Quantifying human capital disclosure in China with textual analysis Open Access

Human capital – encompassing employees’ skills, knowledge and abilities – has been identified in the literature as a corporate strategic intangible asset in the long run. Human capital’s importance as an intangible asset is based on its role in providing a sustainable competitive advantage as a valuable, rare, inimitable and non-substitutable resource (Barney, 1991) and its fostering innovation and adaptability through employees’ ability to learn and apply new skills, adapt to changes and generate creative solutions (Ployhart & Moliterno, 2011). As a consequence, effective human capital management (HCM) can enhance organizational performance by promoting productivity and financial performance (Crook, Todd, Combs, Woehr, & Ketchen, 2011) and drive long-term value creation based on stakeholders seeking to understand how companies integrate human capital considerations into their overarching strategy to create long-term value (Klemash, Neill, & Smith, 2019).

Recently, the COVID-19 pandemic has significantly elevated the importance of human capital in the social dimension of ESG practices. The pandemic has caused enhanced focus on employee well-being and adaptation to new work models, increased emphasis on diversity, equity and inclusion and the need for maintaining organizational resilience and agility via HCM practices during times of uncertainty (Ployhart, Nyberg, Reilly, & Maltarich, 2014; Collings, McMackin, Nyberg, & Wright, 2021).

In the United States, the Securities and Exchange Commission (SEC) updated the HCM disclosure requirements in November 2020, which reflects a significant shift toward greater transparency and accountability in how companies manage and report on their human capital resources. By adopting a principles-based approach and emphasizing materiality, the SEC allows companies flexibility in tailoring their disclosures while ensuring that investors receive relevant and meaningful information (Mayew & Zhang, 2022). The focus on areas such as employee development, health and safety, diversity and inclusion and compensation and benefits underscores the importance of human capital in achieving organizational success and resilience, particularly in the context of the challenges posed by the COVID-19 pandemic ^[1]. Subsequently, an increasing number of studies have examined the disclosure patterns of HCM disclosure after such updated requirements by the SEC (see, e.g. Batish et al., 2021; Bourveau, Chowdhury, Le, & Rouen, 2022). Particularly, a few studies (e.g. Zhang, 2022; Demer et al., 2024) employ textual analysis using natural language processing (NLP) methods and construct HCM disclosure indices.

In China, the central government has made great efforts to promote corporate HCM practices through formal institutions, such as issuing the Employment Promotion Law 2007 and the Labor Contract Law 2008 as well as the latest amendments to the Company Law in 2023. For example, the Employment Promotion Law 2007 focuses on increasing employment opportunities and ensuring fair employment practices. The Labor Contract Law 2008 aims to regulate labor contracts and protect workers’ rights. The Company Law Amendments 2023 introduce significant changes impacting HCM by highlighting employees as important stakeholders. Furthermore, as ESG considerations become more integral to business strategy, companies are expected to disclose how they manage their human capital in alignment with social and governance goals. The China Stock Exchanges have been actively promoting ESG reporting guidelines (involving a broad range of ESG topics, including human capital or employee aspects) for listed firms since the beginning of 2024 ^[2]. This institutional emphasis on HCM in China warrants a close investigation of the determinants, disclosure and effect of HCM. However, existing studies are scarce and provide limited insight, primarily due to the absence of a comprehensive measure of firm-level HCM.

This study intends to fill this void in the literature by constructing a novel measurement of human capital management disclosure (HCM disclosure). We conduct textual analysis of HCM disclosure in the management discussion and analysis (MD&A) section of the annual report using NLP methods. The attention-based view suggests that the allocation of individual attention is a critical determinant of strategic priorities (Ocasio, 1997). In this regard, managers who focus on HCM are more likely to implement practices that develop and leverage human capital, leading to competitive advantages. To proxy for HCM disclosure, we focus on the MD&A section in the annual report. The MD&A section provides a narrative explanation of a company’s operational performance and strategic direction from the management perspective, and its content reflects the priorities and focus areas of the management team. Therefore, we believe that analyzing the extent and depth of HCM-related disclosures in the MD&A can help infer the level of attention the management team assigns to human capital matters.

Based on Batish et al. (2021), we identify ten dimensions of HCM, namely diversity and inclusion, employee development, health and safety, compensation and benefits, employee engagement, employee turnover or tenure, culture, recruiting practices, pay equity and succession planning. We follow Sautner, Van Lent, Vilkov, and Zhang (2023) and construct HCM exposure measures (HCM_Exposure) for these ten dimensions. Finally, we conduct factor analysis of the ten HCM exposure measures and use the principal factor (HCM_Disclosure) as our proxy for HCM disclosure.

