https://orcid.org/0009-0006-9240-4214
Figures
Abstract
While data assets are increasingly recognized as strategic resources in the digital economy, their implications for financial reporting quality remain underexplored. Employing a textual analysis approach based on deep learning techniques to construct a measure of data assets, this study investigates their impact on real earnings management (REM) using a sample of Chinese A-share listed firms from 2010 to 2023. We document a negative association between data assets and REM. Our findings remain robust after addressing endogeneity concerns using instrumental variables based on regional digital infrastructure and propensity score matching. Mechanism analyses reveal that data assets mitigate REM through two distinct channels: curbing managerial short-termism and strengthening external monitoring via increased analyst coverage. Cross-sectional tests indicate that this governance effect is more pronounced among firms facing earnings pressure. Furthermore, data assets primarily constrain cost-side manipulation strategies, specifically abnormal production costs and discretionary expenditures. Finally, we show that the reduction in REM driven by data assets contributes to enhanced long-term firm value. Collectively, these findings uncover a novel governance effect of data assets, suggesting that digital infrastructure investments yield positive externalities for corporate financial behavior.
Citation: Zhang M, Qian H, Ma T (2026) Beyond numbers: The governance role of data assets in real earnings management. PLoS One 21(7): e0351687. https://doi.org/10.1371/journal.pone.0351687
Editor: Yuan-Teng Hsu, Shanghai Business School, CHINA
Received: April 28, 2025; Accepted: May 30, 2026; Published: July 7, 2026
Copyright: © 2026 Zhang et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are available from the Figshare repository (https://doi.org/10.6084/m9.figshare.30882482) and the China Stock Market & Accounting Research (CSMAR) database (https://www.gtarsc.com).
Funding: This work was supported by the 2024 Qinglan Project for Higher Education Institutions in Jiangsu Province, awarded to MZ. The funders provided support for study supervision and writing – original draft.
Competing interests: No authors have competing interests.
1 Introduction
In the digital economy, data have evolved from an operational by-product into a strategic resource that reshapes value creation and corporate competition. Data assets differ from traditional tangible assets such as land or machinery. They are largely non-rivalrous, inexpensive to replicate, and often exhibit increasing returns to scale [1]. These attributes allow firms to improve decision quality, forecast demand and risk more accurately [2]. Recognizing this shift, regulators have begun to formalize the role of data in financial reporting. For instance, China’s Ministry of Finance issued the Interim Provisions on Accounting Treatment of Enterprise Data Resources in 2023. Following the introduction of this policy, some firms have voluntarily disclosed information of data assets in annual reports and regulatory filings to mitigate information asymmetry and facilitate price discovery [3,4].
While extensive literature examines how digital transformation affects firm performance or valuation [5,6], the implications of data assets for financial reporting quality remain underexplored. This study focuses on real earnings management (REM), a practice where managers deviate from optimal business strategies to meet earnings thresholds [7]. Compared with accrual-based earnings management, REM alters actual business activities and is therefore more opaque and more difficult for auditors and investors to detect. Although REM can generate short-term reporting benefits, extensive evidence suggests that it harms long-term performance [8]. Given the concealment advantages of REM and the limitations of traditional monitoring, identifying effective constraints on such behavior is both theoretically and practically important.
By distinguishing data assets from IT investments, we argue that data assets are an important complement to traditional governance mechanisms by reshaping the corporate information environment. Unlike general investments in IT infrastructure to facilitate interconnectivity, data assets reflect the accumulation, integration, and use of granular operational information in the context of the business process [9]. We propose two channels through which data assets discipline REM. As regards internal monitoring, data assets create a verifiable digital record of operational activities [10]. This will significantly reduce the monitoring costs that typically limit the suppression of opportunistic behavior and thus reduce the asymmetry of information on which REM is based. Meanwhile, the substantial investment in data assets provides a credible signal of long-term value creation [11]. This signals to the market that management prioritizes sustainable growth, effectively relieving short-term earnings pressure. Moreover, this focus on sustainability attracts long-term investors who discipline managerial behavior through their monitoring activities [12,13].
Using a sample of A-share listed companies on the Shanghai and Shenzhen Stock Exchanges from 2010 to 2023, we empirically examine the relationship between data assets and REM. We use a textual analysis methodology to construct a measure of data assets based on the frequency of specific keywords in annual reports. This approach allows us to capture the accumulation of data resources rather than merely IT expenditure. Our findings indicate a significant negative relationship between data assets and REM. To address potential endogeneity concerns, we apply instrumental variable (IV) methods and propensity score matching (PSM), with results remaining robust across both methodologies. Furthermore, we explore the mechanisms through which data assets exert their influence. Our analysis suggests that data assets mitigate REM primarily by reducing managerial short-termism and by enhancing analyst coverage, which strengthens external monitoring and promotes information transparency. Further analysis reveals that the mitigating effect of data assets on REM is more pronounced among firms facing greater earnings pressure. We also decompose REM into its components to identify specific channels of influence. The results indicate that data assets significantly reduce earnings manipulation related to production costs and discretionary expenditures, both of which are commonly used cost-side strategies. Finally, we demonstrate that the reduction in REM driven by data asset utilization is associated with improved firm value, underscoring the broader role of data governance in fostering sustainable corporate performance and long-term value creation.
