CSR decoupling and stock price crash risk: Evidence from China

The study investigates whether and how corporate social responsibility (CSR) decoupling—the inconsistency between firms’ CSR disclosures and their actual CSR activities—affects subsequent stock price crash risk (SPCR) in China. Motivated by mixed evidence on the CSR–SPCR link, the authors argue that prior work often assumes disclosure and performance align, potentially obscuring effects when firms overstate CSR in reports. Given China’s volatile equity market and the growing prevalence of CSR reporting, the paper aims to clarify whether CSR decoupling increases crash risk via information asymmetry and managerial bad-news hoarding. The research formulates two hypotheses: H1, CSR decoupling increases SPCR; and H2, the effect is stronger under higher agency risk (weaker monitoring, governance, or institutional environments).

The literature shows competing views on CSR’s impact: proponents argue CSR improves stakeholder relations, transparency, and reduces information asymmetry, thereby stabilizing prices; critics contend CSR can be used opportunistically to mask negative news, raising risk. CSR decoupling research has largely focused on determinants—external environments (laws, markets, stakeholders), firm characteristics (board structure, committees, ownership type), and managerial traits (overconfidence, power, narcissism, political ties)—with fewer studies on consequences. Evidence suggests symbolic CSR can damage reputation and market value and worsen information quality. SPCR research highlights bad-news hoarding due to agency problems (perfect rationality perspective) and behavioral biases from bounded rationality (e.g., overconfidence). Within an agency framework, information asymmetry allows managers to hide adverse information until it is released in lumps, causing crashes. The authors posit that decoupling undermines the informativeness of CSR disclosures, inflates perceptions of firm fundamentals, and heightens crash risk. They therefore hypothesize that CSR decoupling raises SPCR (H1), and this relationship strengthens where agency risks are higher (H2).

Data and sample: Chinese A-share listed firms from 2010–2019. Exclusions: firms with fewer than 30 weeks of annual weekly returns, financial sector firms, ST and PT firms, and observations with missing variables. Final sample: 5,297 firm-year observations. Continuous variables are winsorized at the 1st/99th percentiles. Data sources: CSR performance ratings from Hexun; stock, financial, and governance data from CSMAR. CSR decoupling measure (CSR_GAP): The authors use textual analysis of each firm’s standalone CSR/ESG/Sustainability report (from cninfo.com.cn) to compute an Optimistic Tone index, serving as a proxy for the disclosure dimension. Optimistic Tone is calculated as (frequency of positive tone words − frequency of negative tone words) divided by (positive + negative frequencies), using the Loughran–McDonald dictionary to classify words. This tone index is interpreted inversely as CSR reporting quality (more optimistic tone indicates lower quality due to potential impression management). Actual CSR performance is measured by Hexun’s CSR performance rating. Both Optimistic Tone and CSR Performance are converted to Z-scores. CSR decoupling is defined as CSR_GAP = Optimistic Tone − CSR Performance. A larger positive CSR_GAP indicates greater decoupling (more optimistic disclosure relative to actual performance). An alternative tone measure divides the positive–negative word difference by total words (CSR_GAP2 constructed similarly as robustness). Stock price crash risk (SPCR): Following Hutton et al. (2009) and Kim et al. (2011), two main measures are constructed using weekly idiosyncratic returns. First, weekly residuals are obtained by regressing individual stock returns on market returns with leads and lags to compute firm-specific returns. Weekly idiosyncratic returns are W_it = ln(1 + residual). SPCR proxies are: (1) NCSKEW, the negative conditional skewness of weekly idiosyncratic returns within year t+1; higher values indicate greater left-tail risk; and (2) DUVOL, the log ratio of the variance of down weeks to up weeks within year t+1; higher values indicate more left-skewed return distributions. As a robustness measure, CRASH equals 1 if any weekly return in a year is more than 3.09 standard deviations below the annual mean, 0 otherwise. Baseline model: CrashRisk_{i,t+1} (NCSKEW or DUVOL) is regressed on CSR_GAP_{i,t}, controls, and fixed effects: CrashRisk_{i,t+1} = α0 + α1 CSR_GAP_{i,t} + α2 Controls_{i,t} + Year FE + Industry FE + ε_{i,t}. Controls include volatility (Sigma), cumulative returns (Ret), leverage (Lev), size (Size), book-to-market (BM), profitability (ROA), information opacity (Opaque; absolute discretionary accruals from modified Jones averaged over t, t−1, t−2), and turnover change (Dturn). Standard errors are clustered. Endogeneity and robustness: To address self-selection into CSR reporting and potential reverse causality, the authors implement a Heckman two-stage model. Stage 1 estimates the probability of issuing a CSR report (CSRD) via Probit using determinants (BM, Size, Lev, ROA, ownership concentration, independent director ratio, CEO duality, Tobin’s Q), and a mandatory disclosure indicator as an exclusion restriction. Stage 2 includes the inverse Mills ratio in the crash-risk regressions. Additional robustness includes firm fixed effects, alternative SPCR and CSR decoupling measures, controlling prior-period crash risk and corporate governance variables, and excluding outliers. Mechanism analysis uses subsample splits by agency risk proxies: executives’ overseas and academic experience, external monitoring (institutional ownership; media attention measured by counts of news coverage), and external governance environment (provincial marketization index and legal environment measured by economic case completion rates).

