Policy Coordination and Green Transformation of STAR Market Enterprises Under “Dual Carbon” Goals

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1) Primary question of interest in the research

Strengths. The paper asks a timely question: how does coordinated environmental tax reform, green finance, equity‐network synergies influence firms’ ‘green transformation’ under China's dual‐carbon goals, especially in STAR Market companies. The abstract does this well and places the study in contrast to deficiencies in previous research in traditional sectors.

Major concerns.

Ambiguity in “coordination.” “Policy coordination” is claimed but is not defined in terms of an explicit construct. The approach felt short so far is to use a generic Treat×Post DID framing and add-ons (triple differences, mediation), rather than a specific, observable coordination index across instruments (timing overlap, policy intensity alignment or co-movements of policy shocks). What is not directly modelled in the written models is coordination, but average post-policy effects and heterogeneity are represented.

Object of inference. That question suggests two polarities: (i) a national policy narrative using a wide variety of instruments and (ii) a treatment that is a STAR-listed versus not. Those are conceptually different treatments; the first one is policy-intensity at (region/instrument) level, the latter is listing status (firm-selection). Mixing the two up risks confusing selection into STAR with policy exposure.

Suggestions.

Explicitly define and measure “coordination” (e.g.: a composite policy-coordination index derived from instrument timing, intensity, and scope or a quasi-experimental design leveraging staggered rollouts of each instrument).

Clearly specify a treatment and show how this translates to firm exposure beyond STAR-listed effects (e.g. province-level timing of environmental tax reform, coverage in green-finance pilot, or network-policy triggers).

2) Originality

Strengths. The paper offers a unifying perspective on three other policy levers (tax, finance, equity‑network) in the innovation-intensive market that is less explored than the traditional one.

Major concerns.

Methodological novelty is limited. DID: multi-period, triple-difference, event-study, mediation have been used for a long time. The contribution is really one of context, rather than methodology; at this time, the paper does not present a new method for identifying nor an estimator.

Mechanism test is conventional and possible fragile. This “two-stage” mediation approach with DID variables is subject to post-treatment bias and fails to take advantage of recent advances in causal mediation analysis.

Suggestions.

Focus on (design-based) originality, not simply on context: e.g., a stack DID around discrete reform events; and hetero- geneous-treatment estim- ators that accommodate diffuse timing of policy treat- ment (e.g., Callaway–Sant’Anna; Sun–Abraham) if exposure to the policy is career-streamed.

Estimate effects operating through changes (Imai–Keele–Tingley analysis to causal mediation) or utilize iv or policy-distance based instruments for green finance availability and network centrality to cope with simultaneity.

3) Methodology

What works.

The paper presents specifications for multi-period DID, triple-difference, and an event-study to examine dynamics and heterogeneity; and provides the rationale to test parallel trends and dynamic effects.

The sample description (e.g., the proportion of high-carbon firms was 36%; descriptive statistics on the green-transformation index and covariates) is useful background information.

Major concerns.

Treatment definition & identification. Treat=STAR firm and Post=policy time runs the risk of confounding selection into STAR (endogenous, high-tech, growth) with policy effects. Lacking a plausible policy exposure variable (e.g. province-level reform timing, pilot program coverage, firm-level eligibility), the DID estimator may mix selection with policy effects.

Parallel trends & timing heterogeneity. Though the paper reference the event-study approach, the estimation should use estimators that are robust to heterogeneous adoption and dynamic treatment effects; otherwise, the standard two-way FE DID can suffer with negative weighting and unobserved heterogeneity bias.

Clustering and inference. The paper presents two-way FE and clustering on the provincial level. With firm-panel data and serial correlation in outcomes/treatments: Except for clustering at the firm level or in multi-way clustering (firm × province or firm × time), the standard errors are probably underestimated.

Mechanism/mediation validity. The two-step mediation over environmental cost/green finance/network centrality operates with post-treatment covariates and might suck in a portion of the policy effect, the coefficients of which are biased (post-treatment conditioning).

Construct validity for “GreenTrans.” The index”green-transformation index” construction (components, normalisation, sensitivity) is still too coarse in the visible sections; transparency is a must in order to avoid mechanical correlations with policy proxies.

Suggestions.

Redefine T around policy exposure: (i) intensity of environmental-tax reform, by province and year; (ii) period of coverage of green-finance pilot; (iii) exogenous shocks to equity-network connectivity (e.g., mounting/flooring of the listing/ownership rules). Multiply these established exposures by firm pre-policy characteristics.