We collect data for listed firms in the Shanghai, Shenzhen and Beijing Stock Exchanges during 2009–2022 and construct a sample of HCM_Disclosure with 39,825 firm-year observations for 5,153 unique companies. We find substantial variations in HCM_Disclosure across industries, geographic regions and ownership types. Also, we observed significant increases in HCM_Disclosure in 2021 and 2022, reflecting an increase in HCM disclosure during the COVID-19 pandemic.

Formal multivariate regression analysis shows that firm-fixed effects have strong explanatory power on variations in HCM_Disclosure. Controlling for firm-fixed effects, firms’ operating performance, cash holdings, R&D expenditure, sales growth, labor intensity, leverage and market-to-book ratio appear to be key determinants of HCM_Disclosure: firms with high HCM_Disclosure are better-performing and have more cash holdings and R&D expenditure. These firms are also more labor-intensive but are associated with lower sales growth, market-to-book ratio and leverage.

We also examine the association between HCM_Disclosure and firm performance. We find that HCM_Disclosure is associated with better future accounting performance, measured by return on assets (ROA), but poor future market valuation, measured by Tobin’s Q. The positive relationship with future ROA likely stems from productivity gains from reduced turnover and enhanced employee capabilities (Guiso, Sapienza, & Zingales, 2015). However, the negative association with Tobin’s Q reflects market skepticism about the immediate payoff of HCM investments. Kanodia and Sapra (2016) suggest that markets struggle to price intangibles due to measurement uncertainty. Such a negative association is also consistent with the possibility that the market undervalues intangible investments in human capital due to information asymmetry and difficulties in valuation (Peters & Taylor, 2017). The time lag between HCM investments and market recognition of their value may also contribute to this observed relationship, as suggested by Edmans (2012), who documents that the market does not fully value intangible assets until their benefits materialize in tangible outcomes.

Our study provides the following contributions to the literature: First, based on the attention-based theory, we develop a comprehensive measurement approach for HCM disclosure using textual analysis of MD&A disclosures, providing a quantitative and replicable methodology that can be applied across different markets and institutional contexts ^[3]. This measurement approach involves identifying and categorizing substantial HCM-related discussions in MD&A sections, resulting in a comprehensive, ready-to-use dataset for Chinese listed companies that advances the literature on both HCM practices and MD&A-related studies. Second, we add to the theoretical understanding of corporate disclosure by documenting how managerial attention to specific strategic priorities, particularly human capital, is reflected in narrative reporting. Our findings extend the literature on the information content of MD&A disclosures and their relationship to corporate priorities and outcomes.

The remainder of the paper is organized as follows: Section 2 presents the related literature and institutional background. Section 3 presents the construction of the HCM disclosure measure. Section 4 conducts empirical analyses of the constructed measure. Section 5 discusses and concludes.

In November 2020, the Securities and Exchange Commission (SEC) updated and revised human capital management (HCM) disclosure requirements as a part of a broad overhaul of Form 10-K disclosure. Under the revised rules, firms are required to “provide a description of the registrant’s human capital resources, including in such description any human capital measures or objectives that management focuses on in managing the business, to the extent such disclosures would be material to an understanding of the registrant’s business taken as a whole” ^[4].

Accordingly, an increasing number of studies have examined the issues of HCM disclosure. For example, Batish et al. (2021) examined a sample of one hundred Form 10-K filings following the SEC rule revisions. They use the Equilar database to identify the voluntary HCM content by classifying 11 topics, i.e. diversity, employee development, safety, compensation, employee engagement, tenure/turnover, culture, recruiting, mental health, pay equity and succession planning. They suggest that more than 50% of the sampled firms disclosed the content of diversity and inclusion, employee development and safety, while mental health, pay equity and succession planning were the least frequent topics disclosed by the sampled firms.

Zhang (2022) employs a machine learning technique on the 10-K documents from 1996 to 2017 and develops two dimensions of HCM disclosure: operational-oriented and social-oriented HCM dimensions. She finds that firms are more likely to disclose social-oriented HCM information to differentiate themselves from competitors when facing high market competition and that companies are less likely to disclose operational-oriented HCM information when facing fierce market competition. She also finds that firms with higher socially oriented HCM disclosures obtain higher ratings in the social component of ESG scores and attract more sustainable investors.

Bourveau et al. (2022) hand-collected a large sample of 10-K filings from 2018 to 2023 and examined quantitative human capital disclosures using them. They suggest that after the SEC’s 2020 revision to Regulation S-K on human capital information, firms should increase their disclosure of human capital metrics in 10-K filings.