This study makes several contributions to the literature.
First, we extend the growing body of literature on the economic consequences of data assets by exploring their impact on REM. While prior studies have primarily emphasized how data assets enhance firm performance, drive innovation, and improve market efficiency, we provide evidence that data assets perform a pivotal monitoring function that curbs opportunistic managerial behavior. By distinguishing data assets from IT investments, our study shows their unique capacity to deter REM, thus revealing a new channel through which data-driven infrastructures advance the quality of financial reporting and support the long-term sustainability of companies.
Second, we extend the literature regarding the determinants of REM. While prior studies have concentrated on traditional governance factors such as board composition and ownership structure, we shift the focus toward the influence of data assets. We complement this literature by illuminating how data assets serve as a strategic instrument that modulates managerial reporting decisions. Our findings suggest that in the digital economy, effective governance relies not only on conventional organizational structures but also on the granularity of the corporate information environment.
Third, our study empirically endorses the regulatory push for recognition of data asset. We show that capitalizing and disclosing data assets yields positive externalities that enhance governance and market efficiency. Our findings reveal that these policies are not simply administrative hurdles but are vital for mitigating information asymmetry. By establishing that data assets effectively deter earnings management, we argue for the accelerated adoption of comprehensive disclosure frameworks.
2 Literature review and hypothesis development
2.1 Literature review on real earnings management
The scope of earnings management research has evolved from the original focus on discretionary accruals to investigations of real activities manipulation [14]. While Accrual-based Earnings Management (AEM) relies on adjusting accounting estimates within the bounds of GAAP, REM involves directly altering actual operational activities, making it much more difficult to detect. Consequently, REM is often preferred in strict regulatory environments due to its opacity [15]. As regulatory oversight and financial reporting standards tighten, managers increasingly shift from AEM to REM, leveraging it as a more flexible and covert mechanism for earnings manipulation [16–18].
The motivations behind REM are usually interpreted from two competing theoretical perspectives: the agency theory and the signaling theory [8,19]. The prevailing view in existing literature is that REM stems primarily from managerial opportunism driven by agency conflicts. Given that compensation and job security are often tied to short-term stock performance or earnings targets, managers tend to sacrifice long-term shareholder value to achieve immediate results [20]. Under pressure to meet or beat analyst forecasts, avoid reporting losses, or sustain earnings growth, managers are prone to managerial myopia [21]. Conversely, the signaling perspective argues that managers utilize REM to convey private information regarding a firm’s future potential to the market [8].
Recent governance-oriented studies have highlighted the role of structural factors in mitigating agency costs. Li et al. (2025) indicates that optimizing board and ownership structures, combined with strong eco-technology, can enhance sustainability and circular economy results. This implies that a sound governance system is essential to aligning management decisions with long-term interests [22]. Similarly, Wu et al. (2025) underscores the synergy between ownership structure and financing strategies in driving innovation and environmental fulfillment [23]. However, the efficacy of these structural factors is heavily contingent upon the transparency of corporate information. REM remains a persistent challenge because operational decisions often fall within the realm of “business judgment,” rendering them opaque to traditional oversight mechanisms [24].
2.2 Literature review on data assets
Data assets are data resources owned or controlled by an enterprise that are recorded in electronic or other forms and promise future economic benefits [1,2]. This concept goes beyond the mere accumulation of raw data; it implies the systematic integration and use of data for the purpose of creating value. The early literature on the business value of information technology focused to a large extent on tangible IT investments [5]. However, recent scholarship distinguishes data assets as a specific form of digital capital that generates value through information services rather than infrastructure ownership [9,25]. Unlike traditional physical assets, data assets have a non-rivalry character, which means they can be used simultaneously across multiple algorithms and business lines without any fragmentation [26]. In the meantime, data accumulation often exhibits increasing returns to scale, allowing firms to reduce future uncertainty and maintain long-run growth [27,28].
The utilization of data assets has been linked to various positive economic outcomes. Hu et al. (2022) finds that data assets significantly enhance operational efficiency by optimizing resource allocation [9]. In terms of corporate growth, Zheng et al. (2025) shows that data assets spur corporate R&D investment by improving information integration [10], while Li et al. (2024) finds that data assets further drive sustainable development, evidenced by improved ESG performance [11]. In the context of capital markets, Sun and Du (2024) argue that data asset disclosure reduces stock price synchronicity, facilitating better price discovery [3]. Furthermore, Qian et al. (2025) finds that such disclosure reduces information asymmetry in debt markets, directly aiding firms in acquiring bank lending [29].
2.3 Hypothesis development
The relationship between data assets and REM can be theoretically based on the integration of agency theory, information asymmetry theory, and the resource-based view. REM tends to thrive in environments with high information opacity and limited monitoring [24,30]. This is because external stakeholders and even internal boards often lack granular, real-time information needed to distinguish between necessary strategic adjustments and value-destroying manipulations [14].
Data assets change the information environment within a firm. Compared to traditional physical assets, data assets generate, process and disseminate information in real time [1]. The agency theory states that monitoring provides a method to control behavior that involves opportunism [17]. Data assets reduce monitoring costs by creating a digital record of activities performed. The integration of financial systems with real-time monitoring tools allows for continuous auditing and oversight [31,32]. For example, the use of IOT technologies that interconnect devices greatly enhances supply chain transparency. This reduction in opacity limits areas where manipulative activities may remain hidden [33]. When operational metrics relating to operations that indicate inventory levels and production costs appear in real-time [7], the detection risk for managers attempting to manipulate these activities increases substantially [34].