(i) Replace/augment the TWFE DID by event-study estimators robust to staggered timing (Sun–Abraham) or group-time ATT (Callaway–Sant’Anna). Pre-trend report joint tests, and present placebo leads.

Inference: use firm and time 2-way clustes or wild-cluster bootstrap with few clusters.

Selection into STAR: matching/IPW to create comparable control group of non-STAR firms; Heckman/selection-on-observables sensitivity analysis; or use instrumental variable for STAR listing using plausibly exogenous variation (e.g., eligibility threshold).

Mechanisms: conduct causal mediation analysis with bootstrapped CIs; handle mediator endogeneity (e.g., IV for green finance availability using bank green-credit quotas; network centrality using pre-existing index lags). Pre-specify the DAG and assumptions (sequential ignorability).

Robustness: Pandemic and macro shocks (COVID-19, energy price spikes) should be explicitly controlled for using time×industry fixed effects, and using province×time shocks; consider placebo policies and falsification windows.

4) Conclusions

Strengths. The story connects the findings to policy design; it suggests the dominance of environmental cost pressure and complements by way of financial and network channels, and notes moderation from regional institutions. This policy relevance is a good thing.

Major concerns.

Causal overreach. In view of identification together with (1) treatment definition (2) selection into STAR (3) the limited accounting for staggered policy exposure, the conclusions seem somewhat overstated in terms of causality and “coordination” versus “associations under a DID specification.”

External validity. Prescriptions are general (best use of fiscal instruments, most effective ways of using networks), but evidence is focus- ed on STAR firms (487 firms) and a specific era of the policy; caution is required in applying them to other mar- ket and period.

Mechanism certainty. The mediation statements are robust to possible post-treatment bias and endogeneity of the mediators.

Suggestions.

Reframe conclusions consistent with strength of identification: focus in on credible associations under the stated assumptions; incorporate statement of caveats about selection into STAR and other potential unobserved shocks.

Insert a limitations paragraph (treatment measurement, mediator endogeneity, clustering choice, pandemic confounding) and a next-steps agenda (finer coordination metrics, cleaner policy instruments, more nuanced micro-mechanism data).

Quantify the size of the effects, with confidence intervals and policy-relevant elasticities; provide back-of-the-envelope welfare implications and their sensitivity bounds.

5) Tables, figures and quality of data

What’s solid

You specify the variables and present descriptive statistics and economically robustness tables (Tables 2–3, 10–12), and show distribution plots and a schematic of the DID design.

Major issues

Artifacts of word processing within tables (i.e. “Formatted Table”) show that tables were pasted in with editing metadata, which is a production blocker and just looks unprofessional.

Solution: If a table has this problem, reconstruct every table from scratch out of your nice clean source (LaTeX or Word), delete the field codes, and refresh the PDF.

Table 2 is quite bare—providing names/symbols and measurement units, construction windows (e.g., lags), or data sources per variable are missing; capitalization and notation are not uniform (“green transformation” vs. “Greentransₐᵢₜ”).

Fix: Create a column with definition & unit, period / lag, source, expected sign. Adopt uniform notation (subscripts, italics) throughout the paper.

The table is not only marred by the absence of descriptive statistics (N by firm×time, period) in the table itself, but does not contain the same critical information, inundating the reader to search elsewhere for the 3,896 observations (listed only in the text).

Fix: Place “N=3,896 (487 firms, 2019Q3–2023Q4)” in the Table 3 caption with notes for any sample filters.

Captions for Figures 5–7 are too sparse (no labeled axes, units, bin widths, or N; captions are generic).

Fix: clarify captions (naming x and y-axis or possible international units, N, time period inclusion, for histogram with binning or KDE or DID schematic for treatment and how controls would cluster).

Equation display is corrupted (math not being displayed as it should, rather appears as random symbols and formula place holders), so that makes model somewhat invisible.

Fix: Refactor equations in MathType/LaTeX; proofread all symbols, subscripts after exporting the PDF.

The event-study display (Figure 13) shows dynamics but does not visually convey the presence or absence of uncertainty (CIs) in the figure; the parallel-trend table is there but there is an indirect connection to the plot.

Fix: Present pre-policy leads and post-policy lags with 95% CIs, add a vertical line for the policy and shade the pre-policy band.

Model footnotes are not presented in the results tables. The text states “two-way fixed effects with SE clustered at the provincial level,” although the tables do not report this.

Fix: Include table footnotes – FE structure, clustering level, controls, and legend for stars.