Demers, Wang, and Wu (2024) used a machine learning algorithm (word2vec) trained on a confirmed set of human capital disclosures and developed a list of keywords classified into five human capital dimensions, including compensation and benefits, diversity, equity and inclusion, labor relations and culture, health and safety, demographics and others.

In China, the central government has recently made great efforts to promote corporate HCM practices. Since joining the World Trade Organization in 2001, China has launched or revised HCM-related laws or policies. For example, the Employment Promotion Law 2007 focused on promoting employment through various measures, including vocational training, employment services and support for small and medium-sized enterprises, highlighting fair employment practices in the labor market. The Labor Contract Law 2008 aimed to protect workers’ rights by ensuring that employment contracts were in place and clearly defined. It also introduced measures to improve job security and regulate labor dispatch.

Recently, the latest amendments to the Company Law in 2023, effective from July 1, 2024, include provisions that enhance the protection of employee rights and promote employee participation in corporate governance. Further, the China Securities Regulatory Commission (CSRC) has actively promoted ESG reporting guidelines for listed firms by 2023. The guidelines mandate that listed companies disclose information on a broad range of ESG topics, including human capital/employee aspects. These changes reflect a broader commitment to integrating ESG principles into corporate practices.

Despite the recent promotion of HCM practices by China’s authorities, few studies comprehensively investigate the HCM disclosure of Chinese listed firms. Several studies examine human-capital-related disclosure as a part of their investigation of various dimensions of corporate disclosure. For example, Qu and Leung (2006) developed a corporate governance disclosure index based on the 2003 financial reports of 120 Chinese listed firms, which includes six areas: board structure and functioning, employee-related issues, director remuneration, audit committee, related party transactions and stakeholder interest. Yi and Davey (2010) constructed an intellectual capital disclosure using content analysis based on the 2006 annual reports of 49 cross-listed companies in China. Such intellectual capital disclosure has three dimensions, i.e. internal capital, external capital and human capital (including employee, education, training, work-related knowledge and entrepreneurial spirit).

Our research examines the discussion of HCM-related issues in the MD&A section of corporate annual reports, which we view as a reflection of HCM disclosure. Ocasio (1997) argues that attention is a scarce and critical resource within organizations, and where it is directed can significantly shape organizational outcomes. Becker and Gerhart (1996) suggest that management can ensure that HCM practices are aligned with the organization’s strategic goals. Similarly, Cho and Hambrick (2006) highlight that managerial attention is an important factor influencing corporate strategic change, including HCM practices. Based on this attention-based view, we propose that HCM disclosure is a key dimension of corporate HCM and is closely related to other aspects of operation, such as corporate governance and corporate social responsibility.

To measure HCM disclosure, we follow Sautner et al. (2023) to construct an HCM exposure measure based on the annual reports’ MD&A section. Sautner et al. (2023) developed a methodology that identifies the attention paid by earnings conference call participants to firms’ climate change exposures. They adopt a machine learning algorithm for keyword discovery and capture exposures to various climate change dimensions/topics.

We follow Sautner et al. (2023) and estimate the exposure of each HCM dimension using the following equation:

where $b=0,1,...,Bi,t$ are the set of phrases in the MD&A transcript of firm i in time t, and $1[·]$ is the indicator function. $CHCM_setk$ is the set of the identified HCM key phrases of HCM dimension k. Based on Batish et al. (2021), we consider ten dimensions of HCM, namely diversity and inclusion, employee development, health and safety, compensation and benefits, employee engagement, employee turnover or tenure, culture, recruiting practices, pay equity and succession planning ^[5].

To construct the HCM exposure measure, we first downloaded annual reports of firms listed on the Shanghai, Shenzhen and Beijing Stock Exchanges from the Juchao Information Network (www.cninfor.com.cn), a designated listed firms’ information disclosure website by CSRC. We then convert the downloaded annual reports from PDF to text format using Python. Text and numerical data are retained during the process, while figures and tables are discarded. After that, we isolate the MD&A section from each annual report using regular expressions and pattern-matching techniques. Consequently, we downloaded 40,841 annual reports and isolated 40,574 MD&A section transcripts.

After obtaining the transcript of the MD&A section for each firm-year, we use Jieba ^[6] to perform sentence segmentation and then remove stop words (i.e. a set of commonly used words in any language, such as “the” “is” “and” in English) and irrelevant characters (e.g. Chinese punctuation marks, dates and words starting with numerical or alphabetical values). The remaining cleaned text constitutes the informative phrase set (⁠$Bi,t$⁠) for each MD&A of the sampled firm-year.