Moreover, data assets strengthen internal governance mechanisms. Cheng et al. (2016) highlights that key subordinate executives play a crucial role in monitoring the CEO [35]. However, the effectiveness of this monitoring depends on access to high-quality information. Data assets democratize information access within the executive team. Research indicates that big data quality and diagnosticity significantly improve firm decision quality [36]. When data functions as a shared organizational asset, executives gain objective, data-driven evidence. This empowers them to challenge suboptimal CEO directives, realign managerial actions with the firm’s long-term interests, and reduce the myopic focus that often drives R&D cuts [12].
Beyond the monitoring perspective, the resource-based perspective suggests that data assets improve firm capabilities, which reduces the need for REM. A primary driver of earnings manipulation is the pressure to close the gap between actual performance and market expectations [18]. Data assets help to close this gap by driving genuine improvements. Ma et al. (2024) provides evidence that data assets are useful in optimizing the cost behavior of firms, thus improving operational efficiency [37]. Furthermore, big data analytics enhance dynamic capabilities, allowing firms to adapt to market changes more effectively [36]. When a firm can achieve earnings targets through efficiency gains and successful innovation facilitated by data assets [9,10], the pressure to resort to value-destroying manipulation diminishes.
Finally, data assets enhance external monitoring through the signaling channel. Disclosure and possession of data assets signal to the capital market that the firm maintains superior information processing capabilities and a long-term strategic orientation. Sun and Du (2024) find that disclosure of data assets significantly reduces stock prices synchronicity and facilitates better price discovery [3]. Li et al. (2024) finds that data assets drive corporate sustainable development, signaling a commitment to long-term value creation [11]. As the external information environment becomes more transparent, external stakeholders can scrutinize the firm operations more effectively, as higher coverage is linked to lower earnings management [38]. Based on these arguments, we propose the following hypothesis:
H1: Data assets are negatively associated with the extent of real earnings management.
3 Research design
3.1 Sample selection
We select A-share listed companies on the Shanghai and Shenzhen Stock Exchanges from 2010 to 2023 as our initial sample. To refine the dataset, several filtering procedures are applied. First, firms operating in the financial and insurance sectors are excluded due to their distinct financial reporting structures, which differ significantly from those of other industries. Second, firms classified as *ST or ST statuses are eliminated, as their operational conditions and strategic decisions often diverge from those of normally operating firms. Third, observations with missing values are removed to maintain data integrity. Additionally, to mitigate the influence of outliers, all continuous variables are winsorized at the 1st and 99th percentiles. This procedure ensures that statistical inferences are not driven by extreme values inherent in financial ratios. All firm-level data are obtained from the China Stock Market & Accounting Research (CSMAR) Database.
3.2 Key variable definitions
3.2.1 Real earnings management.
Following Roychowdhury (2006) [7], we first estimate the abnormal levels of production costs (Abprod), cash flow from operations (Abcfo), and discretionary expenses (Abdisx). Each abnormal indicator is derived from the residuals of its corresponding estimation model, as specified below:
Where PRODit denotes the production costs of firm i in year t; CFO𝑖𝑡 represents the operating cash flow of firm i in year t; DISXit denotes the discretionary expenses; TAit-1 is total assets at the end of year t-1; SALESit measures net sales in year t; and ΔSALESit is the change in net sales from year t-1 to year t.
REM typically manifests through abnormally high production costs, abnormally low operating cash flows, and abnormally low discretionary expenses. Recognizing that firms may employ multiple REM strategies simultaneously, we construct a composite measure of REM (REM) by aggregating these three abnormal components, as defined in Eq (4).
3.2.2 Data assets.
Although the Ministry of Finance’s 2023 Interim Provisions authorize the recognition of data resources as assets starting in 2024, relying on financial data for measurement remains impractical for empirical research. As of the 2024 reporting period, only approximately 100 A-share listed companies have recognized data resources in their financial statements, leading to severe sample selection bias and a lack of comparability. Therefore, to capture the substantive utilization of data assets across a broad sample, we adopt a textual analysis approach. Drawing on the definition provided in the White Paper (Version 6.0) (The Data Asset Management Practice White Paper (Version 6.0) was jointly issued by the Big Data Technology and Standard Committee and the China Academy of Information and Communications Technology (CAICT) in June 2023.), we select “information,” “network,” “digital,” and “data” as seed words. We then employ the Word2Vec neural network model and deep learning techniques to identify sets of semantically similar words for each seed term. To ensure semantic accuracy within a corporate context, the model is trained on a corpus comprising the annual reports of all A-share listed companies. Finally, words with the highest cosine similarity scores are retained and manually screened to ensure thematic relevance and validity.
Subsequently, we calculate the frequency of these seed words and their semantically related counterparts in firms’ annual reports. Our final measure of data assets, denoted as DA, is defined as the natural logarithm of the total frequency count. This variable serves as a proxy for the intensity of data asset accumulation and deployment. The final dictionary of keywords used to construct the data assets measure is provided in S1 Table.
