The greenwashing coefficient survives firm and year fixed effects at the 1% level, yet its size remains irreconcilable with the paper's descriptive statistics. Chen describes the regressor plainly. The abstract calls it "a proxy for potential greenwashing based on the standardized gap between ESG disclosure scores (Bloomberg) and ESG performance ratings (Huazheng)". It adds that "this disclosure-rating gap serves as an empirical proxy". One line later, the same abstract turns the proxy into "a significant predictor of corporate debt default". The title drops the qualification and simply says greenwashing. The paper merits attention as evidence on ESG rating disagreement across 11,315 Chinese A-share firm-years from 2009 to 2022. Its default-risk estimate has no scale anyone could sensibly trade or use.
The premise looks plausible. A firm talks loudly about its environmental credentials while an outside assessor judges its actual standing poorly. The literature calls that talk without walk. Chen argues that punishment eventually follows through reputation loss, regulatory fines and litigation cash outflows, or equity market repricing. Each route is meant to weaken debt-servicing capacity. Yet the paper reports no portfolio sort, no out-of-sample test and no cost analysis. The gap therefore receives no trading test.
Chen estimates a panel regression. The greenwashing proxy subtracts the Z-score of the Huazheng ESG rating from the Z-score of Bloomberg's ESG Disclosure Score, whose range is 0.1 to 100. Huazheng's nine letter tiers, C to AAA, are coded 1 through 9. Expected default frequency, EDF, is the dependent variable and comes from the Bharath-Shumway naive distance-to-default model. EDF equals Normal(-DD). Chen multiplies it by 100, making one unit equal to one percentage point of default probability.
The sample covers Chinese A-share firms annually from 2009 to 2022. Financials and ST/PT firms are excluded. Continuous variables are trimmed at 1% and 99%, leaving 11,315 firm-year observations. The controls are Size, Lev, ROE, ROA, Growth, Board, TOP1, TobinQ and Cashflow. Firm and year fixed effects are included, with standard errors clustered by firm.
The main coefficient is 0.949, with a standard error of 0.258 and significance at 1%. Within R-squared is 0.022. Chen next separates the regressor and outcome in time by lagging greenwashing one period and leading EDF one period. Two robustness checks follow. A 2SLS specification using lagged greenwashing as the instrument produces 0.906 (SE 0.416, z reported as 3.05). Winsorizing at the 5th and 95th percentiles yields 0.953 (SE 0.288). The paper then presents three subsample splits and a three-equation mediation analysis covering reputation, regulatory violations and market-adjusted return volatility.
A coefficient without a stable scale
The descriptive table is where the trouble begins. On Chen's rescaled percentage-point basis, EDF has a mean of 0.001, a median of 0.000 and a standard deviation of 0.024. The greenwashing regressor's standard deviation is 1.216. A baseline coefficient of 0.949 therefore implies that a one-unit increase in the gap moves modelled default probability by roughly forty standard deviations of the dependent variable. A handful of high-EDF observations may be driving the regression, since the maximum is 0.680. The other possibility is that the regression and descriptive table use different scales.
The mediation results make the mismatch harder to dismiss. For the same dependent variable, the direct Gws coefficient is 0.653 (SE 0.208) in the reputation column, 0.142 (SE 0.078) in the violations column and 0.013 (SE 0.004) in the volatility column. Those regressions contain 9,770, 8,595 and 10,775 observations. The column nearest the baseline sample size has the smallest coefficient, lower by a factor of seventy. Leverage follows the same pattern, moving from 0.008 in the baseline to 0.011, 0.009 and 0.001.
The violations equation has the scale problem in reverse. A one-unit increase in greenwashing raises recorded regulatory violations by 40.376 (SE 17.856) across 7,446 observations. That amounts to about forty additional violations for a variable with a standard deviation of 1.216. The paper gives no reason for the large changes in coefficient magnitude across the mediation columns. Nor does it account for the fall in observations from 11,315 to 9,770, 8,595 and 10,775.
Chen does address missing observations in the heterogeneity analysis, explaining that "the absence of 247 observations in the full sample stemmed from incomplete information on the ownership nature of certain enterprises". The mediation columns need an equivalent explanation. Until they receive one, the economic magnitude behind the headline remains unresolved. I would not put 0.949 percentage points of default probability before a credit committee.
What does the vendor gap measure?
Chen acknowledges the measurement issue in the conclusion: "our greenwashing measure captures disclosure-rating gaps rather than direct environmental outcomes; future research should validate our findings using hard environmental data." The data section makes the same distinction. This measure tracks the distance between disclosure intensity and perceived performance, rather than direct environmental outcomes.
The paper offers two defences. The method follows established practice, with citations to Yu et al. (2020) and Christopher et al. (2016). Chen also writes that "this limitation is inherent to many studies in this field, given the current state of standardized environmental performance data availability in China." The latter explains why hard outcome data are unavailable. It gives less support to the abstract's use of greenwashing or its claim that the gap predicts debt default.