- Descriptive statistics: CSR_GAP shows substantial cross-firm variation (mean −0.881, SD 1.507). NCSKEW and DUVOL means align with prior literature. - Baseline regressions (Table 3): CSR_GAP is positively associated with future crash risk. With year and industry fixed effects, coefficients are 0.0222 on NCSKEW (t=3.7734, p<0.01) and 0.0114 on DUVOL (t=2.8335, p<0.01). Economically, a 1% increase in CSR decoupling raises crash risk by approximately 2.09% (NCSKEW). N=5,297. - Endogeneity checks: - Heckman two-stage (Table 4): Mandatory disclosure strongly predicts CSR reporting in the first stage. In the second stage, CSR_GAP remains significantly positive for NCSKEW (0.0251, p<0.05) and DUVOL (0.0120, p<0.10). - Firm fixed effects (Table 5): Results persist with CSR_GAP coefficients 0.023 (SE 0.007, p<0.01) for NCSKEW and 0.012 (SE 0.004, p<0.05) for DUVOL. - Additional robustness: - Alternative SPCR (CRASH, Table 8): Logistic regression shows CSR_GAP positively predicts crash events (coef 0.0766, p<0.05). - Alternative CSR decoupling (CSR_GAP2, Table 8): Positive and significant for NCSKEW (0.0233, p<0.01) and DUVOL (0.0119, p<0.10). - Controlling prior crash risk and governance variables; excluding outliers: core results unchanged (Tables 6–7). - Mechanism tests (Table 9): Effects are stronger where agency risks are higher, consistent with information asymmetry as the channel. - Executives’ experience (Panel A): CSR_GAP significantly predicts higher SPCR only when firms lack executives with overseas or academic experience (e.g., without academic experience: NCSKEW coef 0.0198, p<0.05; DUVOL coef 0.0130, p<0.05), but not when such experience is present. - Monitoring forces (Panel B): Effects are significant in low institutional ownership and low media attention subsamples (e.g., low institutional ownership: NCSKEW 0.0272, p<0.05; DUVOL 0.0175, p<0.01), and weaker/insignificant when monitoring is high. - External governance environment (Panel C): Effects are significant in provinces with low marketization (NCSKEW 0.0349, p<0.01; DUVOL 0.0192, p<0.01) and weak legal environments (NCSKEW 0.0238, p<0.01; DUVOL 0.0123, p<0.05), but not in stronger environments.

The findings support the hypothesis that CSR decoupling elevates stock price crash risk by worsening information asymmetry and facilitating managerial bad-news hoarding. By separating CSR disclosure tone from actual CSR performance, the study clarifies mixed prior evidence on CSR’s relation to crash risk: it is not CSR per se, but the divergence between disclosure and reality that matters for tail risk. The mechanism analysis shows the effect intensifies under higher agency risk—where managerial discretion is greater and monitoring or institutional constraints are weaker—reinforcing agency theory predictions. These results highlight that optimistic CSR narratives, when misaligned with real actions, can inflate investor expectations and delay recognition of adverse information, raising the likelihood of abrupt price corrections. The study thus underscores the importance of credible, high-quality sustainability reporting and robust monitoring mechanisms in mitigating market tail risks.

The paper documents that CSR decoupling—optimistic CSR disclosure relative to actual CSR performance—significantly increases future stock price crash risk among Chinese A-share firms. This relationship is robust to endogeneity controls, alternative measures, and additional specifications. The effect is concentrated in contexts of higher agency risk (absence of executives’ overseas/academic experience, weak external monitoring by institutions and media, and weaker marketization/legal environments), consistent with information asymmetry as the mechanism. Contributions include: clarifying mixed CSR–SPCR evidence by focusing on the disclosure–performance gap; adding to the literature on economic consequences of CSR decoupling; and enriching determinants of SPCR by highlighting managerial impression management. Practical implications: firms should strengthen internal governance and provide accurate, transparent CSR reporting; investors should critically assess CSR narratives versus performance; regulators should enhance CSR reporting standards, supervision, and accountability. Future research could develop CSR-specific textual dictionaries to better capture disclosure content and decision usefulness, and examine CSR decoupling and crash risk during crisis periods (e.g., COVID-19) and across different institutional settings.

Key limitations include: (1) CSR decoupling is proxied by an Optimistic Tone metric derived from general financial text dictionaries; this may not fully capture CSR report nuances or the richness of disclosure content. Developing CSR-specific lexicons could refine measurement. (2) The study focuses on 2010–2019 to avoid pandemic effects; results may differ under crisis conditions such as COVID-19. (3) As with observational studies, residual endogeneity cannot be entirely ruled out despite Heckman selection, fixed effects, and robustness tests.