3.3 Model specification
We assess whether data assets mitigate REM by estimating the following baseline regression model:
In this specification, REM denotes the level of REM for firm i in year t. The key explanatory variable, DA, captures the firm’s investment in and utilization of data assets. We control for a set of firm-specific characteristics that may influence REM outcomes. Size is measured as the natural logarithm of total assets. Lev represents the capital structure, calculated as total liabilities divided by total assets. Roa denotes return on assets, computed as net income divided by total assets. Growth is the year-over-year percentage change in operating revenue.
We also incorporate a set of governance-related variables. Dual is a dummy variable equal to 1 if the CEO concurrently serves as the board chair, and 0 otherwise. Soe is a dummy variable equal to 1 if the firm is state-owned. Top1 represents the ownership share held by the largest shareholder, capturing ownership concentration. Big4 is a dummy variable equal to 1 if the firm’s auditor is one of the Big Four accounting firms. Indep measures the proportion of independent directors on the board, while Board is defined as the natural logarithm of the total number of directors serving on the board. Mshare reflects the level of managerial ownership, and Mpay captures the total compensation of the top three executives. AEM controls for accrual-based earnings management. Year and Industry fixed effects are included to account for unobserved time-specific and sector-specific factors. Detailed variable definitions are provided in Table 1.
4 Empirical results
4.1 Descriptive statistics
Table 2 reports the descriptive statistics. The mean value of REM is −0.027, with a standard deviation of 0.224; the interquartile range spans from −0.123 to 0.104, and the maximum reaches 0.524, indicating that a non-negligible number of firms engage in REM.
The independent variable DA has a mean of 3.276 and a standard deviation of 0.696. The 75th percentile and the maximum value are 3.555 and 6.979, respectively. The right-skewed distribution of DA suggests that while many firms maintain moderate levels of data asset investment, a subset exhibits significantly higher engagement. This pattern is consistent with innovation diffusion theory, which posits that early adopters leverage emerging technologies ahead of their peers, providing a robust empirical context for examining the impact of data assets on REM.
4.2 Hypothesis tests
Table 3 presents the main regression results. In Column (1), which controls only for year and industry fixed effects, the coefficient on DA is significantly negative (−0.014, t = −4.917). This negative association becomes even stronger and remains robust (−0.017, t = −7.595) after incorporating firm-level control variables in Column (2). Beyond statistical significance, we further assess the economic magnitude of this effect by examining the sensitivity of REM to an interquartile shift in DA. Given that the interquartile range of DA is 0.665, an increase in data assets from the 25th to the 75th percentile corresponds to a reduction in REM of approximately 0.011, holding other factors constant. Relative to the sample standard deviation of REM, this reduction represents approximately 5.0%, indicating a substantial economic impact. Consistent with these calculations, Fig 1 plots this marginal effect, illustrating a distinct and monotonic downward trend in REM as data asset accumulation increases.
Note: This figure illustrates the predicted level of real earnings management across the interquartile range of data assets, with 95% confidence intervals.
Turning to the control variables, the positive and significant coefficient on Size suggests that larger firms may face greater opportunities or pressures to engage in earnings management. In contrast, Roa is negatively associated with REM, implying that firms with stronger profitability are less inclined toward earnings manipulation. Growth exhibits a significantly positive relationship with REM, indicating that firms experiencing rapid revenue growth may resort to earnings management to meet market expectations.
Regarding governance-related variables, Dual shows a negative coefficient, suggesting that CEO duality may enhance oversight effectiveness and curb opportunistic behavior. The presence of a Big Four auditor (Big4), a higher proportion of independent directors (Indep), and a larger board size (Board) are all significantly associated with lower levels of REM. These findings highlight the pivotal role of corporate governance in curbing REM activities. Finally, the significantly positive coefficient on AEM suggests that firms tend to employ both accrual-based and REM techniques simultaneously.
4.3 Endogeneity tests
4.3.1 Instrumental variable approach.
To address potential endogeneity concerns arising from omitted variable bias and reverse causality, we employ an instrumental variable (IV) approach. Following Huang et al. (2019) and Nunn and Qian (2014) [39,40], we construct the instrumental variable by interacting the number of fixed-line telephones per 100 people in the city where the firm is located in the current year with the number of internet users in the previous year.
Regarding relevance, fixed-line telephones represent the foundational telecommunications infrastructure, while the number of internet users reflects the regional maturity of digital adoption. The interaction of these two factors captures the comprehensive external environment for digital resource aggregation. A region with robust infrastructure and high internet penetration significantly lowers the costs of data acquisition and processing, thereby facilitating corporate data asset accumulation.
Regarding exogeneity, this instrument meets the exclusion restriction for two reasons. First, as a regional macroeconomic indicator determined by historical aggregate demand and urban planning, the number of telephones is exogenous to any individual firm’s specific REM decisions, thereby ruling out reverse causality. Second, the use of a lagged term for internet users effectively severs the link with current-period shocks to firm performance or financial reporting behavior, mitigating concerns regarding simultaneity bias.