In Mandarin Chinese, each phrase may contain one, two, three or more Chinese characters. Our HCM measure is constructed based on the number of phrases instead of Chinese characters. The HCM exposure measure for each of the ten dimensions in specification (1) is essentially a ratio of the number of phrases related to the HCM dimension (i.e. phrases in $CHCM_setk$⁠) to the total number of phrases in the cleaned-up MD&A transcript (i.e. $b=0,1,...,Bi,t$⁠). Following Sautner et al. (2023), we allow duplicated phrases when counting the number of phrases. For example, if the phrase “工资” (wage) shows up five times in the MD&A transcript, it will be counted as five phrases for both the number of phrases related to the compensation dimension (i.e. the numerator), and the number of phrases in the cleaned-up MD&A transcript (i.e. the denominator).

To construct $CHCM_set$⁠, we adopt a multi-pronged approach, incorporating both human expertise and machine learning techniques to ensure $CHCM_set$ has comprehensive coverage of keywords for each HCM dimension.

We recruited five postgraduate students with undergraduate degrees in human resource management to generate seed word lists for each of the ten HCM dimensions. The seed words were compiled from various archival sources, including government regulations (e.g. Labor Contract Law of the PRC; Guidance on Social Responsibility), industry reports (e.g. 2022 Listed Company Talent Demand and Development Environment Report; White Paper on Listed Company Value and High-quality Development) and corporate social responsibility reports of a sample of Chinese firms.

Then, we utilize five LLMs (i.e. GPT-3.5, GPT-4.0, Google Gemini, Perplexity and Perplexity Pro) to validate and expand the seed word list. Each LLM is prompted to generate key phrases for each of the ten dimensions. The generated phrases were then compiled and compared against the student-generated lists. For each HCM dimension, we take the union of the words generated from different LLMs and then take the intersection of this union with the words generated by the students to finalize the seed words for each dimension. This approach leverages the power of LLMs to identify potential keywords while ensuring that the final selection is grounded in human expertise and domain knowledge.

Then, for each seed word, we identify its synonyms using Tencent AI Lab Embedding Corpus for Chinese Words and Phrases ^[7], which is pre-trained on large-scale, high-quality Chinese data. Using the most_similar function of the Python Gensim package ^[8], we identified synonyms for the seed words based on cosine similarity, a measure of the semantic relatedness between words.

Our final HCM dictionary for each dimension includes the seed words and their synonyms with a cosine similarity score exceeding 0.75 with at least one of the seed words. While it is suggested that a cosine similarity of 0.5 or higher implies a meaningful semantic relationship between words (Kee, 2019; Erfani, Cui, & Cavanaugh, 2021), we choose a higher threshold of 0.75 to prioritize precision and ensure that the selected phrases are closely related to the seed words. In Figures A1-A10 in Appendix A, we present the word cloud for each dimension. In Table A1 in Appendix A, we tabulate the ten phrases with the highest frequency for each dimension.

With the downloaded and cleaned-up MD&A transcript and the HCM dictionary we compiled, we can construct the HCM_Exposure measure using equation (1) for each of the ten dimensions for all of the 40,574 firm-years with identifiable MD&A section transcripts. We winsorize all HCM_Exposure measures at the 1% level. Then, we perform the principal factor analysis on the HCM_Exposure of the ten dimensions and construct the HCM_Disclosure measure for each firm-year as the principal factor ^[9]. Our HCM_Disclosure has a mean of zero and a standard deviation of one by construction.

Table 1 presents the sample selection procedure and the sample composition by years. We start with a sample of listed firms in the Shanghai, Shenzhen and Beijing Stock Exchanges from 2009 to 2022, containing 44,871 firm-year observations. After deleting the firm years without annual reports from the Juchao Information Network, we have 40,841 observations. We then excluded the 267 samples with no identifiable MD&A section, resulting in 40,574 observations. After deleting firm-years without financial or stock data from CSMAR and excluding firms headquartered in regions outside mainland China, our final sample has 39,825 firm-year observations.