Testing data management decisions for bias: One percent winsorization and industry-year mean imputation may lead to variance compression and attenuation.

Change: Mention sensitivity to (a) lack of winsorization / 0.5% / 2% winsorization; (b) complete-case only; (c) multiple imputation; and provide a table with ditto missingness.

Build transparency: Green Finance is “sum of green credit and green bonds”; Equity centrality is calculated using top-10 shareholders only. Both risk measurement error.

Fix: Report exact formulas, standardization window (firm- or province-time z-scores), sources, and do alternative constructions (e.g., principal component of finance metrics; ownership network using ultimate owners or all reportable blocks).

Reproducibility: Data sources are all listed, but there’s no formal data & code availability statement.

Fix: Include a Here's My Data note (even if action is provided for a synthetic data case) as well as a README describing variable construction.

Comments on the Quality of English Language

6) Language quality (English)

Major issues

Place holders and incomplete metadata: “Sustainability 2025, 17, x … doi. org/10.3390/xxxxx”; “Correspondence: Correspondence.” These signal an incomplete submission.

Hyphenation/encoding oddities: “Mar- ket,” “emi er,” “pa ern,” and (split words in tables) (“Standard er- ror”)—probably from manually (or encoding-wise) breaking lines during export.

Formatting notes (e.g., “Formatted: Font: Bold”; (Production) Headers within the text) that should be deleted.

Repetitive wording and overformality (“This study constructs…”; “This model captures…”), which can be tightened to avoid wordiness.

Fixes (actionable)

For professional proof editing (American English) to fix grammar, subj–verb agreement and article use; select unified style of capitalization (e.g., “Technology Innovation Board” vs. “Science and Technology Innovation Board”).

Replace all placeholders (DOI, correspondence line, affiliations) Per author instructions is not acceptable for resubmission.

Eliminate hyphenation artifacts via suppressing all automatic hyphenation before PDF export; reflow lines and disable broken lines through tables/figures.

Simplicity: write in easy-to-understand plain English, not jargon; reduce the length and jargon by simplifying equations, by cutting introductory and concluding sections, by getting rid of passive voice when not needed.

Homogeneous statistical notation: either - double asterisks to be defined just once and inserted uniformly (asterisks are not defined, but legend should be repeated in each results table).

Author Response

Comment 1:

Opinion: The lack of clear definition and quantitative indicators in policy coordination

Reply 1:

Thank you for your valuable suggestions. We have supplemented the operational definition of the policy coordination index in Section 2.1, constructing a composite index through three dimensions: environmental tax intensity, green finance coverage, and equity network density, with weights determined using the entropy weight method. This index can dynamically reflect the coordination strength of policy tools at the spatial, temporal, and corporate levels.

 

Comment 2:

Opinion: The methodology lacks innovation, and there is post-processing bias in the mediation test

Reply 2:

Thank you for your professional guidance. We have adopted the Imai-Keele-Tingley causal mediation analysis method to calculate the average mediation effect through 1000 Bootstrap samplings, and introduced the green credit quota of each province as an instrumental variable. Compared with the traditional two-step method, the estimated value of the mediation effect decreased by 12%, but the significance remained stable.

 

Comment 3:

Opinion: The clustering of standard errors needs to be optimized, as the interference from the epidemic has not been controlled

Reply 3:

Thank you for your rigorous correction. We have adjusted the standard error clustering to a dual-dimensionality of "enterprise-time", and added a test of the pandemic dummy variables from 2020Q1 to 2022Q4 in Section 4.4. The results show that the change in the policy effect coefficient is less than 5%, confirming the robustness of the conclusion.

 

Comment 4:

Opinion: The conclusion presents a risk of overgeneralizing causality

Reply 4:

Thank you for your valuable reminder. We have added a new paragraph titled "Research Limitations" in Section 5.2, explicitly pointing out issues such as selection bias in STAR enterprises and time lags in mediating variables, and proposing that future causal inference can be strengthened through quasi-experimental design in pilot cities for green finance.

 

Reviewer 2 Report

Comments and Suggestions for Authors

I had the pleasure of reviewing the manuscript titled “The Driving Effect of Policy Coordination under the "Dual Carbon" Goal on the Green Transformation of Enterprises Listed on the STAR Market” to be considered for publication in "Sustainability journal." The research seems sound and provides fairly interesting findings, yet it requires some substantial improvements. Specifics are below:

The title: The title is okey.

The abstract: The abstract is well-structured and clearly written; however, it is somewhat lengthy. It could be further improved by focusing more on the key points.

