Stock price crash risk, the risk of a sharp stock price decline, poses significant threats to investors and market stability. Company management's concealment of negative information contributes to this risk, as the accumulation of bad news eventually leads to a crash when revealed. Investor psychology and behavior are also crucial factors. ESG (Environmental, Social, and Governance) investing, a prominent concept globally, emphasizes a company's environmental, social, and governance performance beyond traditional financial measures. The Chinese government actively promotes ESG, mandating ESG data disclosure for certain industries. ESG information enhances transparency and can serve as a risk measurement indicator, potentially reducing stock price crash risk. However, inconsistencies in ESG ratings from different agencies create ESG uncertainty, hindering investor assessment. This study innovates by using a PVAR model to explore the bidirectional relationships between ESG uncertainty, investor attention, and stock price crash risk in the Chinese market, employing a newly constructed ESG Uncertainty Index based on data from six major Chinese ESG rating agencies. This approach offers a more comprehensive analysis than previous studies, contributing valuable insights for policymakers, corporate management, and market participants.

Existing research attributes stock price crash risk to principal-agent conflicts and information asymmetry. Studies generally show a negative correlation between strong ESG performance and stock price crash risk, although this relationship's strength and consistency vary. The impact of ESG uncertainty on stock price crash risk is under-explored. Theoretically, ESG uncertainty could exacerbate information asymmetry, weaken investor confidence, and affect the interpretation of management behavior, potentially increasing crash risk. However, in the Chinese market, this relationship might be influenced by factors such as investor awareness of ESG issues, government policies, and company disclosure practices. Research on the relationship between ESG performance and investor attention indicates that high ESG performance attracts more investor attention, driven by both financial returns and ethical investment strategies. The relationship between investor attention and stock price crash risk is generally considered positive, with high attention potentially amplifying market reactions and increasing crash risk. However, in the Chinese market, this relationship might be affected by the high proportion of retail investors and information asymmetry. This paper aims to fill the gap in the literature by comprehensively examining the bidirectional relationships among ESG uncertainty, investor attention, and stock price crash risk in the Chinese context.

This study uses a Panel Vector Autoregression (PVAR) model to analyze the dynamic relationships among ESG uncertainty, investor attention, and stock price crash risk. The sample includes companies listed on the main board and Growth Enterprise Market (GEM) of Shanghai and Shenzhen stock exchanges from 2011 to 2021, excluding ST and *ST companies, financial industry companies, and those with missing data. Data are winsorized at the 1% and 99% quantiles. The ESG Uncertainty Index is constructed by integrating ESG rating data from six major Chinese ESG rating agencies (Hua Zheng, Bloomberg, Syn Tao Green Finance, Wind, FTSE Russell, and MSCI). The index measures the standard deviation of percentile rankings from these agencies. Investor attention is measured using the sum of Baidu search volumes for the company's stock code, short name, and full name, logarithmically transformed. Stock price crash risk is measured using the Negative Coefficient of Skewness (NCSKEW). The PVAR model includes lagged terms of the variables to address endogeneity. The study also incorporates control variables (profitability, asset structure, debt ratio, and company age) to mitigate omitted variable bias. The stability of the panel data is tested using the Augmented Dickey-Fuller and Phillips-Perron tests. The optimal lag order is determined using AIC, BIC, and HQIC criteria. System GMM estimation is used to handle fixed effects and endogeneity. Granger causality tests, impulse response analysis, and variance decomposition are conducted to analyze the dynamic relationships among the variables.

The Granger causality tests reveal no direct causal relationship between ESG uncertainty and stock price crash risk in the Chinese market. However, there is a bidirectional Granger causality relationship between ESG uncertainty and investor attention, and between stock price crash risk and investor attention. The PVAR model results show that while the lagged one-period coefficient of stock price crash risk on ESG uncertainty is not significant, the lagged two-period coefficient is significantly positive, suggesting that the impact of stock price crash risk on ESG uncertainty emerges over time. Both lagged one and two periods of investor attention have significantly positive impacts on ESG uncertainty. The lagged one-period effect of investor attention on stock price crash risk is significantly negative, while the lagged two-period effect is significantly positive. This suggests that increased investor attention can initially reduce crash risk due to improved information dissemination but may eventually lead to market overreaction and increased risk. The impulse response analysis confirms these findings, showing that increased ESG uncertainty leads to increased investor attention, which in turn increases ESG uncertainty. Increased investor attention initially decreases stock price crash risk but increases it in the long run. The variance decomposition analysis indicates that the fluctuations of ESG uncertainty and stock price crash risk are primarily explained by their own shocks. The impact of investor attention on these variables increases over time. Overall, investor attention plays a significant mediating role.

The findings challenge the conventional assumption of a direct link between ESG uncertainty and stock price crash risk. The study demonstrates that ESG uncertainty does not immediately translate to increased crash risk, instead highlighting the role of investor attention as a mediating factor. The bidirectional causality between ESG uncertainty and investor attention emphasizes the dynamic feedback loop where uncertainty attracts attention, and increased scrutiny further reveals uncertainty. The initial negative and subsequent positive effects of investor attention on stock price crash risk underscores the complex dynamics of market behavior, with increased attention potentially initially mitigating risk but later amplifying it. These findings enrich our understanding of the intricate interplay between ESG factors, investor behavior, and market stability in the context of the rapidly evolving Chinese market. The results contribute significantly to the literature on ESG investing, stock price crash risk, and investor behavior, providing valuable insights for both theoretical development and practical applications.

This study demonstrates that in the Chinese market, ESG uncertainty does not directly cause stock price crash risk but indirectly affects it through investor attention. There is bidirectional causality between ESG uncertainty and investor attention, and between stock price crash risk and investor attention. The findings highlight the importance of improving ESG information disclosure to increase market transparency and stability. Future research could explore the differences in these relationships across various industries, company sizes, and investor types, and could also investigate the impact of specific regulatory interventions.

The study's reliance on data from the Chinese market limits the generalizability of the findings to other contexts. The specific metrics used for investor attention and stock price crash risk might introduce biases. The PVAR model, while robust, might not capture all potential causal relationships. Future research could use a broader sample, incorporate additional variables, and employ alternative methodologies to address these limitations.