Table 2 presents the descriptive statistics of the HCM-related variables. The definitions are shown in Table B1 of Appendix B. By construction, HCM_Disclosure has a mean of zero and a standard deviation of one. Figure 1 presents the distribution of HCM_Disclosure, which is comparable to a normal distribution. We observe limited asymmetry (skewness = 0.603) and two short and thin tails ^[10]. On average, the firms in our sample have MD&A sections of about 34,000 phrases in length. Among these phrases, about 0.98% of the phrases are compensation-related dimensions, 0.46% are employment-development-related dimensions, and 0.33% are culture-related dimensions. The percentages of HCM-related phrases for other HCM dimensions ranged from 0.005% (HCM_Pay equity) to 0.12% (HCM_Safety and Health).

Batish et al. (2021) examined the 2020 annual reports disclosed by the first 100 companies with at least one billion USD in market capitalization in the US. Regarding the pattern of the HCM disclosure dimensions, they suggest that the most frequently disclosed dimensions in sequence are diversity, employee engagement, safety and compensation, while the least frequently disclosed dimensions are pay equity and succession planning. Our summary statistics suggest that the most frequently disclosed dimensions in China in sequence are compensation, employee development, culture and diversity, and the least frequently disclosed ones are pay equity and turnover retention. Comparing the suggested disclosure patterns in Batish et al. (2021) with that in our sample, we find that listed firms in both the US and China disclose more compensation-related, employee-development-related, and diversity-related dimensions, while pay-equity-related dimensions are often overlooked.

Table 3 presents the sample composition. Panel A shows the observation numbers and average HCM_Disclosure by years. The observation numbers were relatively small before 2012, consistent with a lack of standardization on disclosing qualitative or nonfinancial information back then. Similarly, the average values of HCM disclosure were negative during 2009–2011 ^[11]. In September 2012, the CSRC officially released “Standards for the Content and Format of Information Disclosure by Companies Offering Securities to the Public No. 2– Content and Format of Annual Reports (2012 Revision)”, stipulating more disclosure of nonfinancial and qualitative information on matters of concern to investors in the MD&A section. Subsequently, the observation numbers steadily increase during 2012–2022. The average HCM_Disclosure, on the other hand, remains stable during 2012–2020 before substantial increase in 2021 and 2022. Such an increase is likely caused by companies’ increased attention to human capital during the COVID-19 pandemic.

Panel B of Table 3 presents the observation numbers and average HCM_Disclosure by industries, sorted by the average of HCM_Disclosure. Because HCM_Disclosure has a mean of zero by construction, industries with positive figures are those with above-average HCM disclosure and vice versa. As can be observed, industries with higher HCM disclosure are often service or manufacturing industries with more skilled/educated employees, whereas those with low HCM disclosure are public utility, real estate or heavy manufacturing industries. Such a pattern is expected because, compared with the latter industries, the operation of the former industries relies more on human capital than on machinery.

We present the observation number and average HCM_Disclosure by the firms’ headquarters province in Panel C of Table 3. Companies from Guangdong, Jiangsu and Zhejiang provinces account for 38.4% of the sample observations and companies in Beijing and Shanghai make up 8.6 and 7.6% of the sample, respectively. There is also substantial variation in HCM_Disclosure across provinces, with the provincial averages ranging from −0.468 (Qinghai) to 0.168 (Sichuan).

We then divide our sample firms into four groups based on their ownership structure: private entities, foreign capital entities and local and central state-owned entities (SOEs). Our ownership structure identification is from CSMAR. Local and central SOEs are companies whose ultimate controlling shareholders are the local (municipal or provincial) and central government, respectively. Foreign capital entities are those with their equity nature being “foreign capital (外资)” in CSMAR. All other companies are private entities, and their equity nature is “private (民营)”. Panel D of Table 3 shows that 90% of our sample observations are private entities and local SOEs. Compared with foreign capital entities and central SOEs, private entities have significantly higher average HCM_Disclosure (0.127), whereas local SOEs have lower HCM_Disclosure on average (−0.259).

Table 4 presents the descriptive statistics of firm characteristics and other variables used in regression analyses. All variables are defined in Table B1 of Appendix B. The statistics are comparable to prior studies (e.g. Firth, Malatesta, Xin, & Xu, 2012; Guan, Su, Wu, & Yang, 2016; Lim, Wang, & Zeng, 2018; Chen, Ma, & Wu, 2019; Huang, Sun, & Xie, 2023; Luo, Wang, & Wu, 2023).