The introduction: The introduction is generally acceptable, as it clearly identifies and explains the research gap. However, it could be further strengthened by providing more background information on 'hard technology,' the 'dual carbon' strategy, the STAR Market, and the Science and Technology Innovation Board.

Theoretical Framework:

• Figure 3 requires further elaboration and clarification, as it contains different types of arrows (e.g., dotted and solid) whose meanings should be explicitly explained. Also, the three paths must be clearly explained.

Research hypothesis:

• Section 2.2 (Research Hypotheses) is generally well-written; however, supporting the justifications with more recent references would enhance its credibility and make the arguments more convincing.
• In my view, Table 1 could be removed, as it does not provide additional value to the manuscript.

Multi-period DID model design: Section 3 (Multi-period DID model design) demonstrates a professional approach to data collection and analysis. However, it would be beneficial to include a table that confirms or rejects the hypotheses proposed in Section 2.2 (Research Hypotheses).

Discussion Section

• Section 5.1 (Discussion of Results) is particularly strong. Nevertheless, it could be further enriched by incorporating references that either align with or contrast the study’s findings.
• As a constructive suggestion, Section 5.2 (Conclusion and Suggestions) could be further refined by dividing it into three parts: a conclusion, a section for practical implications, and another section highlighting limitations and avenues for future research.

Overall, I find the manuscript to be well-structured, with a clear objective, a well-defined study population, and a research gap that has been appropriately addressed.

Author Response

Comment 1:

Comment: The meaning of the arrow in Figure 3 needs to be clarified, and the hypothesis needs to be supported by updated literature

Reply 1:

Thank you for your meticulous review. We have added a legend for Figure 3 in Section 2.1: solid arrows indicate direct influence paths, and dashed arrows indicate moderating effects.

 

Comment 2:

Opinion: The discussion section should strengthen the comparison with similar studies

Reply 2:

Thank you for your valuable suggestions. We have added three comparative analyses in the discussion section of 5.1 (see the third and fourth paragraphs of Section 5.1), systematically comparing STAR Market enterprises with traditional manufacturing industries and ChiNext enterprises in terms of policy elasticity coefficients, network effect contributions, and other dimensions, revealing the differentiated response mechanisms of technology-intensive enterprises.

 

Comment 3:

Opinion: The conclusion section needs to be supplemented with future research directions

Reply 3:

Thank you for your guidance. We have added a new independent paragraph after the conclusion in Section 5.2, proposing three directions for further exploration: 1) measurement of the temporal matching degree of policy coordination indicators; 2) quasi-experimental design for pilot cities of green finance; 3) exploration of collaborative emission reduction mechanisms in supply chain networks.

Reviewer 3 Report

Comments and Suggestions for Authors The review 1. This article addresses the interesting problem of examining whether and how coordinating actions involving environmental tax reform, launching green financing initiatives, and activating capital networks (including VC) yield positive results in the process of implementing the green transformation of enterprises. 2. This topic appears important for public administration to maximize the effectiveness of pro-ecological actions and protect the economy from climate change in the country, including through the involvement of new green technology sectors. 3. The authors attempt to assess the impact of multiple factors on the green transformation in an original way, using a dynamic DiD model, which assumes that factors affect the study group in a multi-stage manner over time. This method is more representative of the actual impact of economic factors, including multiple changes in tax rates, additional subsidies, or adjustments to aid programs to improve their economic effectiveness. Therefore, this approach to research should be assessed positively. 4. The added value of the obtained results is that this method of evaluating aid programs allows public administration to select the optimal composition of aid tools, not just one-off, but also phased over time. This approach can therefore be applied not only to environmental policy but also to solving other economic and social problems. 5. The bibliography used in the article is appropriate and satisfactory. 6. Regarding the editing – The timeframe of the study is not indicated in the Abstract. 7. In my opinion, “Chapter 2.2. Hypothesis” should be titled “2.2. Hypotheses”. It presents all hypotheses, not just one of them. Furthermore, I see no justification for repeating the hypotheses – the first time in the text and the second time in Table 1 (almost identical wording). 8. The variables in equation (1) are not clearly described. It is true that the Green Trans Index description includes some terms whose symbols are found in the equation. However, these are not all variables, and the meaning of the subscripts and superscripts a, t, ait, CR, GP, GC is not explained. 9. In Table 2, some variables are written in uppercase letters (e.g., Financia Leverage, Enterprise size), and others in lowercase letters (e.g., high-carbon, environmental costs). 10. In Table 4 – the correlation matrix, the statistical significance levels for the correlation coefficients are not indicated. 11. Under Tables 5-12, there is no note explaining the meaning of ***, **, and *. In the minimum option, such a note would suffice only under Table 5 and possibly Table 4 (if the statistical significance level is added in the correlation matrix).

