ROR of both frequency groups in Table 9, Section 4.2, final paragraph.]

["The 95% confidence intervals for the mean ROR are also reported."]

Comments 11: [The manuscript would benefit from proofreading to improve English clarity and avoid redundancy (repetition in Sections 2.2 and 2.3).]

Response 11: [We have thoroughly proofread the entire manuscript and eliminated redundant content, particularly in Sections 2.2 and 2.3, to improve clarity and flow. Specifically, we have:

1. Streamlined Section 2.2 to focus on general RL concepts and methodology, reducing repetitive discussions of cryptocurrency applications
2. Enhanced Section 2.3 with explicit connections to Section 2.2 using phrases such as "Building upon the RL foundations discussed in Section 2.2"
3. Consolidated redundant references and varied expressions to eliminate repetitive phrasing]

Comments 12: [Figures and tables are informative but could be better integrated into the discussion. For instance, Figure 1 (CPTS framework) could be more detailed to highlight the data flow and agent-environment interactions.]

Response 12: [We have improved the integration of tables and figures into the narrative throughout the manuscript. While maintaining the current Figure 1 structure, we have enhanced the textual description of the CPTS framework in Section 3 to provide clearer explanation of data flow and agent-environment interactions within the three-stage process.]

Reviewer 2 Report

Comments and Suggestions for Authors

This study sets out to design a Cryptocurrency Portfolio Trading System (CPTS) that utilizes reinforcement learning (RL) and timeframe analysis to optimize trading strategies in the cryptocurrency futures market. It aims to address the challenges of high volatility and nonlinear dynamics in cryptocurrency markets through algorithmic learning and statistical testing. Results show strong in-sample (training) performance for high-frequency portfolios, but the out-of-sample (testing) results heavily favor low-frequency portfolios. ANOVA confirms statistical significance in timeframe-based performance. 
Perhaps a better understanding of these results would be gained by considering that several dynamic characteristics, such as the Hurst exponent or the distribution of return fluctuations, also depend on the sampling frequency of price changes (see "Multiscale characteristics of the emerging global cryptocurrency market" in https://doi.org/10.1016/j.physrep.2020.10.005). There might be some connection. Perhaps, for example, if time were scaled to the average number of transactions per unit of time characteristic for a given cryptocurrency and such a "market time" were used, the results would become similar? Such a discussion would be highly recommended here.

 

Author Response

Comments 1: [Perhaps a better understanding of these results would be gained by considering that several dynamic characteristics, such as the Hurst exponent or the distribution of return fluctuations, also depend on the sampling frequency of price changes. There might be some connection. Perhaps, for example, if time were scaled to the average number of transactions per unit of time characteristic for a given cryptocurrency and such a "market time" were used, the results would become similar? Such a discussion would be highly recommended here.]

Response 1: [Thank you for this insightful comment regarding the concept of "market time" and its potential relationship to our timeframe-dependent results. This is indeed a fascinating research direction that could provide deeper insights into why different timeframes exhibit varying performance characteristics. We have added a discussion of this concept in Section 5, paragraph 2, acknowledging that future research should explore the relationship between sampling frequency, market microstructure characteristics, and performance outcomes using market time scaling approaches.]

["Additionally, exploring the concept of 'market time' could provide valuable insights into our timeframe-dependent results. The relationship between sampling frequency, market microstructure characteristics such as the Hurst exponent, and performance outcomes warrants investigation using market time scaling approaches. "]

Reviewer 3 Report

Comments and Suggestions for Authors

The paper introduces the CPTS framework for cryptocurrency futures trading by incorporating reinforcement learning in the development of strategies. The topic is timely and of considerable interest academically and practically in the realms of algorithmic trading and cryptocurrency markets. The method is well-modeled, and the results present some preliminary evidence of the potential of the framework. However, certain methodological improvements are warranted to improve the robustness and interpretability of the results. My comments concern aspects relating to improved evaluation metrics, comparative validity, followed by dataset extension in future work.

1. The paper is evaluated mainly based on the return (%). However, returns alone do not take into account the corresponding risk or volatility of the returns. In a highly volatile market such as cryptocurrency futures, two strategies that have the same return could differ significantly in risk exposure. I think the paper could supplement this by including the Sharpe ratio, which can serve to measure risk-adjusted return and thus provide more rounded performance information, which values a much more equitable and informative assessment between strategies.
2. The current evaluation lacks a direct comparison with underlying trading strategies. For example, a buy-and-hold strategy as a reference would certainly provide valuable context for interpreting the results. This comparison is important to determine whether the CPTS framework performs well relative to the passive approach of simply investing money; thus, further increasing the credibility of the obtained performance.
3. The authors acknowledged the limitation of their on-chain data set (basically just the circulating supply of USDT) in the conclusion section of their paper. While this acknowledgement is valuable, I suggest that future research consider an expanded on-chain data set, such as active addresses, transaction volume, hash rate, and cryptocurrency activity. These additional indicators would help increase the predictive power of the model and provide a more comprehensive overview of cryptocurrency market dynamics.
4. Some online resources include the word "[Online]" without specifying the "(accessed on DD Month YYYY)" tag, which is usually required, while URLs appear in two forms: a hyperlink and "Available online:". The capitalization of the title is not consistent: some titles appear in uppercase or lowercase, while others are in uppercase or lowercase in the sentence. Certain conference proceedings do not have the entire phrase "In Proceedings of..." with location, country, and dates. For some books, the edition and publisher location are missing. Punctuation marks appear in some entries and not in others, whether with commas or semicolons.