Panel A of Table 5 presents the correlations among HCM_Disclosure and its components. HCM_Disclosure is highly correlated with HCM_Succession Planning, HCM_Recruiting, HCM_Employment Engagement, HCM_Employee Development and HCM_Culture, with correlations above 0.50. The correlations between HCM_Disclosure and HCM_Turnover Retention, HCM_Safety and Health and HCM_Compensation were comparatively lower, ranging from 0.31 to 0.41. The correlation between HCM_Disclosure and HCM_Pay Equity (HCM_Diversity) is only 0.18 (0.20), which reflects the low correlations between HCM_Pay Equity, HCM_Diversity and other HCM dimensions, which are all below 0.10. Panels B and C of Table 5 show the correlations among HCM_Disclosure, firm characteristics, and other control variables. We find that HCM_Disclosure is positively correlated with Tobin’s Q, ROA, Cash/assets, R&D/assets, Log(employee/total assets) and negatively correlated with Leverage and Log(assets). HCM_Disclosure is weakly correlated with Sales growth, Tangibility and MTB. Consistent with Panel D of Table 3, we find HCM_Disclosure positively correlated with the private entity dummy and negatively correlated with the local SOE dummy. The provincial GDP per capita positively correlates with HCM_Disclosure, whereas industry competition, defined as the Herfindahl index of firm revenue for each industry, negatively correlates with HCM_Disclosure.

We conducted formal regression analysis to investigate the determinants of HCM_Disclosure in Panel A of Table 6. In columns (1)–(4), we focus on the explanatory power of various fixed effects. Column (1) presents the benchmark results, where we regress HCM_Disclosure on year-fixed effects only. The R-squared is 6.1%. In column (2), we include both industry and year-fixed effects, and the R-squared increases to 13.4%, indicating the strong explanatory power of industry time-invariant characteristics. We replace industry-fixed effects with province-fixed effects in column (3), and the R-squared dropped to 7.5%, which is only 1.4% higher than the benchmark in column (1). Recall that in Panel C of Table 3, there are substantial differences in HCM_Disclosure across provinces. The low incremental explanatory power of province fixed effects in column (3) of Table 6 is likely because 58.9% of sample firms are from seven provinces (Jiangsu, Zhejiang, Anhui, Guangdong, Hunan, Shandong and Beijing) which have similar average HCM_Disclosure, ranging from −0.004 to 0.132 only. In column (4), we replace province-fixed effects with firm-fixed effects, and the R-squared increases to 55.9%. The result shows that about half of the variations in HCM_Disclosure are cross-sectional and can be explained by time-invariant firm fixed effects.

In column (5), we include a set of firm-level characteristics together with industry competition and provincial GDP per capita as explanatory variables. For easier interpretation of the economic explanatory power, we standardize all continuous explanatory variables by deducting the variable mean and then dividing it by its standard deviation (Mitton, 2024). Hence, in the regression, all continuous variables, including both the dependent and independent variables, have a mean of zero and a standard deviation of one. We control for year, industry, and province fixed effects and adjust standard errors for clustering at the firm level. Our findings suggest that almost all firm-level variables are statistically significant, whereas industry competition and provincial GDP per capita are not. The results suggest that firms with high HCM_Disclosure are better-performing and have more cash holdings and R&D expenditure. These firms are also more labor-intensive and have more fixed assets, but they are less leveraged and smaller in size and are associated with lower sales growth and market-to-book ratio. Compared with private entities (the benchmark), local SOEs and foreign capital entities have lower HCM_Disclosure, whereas central SOEs have similar HCM_Disclosure.

We then replace industry and province fixed effects with firm fixed effects in column (6) and examine the association between firm characteristics and time-series within-firm variations of HCM_Disclosure. Not surprisingly, the loadings on firm size, tangibility and ownership type dummies become insignificant, whereas those on other firm characteristics remain similar. Based on the economic magnitude and the statistical significance in columns (5) and (6), we find the firm’s operating performance, R&D expenditure, labor intensity and leverage to be the strongest determinants of HCM_Disclosure.

The relationships between these firm characteristics and HCM disclosure can be explained through established theoretical frameworks. First, better-performing firms with higher cash holdings demonstrate greater HCM disclosure, as financial slack enables strategic investments in employee development and retention, which is consistent with the resource-based view of intangible asset accumulation (Mishina, Pollock, & Porac, 2004). Second, labor-intensive firms prioritize HCM due to human capital’s centrality to operational success. When labor costs constitute a large proportion of total expenses, managers face stronger incentives to optimize workforce productivity and reduce turnover costs (Ghaly, Dang, & Stathopoulos, 2020). The attention-based view predicts that structural factors like labor intensity filter managerial focus toward issues critical to value creation, making HCM a natural priority. Third, R&D-intensive firms allocate attention to HCM to retain specialized talent critical for innovation, aligning with human capital theory’s emphasis on firm-specific skills (Raffiee & Coff, 2016). Fourth, high-leverage firms exhibit reduced HCM disclosure, as debt obligations prioritize short-term financial stability over long-term intangibles, with creditors often pressuring managers toward cost-cutting measures (Acharya, Almeida, & Campello, 2013). Further, lower sales growth firms focus more on HCM disclosure as part of operational efficiency strategies (e.g. firms facing stagnant revenues may often invest in human capital to enhance productivity and offset growth limitations), which aligns with the behavioral theory of the firm, where performance shortfalls trigger a problemistic search for efficiency improvements (Greve, 2003). Finally, lower market-to-book ratios correlate with higher HCM disclosure, suggesting that undervalued firms strategically emphasize human capital to signal long-term potential. Edmans (2012) documents that markets systematically undervalue intangible investments early in their lifecycle, creating incentives for firms to double down on HCM to bridge valuation gaps.