The results are presented in columns (1) and (2) of Table 4. In the first stage of the regression, DA is the dependent variable. Results indicate that the coefficient for the instrument variable is positive and significant. In the second step regression, REM is the dependent variable. Results show that the instrumented DA coefficient is still negative and significant (−0.171, t = −5.795). The results indicate that after correction for possible endogeneity bias, the data source continues to have a significant REM inhibitory effect, in line with our baseline findings.
4.3.2 Propensity score matching.
To address potential endogeneity arising from self-selection, we implement a propensity score matching (PSM) approach. Firms with data asset levels above the sample median are classified as the treatment group, while those below the median constitute the control group. Propensity scores are estimated using a logit model based on the following covariates: Size, Lev, Roa, Growth, Board, Age (measured as the number of years since listing), and FC (financial constraint, measured using the FC index). We apply 1:1 nearest-neighbor matching without replacement.
The post-matching covariate balance test, presented in Table 5, indicates that most covariates achieve satisfactory balance. While there remains a marginal difference in Size between the treatment and control groups, the overall matching significantly reduces the differences across the covariates. Re-estimating the regression model using the matched sample, as reported in Column (3) of Table 4, we find that the coefficient on DA remains significantly negative (−0.019, t = −6.284), further reinforcing the robustness of our main findings.
4.3.3 Alternative clustering methods.
To address potential cross-sectional correlations at the firm and regional levels, we cluster standard errors by firm, region, region-year and firm-year, respectively. Columns (4) to (7) of Table 4 report the regression results under these alternative clustering methods. Across all specifications, the coefficient on DA consistently remains −0.017 and is statistically significant at either the 1% or 5% level, further strengthening the robustness of our main findings.
4.4 Other robustness tests
4.4.1 Alternative samples.
To ensure the robustness of our findings, we conducted two tests by altering the sample composition. First, we excluded observations prior to 2013, focusing on a period characterized by rapid internet development and widespread adoption of smart devices in China. Since 2013, technological advancements have triggered an exponential growth in data generation, prompting firms to invest more heavily in data assets to enhance management and operational efficiency. This period also coincides with the launch of China’s National Broadband Strategy, which aimed to expand broadband infrastructure and accelerate 4G network coverage nationwide. By restricting the sample to the post-2013 period, we better capture an environment where data assets became increasingly critical to corporate operations.
Second, we exclude industries with inherent data advantages to mitigate potential bias. Firms in the information or communication sectors naturally generate and process large volumes of data, granting them a distinct advantage in accumulating and leveraging data assets. To address this concern, we exclude all firms classified under the Information Transmission, Software, and Information Technology Services sectors.
Columns (1) and (2) of Table 6 present the results based on these alternative samples. In both cases, the coefficient on DA remains significantly negative at the 1% level, further strengthening the robustness of our main findings.
4.4.2 Changing the measurement of data assets.
We adopt an alternative approach following the methodology of Sun and Du (2024) [3]. Specifically, we select “data assets” and “data resources” as seed words and apply deep learning techniques using the Word2Vec neural network model to identify semantically similar terms. The top ten words with the highest similarity scores are retained to construct an expanded data asset dictionary. We then define an alternative measure, DA_1, as the frequency of these dictionary words in each firm’s annual report divided by the total word count, as specified by the following equation:
where Dictionary Worditn is the frequency of the n-th dictionary word in firm i’s annual report in year t, and Total Wordsit is the total word count of the annual report.
In addition, we construct a relative measure, DA_2, defined as the difference between a firm’s data assets and the corresponding industry-year average. This variable captures the firm’s data asset intensity relative to its industry peers. The regression results, presented in Columns (3) and (4) of Table 6, show that both DA_1 and DA_2 exhibit significantly negative coefficients at the 1% level, further strengthening the robustness of our main findings.
5 Mechanism analysis
5.1 Managerial short-termism
Building on the preceding hypothesis, we propose that data assets help mitigate REM by reducing managerial short-termism. Short-termism refers to a managerial bias toward achieving immediate financial outcomes at the expense of long-term value creation [41]. Driven by market pressures and incentive structures, such myopic behavior can be moderated by factors that enhance governance quality or extend managerial time horizons, thereby reducing reliance on earnings manipulation [35].
Data assets play a critical role in curbing managerial short-termism. First, data assets enhance internal controls and improve information transparency. By leveraging big data resources, comprehensive training datasets, and advanced recognition algorithms, firms can generate more accurate earnings forecasts and provide real-time information to stakeholders [42]. This increased transparency raises the cost of opportunistic manipulation by amplifying the risk of detection, thereby discouraging myopic behavior. Moreover, enhanced informational clarity alleviates the uncertainty encountered by managers, which constitutes a key driver of short-term decision-making under performance pressure [43].
Second, data assets help broaden managerial time horizons. By integrating key business dimensions—such as production efficiency, customer engagement, market pricing, and sales performance—into centralized, data-driven systems, firms gain a comprehensive view of their operations. Machine learning models and predictive analytics highlight long-term growth opportunities and guide more efficient resource allocation, aligning firm performance with sustainable value creation [44]. In such a data-intensive ecosystem, the marginal benefit of engaging in REM diminishes, as decisions become increasingly evidence-based rather than tactic-driven. Consequently, managers are incentivized to prioritize durable performance outcomes over short-term financial optics.