The paper itself identifies the deeper measurement concern. Huazheng evaluates a firm's ESG position by "synthesizing various information sources, including both disclosed data and analyst assessments of actual practices". Those analyst assessments support Chen's interpretation. Disclosed data also enter Huazheng's rating, meaning the rating already contains some disclosure information.
The resulting gap subtracts a composite containing disclosure from a disclosure score, using vendors with different coverage, methodologies and industry treatments. Bloomberg measures reporting completeness. Huazheng maps firms into a nine-point letter tier. A company can report thoroughly in an industry that Huazheng judges harshly, producing vendor disagreement without establishing misrepresentation.
The subsample evidence permits either reading. In the East, the effect is 1.133 (SE 0.517), compared with an insignificant 0.694 in the West and 0.542 in the Central region. Chen explains the eastern estimate through "stricter market regulation, and more developed information dissemination channels", alongside more sensitive investors and a more mature price discovery function.
The industry split produces 1.577 (SE 0.594) among non-high-tech firms and 0.112 among high-tech firms. Chen invokes pollution intensity and heavier asset structures here. The paper also says that "traditional industries often face higher environmental compliance costs, meaning that reputational damage caused by greenwashing can more directly undermine operational stability and debt-servicing capacity". High-tech firms, it adds, have greater financing flexibility to absorb the damage. Both explanations hang together. The same patterns could also arise from differences in where Bloomberg and Huazheng disagree because of coverage and methodology.
Prediction only reaches 10%
The abstract describes greenwashing as "a significant predictor of corporate debt default." Yet both specifications that put the regressor before the outcome reach only 10%. Lagged greenwashing on current EDF has a coefficient of 0.658. Current greenwashing on next-period EDF gives 0.652. Each has a t of 1.81, using 9,615 observations and 1,244 firms.
Chen reports this openly and argues that the direction and magnitude remain aligned with the baseline. Fair enough. The result significant at 1% is contemporaneous. Meanwhile, the 2SLS instrument is the lagged regressor itself. If the vendor gap persists and EDF is autocorrelated, that instrument cannot resolve reverse causality. Both the annual market-based EDF and the difference between two vendor scores are likely to move slowly. The paper supplies no persistence or autocorrelation statistics that settle the issue.
The channel with the wrong sign
Among the three channels, regulatory violations provide the clearest sequence. Greenwashing increases the count of regulatory violations at the 5% level, while violations enter EDF positively at 5%. Reputation also moves in the expected direction: greenwashing lowers the reputation measure at 1%. Its coefficient is -6018.095 (SE 1383.555). Chen describes that mediator as a factor score grouped into ten deciles. I could not reconcile that description with the coefficient.
Volatility creates a direct contradiction. Greenwashing raises market-adjusted volatility by 20.510 (SE 5.428). Chen then states that "the volatility-EDF association in our sample is negative, implying that higher volatility is associated with lower perceived default risk," with significance at 1%. In a Merton-type model, equity volatility feeds directly into asset volatility and default distance. The sign is therefore diagnostic trouble.
Chen interprets it as a pre-exposure phase in which green signalling attracts attention and temporarily improves market-based health indicators. The story is coherent. A negative in-sample mediator coefficient does not establish it. EDF and VAR_ADJ also come from the same return series, pulling the relationship closer to arithmetic than economics. The mediator coefficients on EDF appear as 0.000 and -0.000, both with standard errors of 0.000. The table provides no readable indirect magnitude. I did not find bootstrapped indirect effects or a Sobel test anywhere in the paper.
One result worth carrying forward
The SOE split survives.
For state-owned enterprises, the gap-default coefficient is 1.673 (SE 0.636). For non-SOEs it is 0.010 (SE 0.099, insignificant), based on 6,162 and 4,906 observations. The measured effect resides entirely in the SOE book. Chen argues that SOEs face greater social responsibilities and stricter public oversight. Exposure could also cost them policy advantages and financing facilities, while agency problems are more pronounced in their governance structures. This is the result I would want to see replicated with hard emissions data.
I could not test the finding independently. Constructing the signal requires Bloomberg ESG disclosure scores and Huazheng ratings, both available under commercial licence. The paper's data statement says they are not redistributable. Its universe is Chinese A-shares, where we have no equity, credit or default-event coverage. Reconstructing a disclosure-rating gap for US equities would substitute one market for another, with a different measure and disclosure regime, and would provide no reproduction of Chen's signal.
The paper is best used as a reason to reconcile ESG vendors for China-exposed issuers and to scrutinize SOE names where loud disclosure accompanies a mediocre external rating. My view of the headline would change if the baseline coefficient could be shown on a scale consistent with the descriptive statistics, and if the lead specification cleared 5%.