In Panel B of Table 6, we present the results for regressing HCM_Exposures from the ten HCM dimensions on firm characteristics and industry, province and year fixed effects. To aid interpretation, all continuous variables – HCM_Exposures and controls – are standardized. Focusing on statistically significant coefficients, the signs are largely consistent across models, with a few exceptions. Notably, market-to-book ratio (MTB) and firm size [Log(assets)] are negatively associated with almost all HCM_Exposures except HCM_Compensation, implying that larger or high-growth firms favor monetary-incentive–based human capital practices. Both Local_SOE and Central_SOE exhibit negative relationships with most HCM_Exposures – saving for HCM_Safety and Health – suggesting that state-owned enterprises place greater emphasis on employee safety and health. Finally, industry competition positively correlates with HCM_TurnoverRetention and HCM_Compensation but negatively with HCM_Diversity and HCM_EmployeeEngagement. This pattern aligns with the notion that in more competitive industries, firms allocate resources toward financial incentives to retain key talent, often at the expense of diversity and engagement initiatives.

Panel A of Tables 5 and 6 shows that although HCM_Disclosure captures the bulk of the shared information across the ten HCM dimensions, it can overlook each dimension’s unique attributes. Hence, we acknowledge that while HCM_Disclosure can serve as a proxy for the overall HCM of the firm, over-reliance on HCM_disclosure alone is not encouraged.

We estimate the association between HCM disclosure and future firm performance in Table 7. We use ROA and Tobin’s Q at year t+1 to measure firm performance and control for a battery of control variables. We control for industry × year fixed effects in all regressions and add firm-fixed effects in columns (2), (4) and (6). In columns (1) and (2), we use HCM_Disclosure as the key independent variable. In columns (3) and (4), we employ the comprehensive score using the entropy method as an alternative measure of HCM_Disclosure (Li, Sun, & Zhang, 2024). In columns (5) and (6), we use HCM_Exposures from the ten HCM dimensions. Similar to Table 6, we standardize HCM_Exposures for easier interpretation.

The dependent variable in Panel A is ROA at year t+1. In columns (1) through (2), we find that HCM_Disclosure is positively associated with future ROA. In column (5), we find that HCM_Turnover Retention and HCM_Employee Development are negatively related to future ROA, whereas the coefficient estimates for HCM_Safety and Health, HCM_Succession Planning, HCM_Culture and HCM_Compensation are strongly and significantly positive. The coefficient estimates of HCM_Exposures on other HCM dimensions are not significant. In column (6), with the addition of controlling for firm-fixed effects, we find that HCM_Compensation, HCM_Succession Planning and HCM_Recruiting are significantly and positively related to future ROA. Other dimensions are generally insignificant.

These empirical results reveal nuanced relationships between specific HCM dimensions and future ROA, reflecting both the strategic value and potential complexities of human capital investments. For example, HCM_Safety and Health demonstrates a strong positive relationship with future ROA, consistent with evidence that workplace safety reduces absenteeism, healthcare costs and operational disruptions while enhancing employee productivity (Cohn & Wardlaw, 2016). The positive association between HCM_Succession Planning and ROA underscores the importance of leadership continuity in sustaining operational performance. HCM_Culture exerts a significant positive influence, suggesting that a values-driven culture reduces coordination costs and enhances decision-making efficiency, directly impacting profitability (Guiso et al., 2015). The results concerning HCM_Compensation reflect the role of efficient pay structures in attracting talent and aligning employee incentives with organizational goals (Edmans, Gabaix, & Jenter, 2017). HCM_Recruiting emerges as significant in fixed-effects models, suggesting that strategic hiring practices adapt to firm-specific needs. For instance, Tambe, Cappelli, and Yakubovich (2019) link data-driven hiring to higher innovation output and operational agility.