To empirically examine whether data assets mitigate REM by reducing managerial short-termism, we construct a Myopia Index through textual analysis of the Management’s Discussion and Analysis (MD&A) sections in annual reports. Drawing on the short-term orientation lexicon developed by Brochet et al. (2015) [45], we adapt and expand this word list to better align with the Chinese linguistic and business context. First, we manually select a set of seed words categorized into direct expressions (e.g., “within days,” “several months,” “immediately,” “at once”) and indirect expressions (e.g., “opportunity,” “occasion,” “pressure,” “challenge”) that subtly imply a short-term mindset. Next, we employ the Continuous Bag-of-Words (CBOW) model within the Word2Vec framework to train word embeddings on a corpus of annual financial reports. By calculating vector similarities, we identify additional words closely related to our seed terms, ultimately forming an expanded vocabulary representing short-term orientation. Finally, we compute the Myopia Index (Short-termism) as the frequency of short-term-oriented words divided by the total word count of the MD&A section, multiplied by 100. A higher index value indicates a greater degree of managerial short-termism.
Columns (1) and (2) of Table 7 present the baseline mechanism analysis using the text-based Myopia Index. In Column (1), where managerial Short-termism is the dependent variable, the coefficient on DA is significantly negative (−0.009, t = −20.153), suggesting that firms with higher levels of data assets tend to exhibit a lower degree of short-term orientation. In Column (2), with REM as the dependent variable, DA remains significantly negative (−0.016, t = −7.000), while Short-termism enters with a significantly positive coefficient (0.126, t = 3.354). This positive association indicates that greater managerial myopia is indeed a driver of REM.
To validate these findings and address potential concerns regarding the subjectivity of textual analysis, we introduce an objective, asset-based proxy for short-termism. Drawing on the work of Demir (2009) [46], we construct Shortinv, defined as the proportion of short-term investments (trading financial assets, available-for-sale financial assets, and held-to-maturity investments) to total assets. A higher ratio implies a managerial preference for liquidity and short-term financial performance over long-term value creation. Columns (3) and (4) report the results of this validation test. In Column (3), the coefficient on DA is significantly negative (−0.004, t = −4.038), indicating that data assets are associated with a reduction in the allocation of short-term assets. Furthermore, in Column (4), Shortinv is positively associated with REM (0.046, t = 2.621), while the negative effect of DA on REM remains robust (−0.016, t = −7.338).
Taken together, results from both the text-based and asset-based measures provide consistent empirical support for the proposed mechanism: data assets mitigate REM by curbing managerial short-termism.
5.2 Analyst coverage
A firm’s proactive disclosure of data assets conveys a strong strategic signal to external stakeholders, highlighting its capacity to manage data as a core organizational resource [29]. Prior research suggests that detailed and comprehensive corporate disclosures attract greater analyst coverage by reducing information asymmetry and enhancing firm transparency [47]. In the context of the digital economy, where data assets are increasingly recognized as critical drivers of long-term value, such disclosures provide analysts with valuable inputs for assessing firm-specific risks, opportunities, and strategic capabilities.
Financial analysts serve as influential external monitors within capital markets, and their continuous scrutiny imposes market discipline on corporate managers [48]. Firms subject to more intensive analyst coverage face greater accountability for financial reporting quality, as analysts not only disseminate firm-specific information but also act as gatekeepers for investors and regulators. Empirical evidence confirms that firms under heightened analyst scrutiny are less likely to engage in opportunistic earnings manipulation due to improved transparency and external monitoring pressure [38]. Consequently, we posit that the disclosure of data assets attracts increased analyst attention, which in turn constrains managers’ propensity to engage in REM.
To empirically test the proposed mechanism, we construct the variable Analyst coverage, measured as the natural logarithm of the number of financial analysts following a firm. Columns (3) and (4) of Table 7 present the results of the mechanism analysis investigating whether analyst coverage mediates the relationship between data assets and REM. In Column (3), the regression results show that DA has a significantly positive effect on Analyst coverage (0.093, t = 9.315), indicating that firms with greater engagement in data assets attract more attention from financial analysts. In Column (4), where REM is the dependent variable, DA continues to exhibit a significantly negative coefficient (−0.014, t = −6.414), while Analyst coverage also shows a significant and negative association with REM (−0.028, t = −16.934). These findings suggest that analyst coverage acts as a mediating channel through which data assets reduce REM, providing further empirical support for the proposed mechanism.
6 Further analysis
6.1 Moderating effect of data assets on real earnings management under earnings pressure
We further investigate the moderating effect of data assets on REM in the presence of earnings pressure. Specifically, we focus on situations where firms fail to meet analyst consensus earnings forecasts, a widely used proxy for capital market performance pressure. Firms that fall short of market expectations are more likely to experience negative investor reactions and thus face stronger incentives to engage in earnings manipulation [49]. However, data assets can help mitigate such opportunistic behavior by enhancing internal controls and improving transparency [4, 9]. Accordingly, the governance effect of data assets is expected to be more pronounced among firms facing greater earnings pressure.