 

Author Response

Comments 1: [The paper is evaluated mainly based on the return (%). However, returns alone do not take into account the corresponding risk or volatility of the returns. In a highly volatile market such as cryptocurrency futures, two strategies that have the same return could differ significantly in risk exposure. I think the paper could supplement this by including the Sharpe ratio, which can serve to measure risk-adjusted return and thus provide more rounded performance information, which values a much more equitable and informative assessment between strategies.]

Response 1: [Thank you for this excellent and important suggestion. We completely agree that risk-adjusted performance metrics are crucial in highly volatile cryptocurrency markets and represent a significant area for improvement in our evaluation framework. We acknowledge that the evaluation based solely on returns is a notable limitation that does not capture the risk-adjusted performance characteristics that are essential for comprehensive assessment in volatile cryptocurrency markets.]

["Sixth, the evaluation based solely on returns does not capture risk-adjusted performance, which is crucial in volatile cryptocurrency markets. Future research should incorporate comprehensive risk-adjusted performance metrics such as the Sharpe ratio, Sortino ratio, maximum drawdown analysis, and Value at Risk (VaR) to provide more thorough evaluation of trading strategies in highly volatile markets.", “Additionally, implementing comprehensive risk-adjusted performance analysis including the Sharpe ratio and other risk metrics would significantly enhance the evaluation framework.”]

 

Comments 2: [The current evaluation lacks a direct comparison with underlying trading strategies. For example, a buy-and-hold strategy as a reference would certainly provide valuable context for interpreting the results. This comparison is important to determine whether the CPTS framework performs well relative to the passive approach of simply investing money; thus, further increasing the credibility of the obtained performance.]

Response 2: ["We acknowledge this important limitation in our current study. You are absolutely correct that direct comparison with benchmark strategies such as buy-and-hold is essential for demonstrating the practical value of our CPTS framework. However, due to computational constraints and data limitations in our current research setup, we were unable to implement comprehensive buy-and-hold strategy comparisons during this study. We have explicitly acknowledged this limitation in our manuscript in Section 4.2, paragraph 3 and limitation discussion in Section 5, paragraph 2, 5"]

[“We acknowledge that comparisons with buy-and-hold strategies and cryptocurrency market indices would provide more comprehensive performance evaluation but were unable to conduct these comparisons due to data limitations in our current study design.”,” Fourth, due to computational constraints and data limitations, we could not conduct comprehensive comparisons with buy-and-hold strategies or cryptocurrency market indices, which would provide more meaningful benchmarks for evaluating the practical effectiveness of our CPTS framework.”,” Furthermore, future studies should include meaningful benchmarks such as buy-and-hold strategies, equal-weight portfolios, and cryptocurrency market indices to provide more comprehensive performance evaluation and establish the practical value of algorithmic trading approaches.”, ”Furthermore, future studies should include meaningful benchmarks such as buy-and-hold strategies, equal-weight portfolios, and cryptocurrency market indices to provide more comprehensive performance evaluation and establish the practical value of algorithmic trading approaches.”]

 

Comments 3: [The authors acknowledged the limitation of their on-chain data set (basically just the circulating supply of USDT) in the conclusion section of their paper. While this acknowledgement is valuable, I suggest that future research consider an expanded on-chain data set, such as active addresses, transaction volume, hash rate, and cryptocurrency activity. These additional indicators would help increase the predictive power of the model and provide a more comprehensive overview of cryptocurrency market dynamics.]

Response 3: [Thank you for this insightful suggestion. We have significantly expanded our discussion of on-chain data limitations and future research directions in Section 5, paragraph 2, 3.]

["Second, the utilization of on-chain data is significantly limited, as we consider only the circulating supply of USDT and do not incorporate various other critical on-chain indicators. Future research would benefit from incorporating a broader range of on-chain metrics such as active addresses (indicating network usage and adoption), transaction volume and velocity (reflecting network activity), hash rate (for proof-of-work cryptocurrencies, indicating network security), network value to transactions ratio (NVT, serving as a valuation metric), realized market capitalization, and social sentiment indicators from various platforms. These additional on-chain indicators could provide richer insights into cryptocurrency market dynamics, network health, adoption trends, and investor behavior, potentially enhancing the predictive power of our model.", “Future research should address these limitations by conducting training and testing over longer periods, incorporating the broader range of on-chain data indicators mentioned above, and experimenting with a larger and more diverse set of cryptocurrencies.”]

