Information asymmetry persists in China's capital market despite improved transparency, leading to risks for investors. While overall information efficiency has progressed, the responsiveness and accuracy of non-financial information lag. This study addresses this gap by investigating the impact of forward-looking non-financial information disclosure on capital market information efficiency. The increasing focus on non-financial information, its qualitative nature, and the growing investor demand for future projections necessitate this research. Regulations in China and the US encourage forward-looking disclosures. This study uses text analysis techniques to quantify the forward-looking content of non-financial information and analyzes its impact on market efficiency, considering the moderating role of analyst attention and mediating roles of analysts' earnings forecast accuracy, audit opinions, and corporate financialization. The study also examines heterogeneity based on media attention and internal control quality.

Existing literature primarily focuses on forward-looking *financial* information, neglecting the impact of forward-looking *non-financial* information on capital market efficiency. Studies highlight the importance of forward-looking disclosures in various contexts, including investor demand, managerial disclosures under uncertainty, and the influence of cultural and institutional factors on disclosure quality. Research also shows a correlation between forward-looking disclosures and stock returns, debt levels, and adverse news. This paper bridges this gap by focusing on the impact of non-financial forward-looking information on capital market information efficiency. The literature on capital market information efficiency highlights the roles of media attention, regulatory scrutiny, analyst forecasts, and financing mechanisms in shaping information efficiency. This study extends this literature by incorporating non-financial information as a key determinant.

The study uses data from Chinese A-share listed companies from 2007 to 2022, excluding financial institutions, ST/ *ST companies, firms with gearing ratios exceeding 1, and those with incomplete data. Share price synchronization, measured using a regression model (R-squared), serves as the proxy for capital market information efficiency. Forward-looking non-financial information content is quantified using machine learning techniques and Loughran and McDonald's lexicon, calculating the proportion of positive and negative words. Analyst coverage is measured by the number of analysts following the company. A four-step multilevel regression model is employed to test the hypotheses, including controls for firm characteristics and industry and year fixed effects. Robustness checks involve altering the stock synchronization measurement method and substituting the non-financial information quantification technique. Endogeneity is addressed using instrumental variable and Heckman two-stage models, and lagged explanatory variables are included. Mediation analysis explores the roles of analysts' earnings forecast biases, audit opinions, and corporate financialization. Heterogeneity analysis is conducted based on media attention and internal control quality.

The study's primary finding is that positive non-financial information disclosures significantly increase capital market information efficiency by reducing stock price synchronization. This supports the hypothesis that more positive forward-looking non-financial information improves market efficiency. Analyst attention moderates this relationship; the positive impact is stronger for companies with high analyst coverage. Mediation analysis reveals that the accuracy of analysts' earnings forecasts, audit opinions, and corporate financialization mediate the relationship between forward-looking non-financial information and market efficiency. Specifically, more positive forward-looking disclosures lead to more accurate analysts' forecasts, a reduced likelihood of non-standard audit opinions, and lower corporate financialization, each of which contributes to improved market efficiency. Heterogeneity analysis shows that the positive impact is more pronounced for companies with high media attention and high internal control quality. Robustness tests using alternative measures of stock synchronization and non-financial information content confirm these findings. Endogeneity tests using instrumental variables and the Heckman two-stage model show that the results remain robust even when addressing endogeneity concerns. The inclusion of lagged variables further strengthens the validity of the results.

The findings confirm the significant role of forward-looking non-financial information in enhancing capital market information efficiency. The moderation by analyst attention highlights the importance of information intermediaries in processing and disseminating this information. The mediation effects underscore the mechanisms through which forward-looking non-financial information impacts market efficiency, improving the accuracy of analysts' forecasts, reducing the risk of non-standard audit opinions, and decreasing the level of corporate financialization. The heterogeneity analysis points to the significance of factors such as media attention and internal control quality in amplifying the benefits of forward-looking disclosures. This has major implications for promoting transparency, reducing information asymmetry, and fostering a more efficient capital market.

This study contributes to the literature by empirically demonstrating the positive impact of forward-looking non-financial information disclosure on capital market information efficiency, particularly in the context of China’s A-share market. Future research could explore the generalizability of these findings to other markets, examine the effects of different types of non-financial information, and further investigate the interaction between non-financial and financial disclosures on market efficiency. The implications of this research are relevant for regulators, companies, and investors in designing disclosure policies and investment strategies.

The study's limitations include the use of data primarily from publicly listed companies, potentially limiting the generalizability of the results to privately held firms. The choice of machine learning model and parameters might affect the accuracy of the quantification of forward-looking content. The measurement of analyst attention might also be susceptible to subjective factors. Future research could address these limitations by including private firm data, exploring alternative quantification methods, and developing more objective measures of analyst attention.