Author Response

Comment 1:

Comment: Variable symbols are not clearly defined, and the table format is not uniform

Reply 1:

Thank you for pointing out the technical specification issues. We have conducted a comprehensive inspection of variable symbols: we have unified the naming format of variables across all tables. Please refer to the revised content in Section 3.2 for details.

 

 

Comment 2:

Opinion: The correlation matrix lacks significant indicators

Reply 2:

Thank you for your professional reminder. We have added the notation "*p<0.1, **p<0.05, ***p<0.01" below Table 4, and uniformly included the standard error clustering specification in all result tables.

We sincerely thank all the reviewers for their valuable feedback! We have implemented all suggestions item by item and marked the changes through the highlighting revision mode. Regarding the methodological limitations and insufficient policy implications pointed out by you, we will explore them in depth in our subsequent research. We kindly request your continued guidance!

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

What’s improved (quick)

Clearer theoretical framing and an explicit policy-coordination index (including entropy-weighting).

The proposed variables and the composite greentrans index C. 

DID/event-study + heterogeneity/mediation sections are finally there. 

What’s not there / still need to clean up Which (if any) of the following attempts do you want me to consider, so that I can deal with this at some point?

Front-matter & production placeholders

DOI is currently “https://doi.org/10.3390/xxxxx”; “Correspondence: Correspondence”.; and we’re left with a stray “Microsoft Word — …” header. Replace with the actual DOI (or delete), full correspondence line (name, affiliation, address, email) and then remove the Word header. 

Broken equation rendering

Summary (Plain text) The GreenTrans formula is missing (lost-it ization) Please correct. Rebuild with MathType/LaTeX and reload the PDF. 

Hyphenation/encoding artifacts

Examples include “STAR Mar-ket” and “carbon emier”. Turn off auto-hyphenation and reflow. 

Methods: staggered-adoption DID

You leverage two-way FE DID w/ province clusters and event study, but you don’t use a more sophisticated staggered timing estimator (group-time ATTs). Either adopt a Sun–Abraham/Callaway–Sant’Anna-type approach, or provide a rationale for not staggering policies. Report group-specific pretrends.

Clustering & serial correlation

Tables report star legends only; footnotes fail to document FE structure and clustering, and the text acknowledges that provincial clustering may understate serial correlation. Include table notes: not looking for grand robust standard errors, would like to see them drug-tested though that would mean specifying FE (firm & time), clustering level(s), re-estimate with multi-way/wild cluster as robustness. 

Consistency between tables and narrative

The text reports the Leverage effect as −0.109 and “significant,” but Table 5 has miniscule coefficients (e.g., −0.007) and no SEs there for Lev. If you do it Signature 5 style, you harmonize number and significance claims, which guarantees that every numeric claim can be traced back to the table on the screen. 

Definition of treatment and timing

Date and era is at 2021Q4 (Action Plan release). Explain how this relates to concrete policies times to kick-in (tax reform, green-finance pilots, network updates). If instruments/policies are implemented at different times in different provinces then write staggered exposure rather than one global Post.

Policy coordination index—operational detail

You talk about entropy weighting, but no normalization windows, the formulae for the entrpy/weights, the resultant weights, and the sensitivity with constant weights/PCA based optimization are given. Put in a brief Appendix with the formulas, the weight values by year/province, and robustness to alternative constructions. 

Data handling transparency & robustness

You still use 1% winsorization and industry–year mean imputation, but don’t demonstrate the robustness of these choices or a completeness check. Include: (a) 0.5%/2% winsorization checks; (b) complete-case results; (c) multiple-imputation check; (d) table of missingness by variable/year. 

Mediation & IV details

You cite Imai–Keele–Tingley and only report first-stage F>10 and lag(centrality) IV, yet there’s no first-stage specification table, instrument rationale (exclusion) or CI for ACME/ADE. Include the two model specs, first-stage table, and bootstrapped CIs for mediation effects. 

Figures and captions

Captions fail to report axes, units, N, time window, binning; event-study/dynamics figure does not display 95% CIs and policy cut-off line visually apparent. Expand label titles and redraw dynamic plot with CIs and vertical line on policy start. 