Conversely, the observed negative relationship between HCM_Turnover Retention and future ROA seems to be counterintuitive. However, the literature suggests that firms with very low turnover rates often retain employees who lack the skills or motivation to contribute effectively, creating operational drag and imposing significant hidden costs on productivity and innovation (Shaw, 2011). The negative relationship between HCM_Employee Development and short-term ROA suggests that, while employee development programs are widely recognized as drivers of innovation and productivity, their benefits often materialize over multi-year horizons, creating short-term productivity trade-offs (Huselid, 1995).

When we use Tobin’s Q at year t+1 as the dependent variable. The empirical results are shown in Panel B. Compared to those in Panel A, the results show different implications, except for HCM_recruiting. We find that HCM_Safety and Health and HCM_Employee Development are significantly negatively associated with future Tobin’s Q, while HCM_Turnover Retention, HCM_Succession Planning and HCM_Employee Engagement exhibit no significant relationship. These findings reflect investor skepticism toward certain human capital investments and challenges in valuing intangible assets in the short run. For example, the negative association between HCM_Safety and Health and future Tobin’s Q suggests that, while firms may view such investments as proactive risk management, markets often interpret them as signals of unresolved operational deficiencies or regulatory noncompliance (Corbet, Larkin, & McMullan, 2020). The negative association between HCM_Employee Development and future Tobin’s Q suggests that development programs depress short-term market valuations due to upfront costs (e.g. training expenses, reduced productivity during skill acquisition), even though they enhance long-term competitiveness (Tharenou, Saks, & Moore, 2007).

Furthermore, the estimates on HCM_Turnover Retention lack significance, suggesting that markets do not uniformly view retention as value-enhancing. This aligns with the argument that retention’s impact depends on who is retained: retaining high performers boosts value, while retaining low performers harms it (Shaw, 2011). Similarly, homophily-driven retention (e.g. favoring tenure over merit) may neutralize any signal of stability, as shown by Harrison and Klein (2007), who found homogeneous teams reduce innovation potential, offsetting retention’s perceived benefits. The lack of a significant relation between HCM_Succession Planning and future Tobin’s Q reflects that the market may perceive succession planning as a baseline governance practice rather than a value-enhancing differentiator. The insignificant relationship between HCM_Employee Engagement and future Tobin’Q is possibly due to measurement ambiguity, which suggests that engagement metrics may vary widely depending on survey design (e.g. question framing and scale granularity) and lack standardization and comparability across industries, thereby undermining investor confidence in engagement metrics and leading to inconsistent correlations with performance outcomes (Guest, 2011).

Taken together, Panels A and B of Table 7 provide somewhat mixed results regarding HCM disclosure and firms’ future performance: HCM disclosure is associated with better future accounting performance, measured by ROA, but poor future market valuation, measured by Tobin’s Q. Note that the results in Table 7 do not suggest any causal inference because of the obvious endogenous nature of HCM. Further research is warranted to reconcile the mixed evidence regarding HCM disclosure, operating performance and market valuation and address the endogeneity concerns.

This study set out to estimate and examine human capital management (HCM) disclosure among listed firms in China. Employing natural language processing methods, we constructed a comprehensive measure of HCM disclosure for 5,153 companies over the period 2009–2022 based on their annual reports’ management discussion and analysis (MD&A) sections.

Our analysis reveals some important findings. First, HCM disclosure has significantly increased through time, particularly in 2021 and 2022. This surge aligns with the heightened focus on human capital during the COVID-19 pandemic, which underscored the importance of employee well-being, adaptability and resilience. Second, the level of HCM disclosure varies across industries, geographic regions and ownership types. Third, firms that exhibit high HCM disclosure tend to perform better regarding accounting metrics such as return on assets but have lower market valuation, as captured by Tobin’s Q. These firms also hold more cash and invest more in R&D, indicating a strategic emphasis on long-term innovation and stability.

This study provides important implications for both practitioners and policymakers. The increased HCM disclosure and its association with firm operating performance and market valuation indicate the relevance of HCM in corporate management and underscore the need for more robust and standardized disclosure of HCM in China. Our findings support recent regulatory efforts by CSRC to enhance the transparency and accountability in HCM disclosures and advocate for more explicit and specific HCM disclosure requirements in the future.

We thank the discussant and conference participants at CAFR Special Issue Conference 2024. Gong and Lu thank the financial support from Xi'an Jiaotong-Liverpool University. All errors are our own.

The supplementary material for this article can be found online.