To empirically test this expectation, we introduce an interaction term between DA and an indicator variable (Miss_forecast) that equals one if the firm fails to meet the analyst consensus earnings forecast. The regression results, presented in Column (1) of Table 8, show that the coefficient on Miss_forecast is significantly positive (0.037, t = 2.679), indicating that firms missing market expectations are more likely to engage in REM. More importantly, the significantly negative coefficient on the interaction term Miss_forecast*DA (−0.012, t = −2.802) indicates that data assets weaken the positive impact of missing analyst forecasts on REM. In other words, the presence of stronger data assets mitigates the tendency of firms under earnings pressure to engage in REM.
6.2 Impact of data assets on components of real earnings management
This subsection examines how data assets affect individual components of REM, namely abnormal production costs (Abprod), abnormal operating cash flows (Abcfo), and abnormal discretionary expenses (Abdisx). By disaggregating REM into these three components, we aim to provide a more nuanced understanding of the channels through which data assets influence managerial financial reporting behavior. Theoretically, we expect a negative relationship between DA and Abprod, indicating reduced manipulation of production costs. Conversely, we anticipate positive relationships between DA and both Abcfo and Abdisx, reflecting reduced manipulation in cash flows and discretionary expenses, respectively.
Columns (2), (3), and (4) of Table 8 present the regression results. In Column (2), the coefficient of DA on Abprod is −0.011 (t = −6.972), consistent with the expectation of reduced production cost manipulation. In Column (3), the coefficient of DA on Abcfo is −0.003 (t = −3.387), which contradicts the expected positive relationship. In Column (4), the coefficient of DA on Abdisx is 0.009 (t = 8.984), significant at the 1% level, aligning with the theoretical prediction. These findings suggest that data assets primarily mitigate REM by reducing the propensity for cost-side operational manipulation, particularly in production activities and discretionary expenditures. However, the observed negative association between DA and Abcfo introduces an intriguing nuance. This result may reflect structural changes in operational cash flow patterns, potentially driven by more data-informed sales strategies or the adoption of more conservative revenue recognition practices.
6.3 The effect of data assets on long-term performance through reduced real earnings management
REM is often employed to inflate reported earnings by manipulating operational activities. However, such practices can undermine investment efficiency and distort resource allocation, ultimately impairing long-term firm value [21]. This subsection examines whether data assets enhance long-term firm performance by mitigating REM. We use Tobin’s Q as a forward-looking measure of firm value, capturing investors’ expectations regarding the firm’s growth potential.
Columns (5) and (6) of Table 8 report the regression results. In Column (5), the coefficient on DA is significantly positive (0.055, t = 3.467), suggesting that firms with greater engagement in data assets tend to exhibit higher long-term performance. In Column (6), after controlling for REM, the coefficient on DA remains significantly positive (0.042, t = 2.678), while the coefficient on REM is significantly negative (−0.758, t = −11.848). These findings indicate that data assets not only directly contribute to long-term value creation but also indirectly enhance long-term performance by mitigating REM.
7 Conclusions
This study examines whether data assets can mitigate REM. Using a panel of 20,770 firm-year observations from A-share listed companies in China between 2010 and 2023, we document a significantly negative association between data assets and REM, which remains robust across multiple endogeneity and robustness tests.
Our mechanism analysis sheds light on the underlying processes driving this relationship: data assets curb REM by weakening managerial short-termism and by attracting greater attention from financial analysts, thereby enhancing both internal discipline and external oversight. Further analyses reveal that the mitigating effect of data assets on REM is particularly pronounced among firms facing earnings pressure, especially those failing to meet analyst forecasts. Moreover, our results show that data assets primarily mitigate REM by reducing manipulation in production costs and discretionary expenses, rather than by influencing operating cash flows. Finally, we find that the reduction of REM through the use of data assets contributes to enhanced long-term firm performance.
Our findings offer several important practical implications. For firms, they underscore the strategic value of investing in data infrastructure, not only for operational efficiency but also as a component of a robust governance framework. For policymakers, our results suggest that promoting standardized and transparent data assets reporting can strengthen external governance. This aligns with recent policy initiatives, such as China’s 2023 Interim Provisions on Accounting Treatment of Enterprise Data Resources. By formalizing data as a balance-sheet asset, this policy is likely to amplify the transparency and monitoring effects we document. In jurisdictions without such mandatory disclosure requirements, our findings imply that voluntary disclosure of data assets can serve as a credible signal of governance quality, allowing high-quality firms to distinguish themselves. Furthermore, the governance mechanisms identified in this study are likely to be strengthened by the adoption of AI-driven audit tools. As noted in prior literature, technologies that enable continuous, full-population transaction analysis can leverage corporate data assets to automate the detection of operational anomalies in real time. This technological synergy significantly raises the detection risk for managers attempting to conceal REM. Finally, from a strategic perspective, the deployment of data assets should be viewed not merely as a compliance exercise but as a value-enhancing initiative that fosters financial discipline and sustainable competitive advantage.
This study is subject to several limitations. First, its focus on a single country means the findings may not be fully generalizable to economies with different institutional environments, legal frameworks, and levels of digital development. Second, our measure of data assets relies on a textual-frequency proxy. While grounded in established methods, this approach captures the disclosure of data assets and may not perfectly reflect the quality or efficiency of a firm’s actual data utilization capabilities. Future research could address these limitations through cross-country comparative studies to examine how varying institutional factors moderate the governance role of data assets.