Results reproducibility & availability

“Data provided in article/supplement” is insufficient for empirical replication. Include a code & data availability subsection (scripts, log of variable construction, pseudonymized competing interests competing interests This is not to be updated) that contains:1.

Comments on the Quality of English Language

A lot of tiny split words/clunky parts still linger (as for #3); as do meaningless figure titles (“Distribution of…”). Put an eye of professional copy-edit on it and standardize capitalization and terms.

Author Response

We sincerely thank the reviewers for their rigorous evaluation and valuable suggestions, which have significantly enhanced the academic quality of this study. All comments have been carefully addressed through comprehensive revisions to the manuscript. Key methodological improvements include: formalizing the entropy weight calculation process in the methodology section and validating the robustness of this approach through detailed sensitivity analyses; further substantiating the rationale for the double-fixed-effects DID framework via time-series clustering analysis of policy implementation timelines, supplemented by empirical decomposition tests in the supplementary materials. All regression results employ standardized error clustering protocols, explicitly label provincial adjustment terms, and systematically compare alternative clustering methods in the robustness testing section. Comprehensive verification against the original dataset corrected critical data biases, such as leverage effect coefficients in the main results table. Policy timing consistency is now systematically explained through analyses of administrative directive coherence, corporate data disclosure thresholds, and provincial implementation windows. Sensitivity tests on truncation thresholds and missing data handling methods confirm the reliability of benchmark processing techniques. The argument for mediation is strengthened by formal reporting of instrumental variable first-stage regression statistics and bootstrap confidence intervals. Inconsistencies in communication details and document metadata formatting have been fully resolved. We commit to continuously refining this study under academic guidance. Detailed point-by-point responses are provided in the document below.

Author Response File: Author Response.docx

Round 3

Reviewer 1 Report

Comments and Suggestions for Authors

Comment 1 — Formatting
I have revised the front matter of the PDF to clearly present the corresponding author’s name and email address. DOI and header formatting have also been aligned with journal guidelines.

Comment 2 — Entropy weight method: add formula + sensitivity
Section 3.2.2 now includes the explicit entropy-weight method formulas, labeled as equations (2)–(4). Additionally, we have conducted a sensitivity analysis to verify the robustness of the entropy-based index, as now stated clearly in the revised section.

Comment 3 — Explain non-staggered DID choice and provide robustness
Section 3.3 explains the rationale for using a two-way fixed-effects DID rather than a staggered approach, given the concentrated timing of the policy shocks in Q4 2021. Robustness checks—including dynamic/event study analysis, parallel trend validation, and placebo tests—are documented in Section 4.4.

Comment 4 — Clarify SE clustering and serial correlation
Partially addressed; further revision made.
The authors originally indicated that standard errors were clustered at the provincial level, as reflected in Tables 5 and 8. However, their cover letter also stated that results with firm–time double clustering and wild bootstrap p-values would be reported. These elements were not initially visible in Section 4.4. The authors have now rectified this by adding a paragraph and a robustness table in Section 4.4, which reports (i) two-way clustered standard errors (firm and time) and (ii) wild-cluster bootstrap p-values for key coefficients. Table footnotes have been updated to clearly indicate “two-way (firm and time) clustering” and “wild bootstrap with 1,000 replications,” resolving the prior inconsistency.

Comment 5 — Fix inconsistent leverage (Lev) coefficient
Addressed.
Table 5 now reflects the corrected coefficient for leverage (Lev = −0.109; SE = 0.043), consistent with the cover letter and underlying estimation.

Comment 6 — Define policy time and justify uniform timing
Addressed.
Section 3.2.2 now includes three justifications for the uniform timing assumption:
(i) uniform national rollout of the State Council’s plan,
(ii) synchronized onset of enterprise disclosure requirements, and
(iii) ±1-quarter provincial implementation windows. These collectively support the uniform policy-time specification.

Comment 7 — Sensitivity to 1% winsorization and mean imputation
Addressed.
Section 3.1 details comparisons using 0.5% and 2% winsorization, as well as multiple imputation versus complete-case analysis. Section 4.4 (Table 15) summarizes these results, confirming the stability of our findings (e.g., estimated coefficients remain 0.342 and 0.341).

Comment 8 — Mediation: include one-stage (first-stage) table + CIs.

Addressed.
Section 4.3 now includes first-stage statistics, with F = 12.7 (Table 8), confirming the relevance of the instrument. Table 9 reports 95% bootstrap confidence intervals for ADE and ACME based on 1,000 resamples.