 

Comments 4: [Some online resources include the word "[Online]" without specifying the "(accessed on DD Month YYYY)" tag, which is usually required, while URLs appear in two forms: a hyperlink and "Available online:". The capitalization of the title is not consistent: some titles appear in uppercase or lowercase, while others are in uppercase or lowercase in the sentence. Certain conference proceedings do not have the entire phrase "In Proceedings of..." with location, country, and dates. For some books, the edition and publisher location are missing. Punctuation marks appear in some entries and not in others, whether with commas or semicolons.]

Response 4: [Thank you for pointing out these important formatting inconsistencies. We have completely revised and standardized all 53 references in our bibliography to ensure consistency and compliance with MDPI formatting guidelines.]

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have carefully taken into account the previous remarks and have substantially revised the manuscript. The modifications are clear, relevant, and well-integrated into the overall structure of the paper. The quality of the manuscript has improved, and the authors have adequately addressed the key concerns raised in the earlier review.

Author Response

Comments 1 : The authors have carefully taken into account the previous remarks and have substantially revised the manuscript. The modifications are clear, relevant, and well-integrated into the overall structure of the paper. The quality of the manuscript has improved, and the authors have adequately addressed the key concerns raised in the earlier review.

Response 1 : Thank you for your positive evaluation and thorough review. We have carefully addressed all the comments from the previous review and made comprehensive revisions to strengthen the manuscript. We greatly appreciate your acknowledgment that the changes have been well-integrated and improved the overall quality of the paper. We hope that this revised version meets the high academic standards of your journal, and we will continue striving to produce high-quality research.

Reviewer 2 Report

Comments and Suggestions for Authors

Revisions are satisfactory and thus this manuscript can be accepted for publication in its current form.

Author Response

Comments 1 : Revisions are satisfactory and thus this manuscript can be accepted for publication in its current form.

Response 1 : Thank you very much for your positive assessment and recommendation. We are delighted to hear that the revisions are satisfactory and that the manuscript is considered suitable for publication in its current form. We sincerely appreciate your time, effort, and valuable feedback throughout the review process, which have greatly contributed to improving the quality of our work.

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have adequately addressed the main concerns. Some limitations remain (e.g., lack of benchmark comparisons), but these are at least sufficiently acknowledged and left for future work.
Lastly, a final check should be made to ensure full compliance with MDPI guidelines regarding reference formatting (Some journal names are written in full, like Finance Research Letters or International Review of Financial Analysis, instead of their official abbreviated forms required by the template. Proceedings of conferences also exhibit some inconsistency in their format: whereas some are correctly detailed with publisher information for IEEE (e.g., McNally et al. 2018), others (e.g., Carta et al. 2021) lack this information, including city, country, and pages, which should be consistent for all. The arXiv citations appear as arXiv 2017, arXiv:1701.07274, which is just half correct; for uniformity, the year should always appear, and the term arXiv italicized. When referring to books, some include the publisher and location (for example, Murphy, 1999; Wilder, 1978), whereas others omit the city (i.e., Kaufman, 2013), and it should be added. Finally, most online references (white papers, SSRN, Econstor, GitHub, etc.) are formatted correctly, except a few are lacking an access date, like SSRN 2019 (#7), and certain arXiv papers, which should be updated to comply fully with the style guide).

Author Response

Comments 1 : The authors have adequately addressed the main concerns. Some limitations remain (e.g., lack of benchmark comparisons), but these are at least sufficiently acknowledged and left for future work.
Lastly, a final check should be made to ensure full compliance with MDPI guidelines regarding reference formatting (Some journal names are written in full, like Finance Research Letters or International Review of Financial Analysis, instead of their official abbreviated forms required by the template. Proceedings of conferences also exhibit some inconsistency in their format: whereas some are correctly detailed with publisher information for IEEE (e.g., McNally et al. 2018), others (e.g., Carta et al. 2021) lack this information, including city, country, and pages, which should be consistent for all. The arXiv citations appear as arXiv 2017, arXiv:1701.07274, which is just half correct; for uniformity, the year should always appear, and the term arXiv italicized. When referring to books, some include the publisher and location (for example, Murphy, 1999; Wilder, 1978), whereas others omit the city (i.e., Kaufman, 2013), and it should be added. Finally, most online references (white papers, SSRN, Econstor, GitHub, etc.) are formatted correctly, except a few are lacking an access date, like SSRN 2019 (#7), and certain arXiv papers, which should be updated to comply fully with the style guide).

Response 1 : Thank you for your thorough review and constructive feedback. We appreciate your acknowledgment that the main concerns have been adequately addressed. In response to your comments regarding reference formatting, we have carefully rechecked and revised the entire reference list to ensure full compliance with the MDPI guidelines. Specifically, journal names have been updated to their official abbreviated forms, conference proceedings have been standardized with publisher information (including city, country, and page numbers), and all arXiv citations have been corrected to include the year and italicized “arXiv.” In addition, book references have been updated to include publisher locations consistently, and all online references (SSRN, Econstor, GitHub, white papers, etc.) have been reviewed to ensure the inclusion of access dates. We are confident that the reference list now fully adheres to the MDPI style guide. Thank you again for your helpful suggestions, which have improved the quality and consistency of the manuscript.