Supporting information
S1 Table. The Dictionary of Data Asset Keywords.
https://doi.org/10.1371/journal.pone.0351687.s001
(XLSX)
References
- 1. Xie K, Wu Y, Xiao J, Hu Q. Value co-creation between firms and customers: the role of big data-based cooperative assets. Information & Management. 2016;53(8):1034–48.
- 2. Lin C, Zhang M. A review of the impact of data assets on the operation and development of enterprises. JRACR. 2024;14(2).
- 3. Sun X, Du Z. Enhancing capital market efficiency: the role of data assets disclosure in reducing stock price synchronicity. International Review of Economics & Finance. 2024;94:103351.
- 4. Wei Y, Zhang J, Cheng M, Liu T. Does data asset disclosure contribute to the market efficiency? Evidence from China. Research in International Business and Finance. 2025;73:102549.
- 5. Vial G. Understanding digital transformation: a review and a research agenda. Managing Digital Transformation. 2021. p. 13–66.
- 6. Wang S, Zhang H. Enhancing SMEs sustainable innovation and performance through digital transformation: insights from strategic technology, organizational dynamics, and environmental adaptation. Socio-Economic Planning Sciences. 2025;98:102124.
- 7. Roychowdhury S. Earnings management through real activities manipulation. Journal of Accounting and Economics. 2006;42(3):335–70.
- 8. Gunny KA. The relation between earnings management using real activities manipulation and future performance: evidence from meeting earnings benchmarks*. Contemporary Accting Res. 2010;27(3):855–88.
- 9. Hu C, Li Y, Zheng X. Data assets, information uses, and operational efficiency. Applied Economics. 2022;54(60):6887–900.
- 10. Zheng X, Zhang X, Yang G. Can data assets spur corporate R&D investment?. Applied Economics Letters. 2025;32(4):538–45.
- 11. Li Y, Wang X, Zheng X. Data assets and corporate sustainable development: evidence from ESG in China. Pacific-Basin Finance Journal. 2024;85:102378.
- 12. Bushee BJ. The influence of institutional investors in myopic R&D investment behavior. The Accounting Review. 1998;73(3):305–33.
- 13. Harford J, Kecskés A, Mansi S. Do long-term investors improve corporate decision making?. Journal of Corporate Finance. 2018;50:424–52.
- 14. Dechow PM, Skinner DJ. Earnings management: reconciling the views of accounting academics, practitioners, and regulators. Accounting Horizons. 2000;14(2):235–50.
- 15. Ewert R, Wagenhofer A. Economic effects of tightening accounting standards to restrict earnings management. The Accounting Review. 2005;80(4):1101–24.
- 16. Cohen DA, Dey A, Lys TZ. Real and accrual-based earnings management in the pre- and post- Sarbanes-Oxley periods. The Accounting Review. 2008;83(3):757–87.
- 17. Cohen DA, Zarowin P. Accrual-based and real earnings management activities around seasoned equity offerings. Journal of Accounting and Economics. 2010;50(1):2–19.
- 18. Vorst P. Real earnings management and long-term operating performance: the role of reversals in discretionary investment cuts. The Accounting Review. 2015;91(4):1219–56.
- 19. Habib A, Ranasinghe D, Wu JY, Biswas PK, Ahmad F. Real earnings management: a review of the international literature. Accounting & Finance. 2022;62(4):4279–344.
- 20. Graham JR, Harvey CR, Rajgopal S. The economic implications of corporate financial reporting. Journal of Accounting and Economics. 2005;40(1–3):3–73.
- 21. Stein JC. Efficient capital markets, inefficient firms: a model of myopic corporate behavior. The Quarterly Journal of Economics. 1989;104(4):655.
- 22. KaoDui L, Kongkuah M, Alessa N. Beyond compliance: how board–ownership structures and eco‐technology intensity drive circular economy performance in MENA firms. Corp Soc Responsibility Env. 2025;32(5):6658–79.
- 23. Ning W, Saeed UF, Kongkuah M. Saving the environment in emerging markets: the synergistic roles of corporate ownership structure, financing strategy, and innovation capacity. Bus Strat Env. 2025;34(4):5114–38.
- 24. Kothari SP, Mizik N, Roychowdhury S. Managing for the moment: the role of earnings management via real activities versus accruals in SEO valuation. The Accounting Review. 2015;91(2):559–86.
- 25.
Tambe P, Hitt L, Rock D, Brynjolfsson E. w28285. National Bureau of Economic Research; 2020.
- 26. Jones CI, Tonetti C. Nonrivalry and the economics of data. American Economic Review. 2020;110(9):2819–58.
- 27. Farboodi M, Veldkamp L. Long-run growth of financial data technology. American Economic Review. 2020;110(8):2485–523.
- 28. Brynjolfsson E, McElheran K. The rapid adoption of data-driven decision-making. American Economic Review. 2016;106(5):133–9.
- 29. Qian K, Pan C, Liang X. Does data asset disclosure affect bank lending acquisition?. International Review of Financial Analysis. 2025;97:103834.
- 30. Kothari SP, Shu S, Wysocki PD. Do managers withhold bad news?. J of Accounting Research. 2009;47(1):241–76.
- 31. Vasarhelyi MA, Kogan A, Tuttle BM. Big data in accounting: an overview. Accounting Horizons. 2015;29(2):381–96.
- 32. Alles MG, Kogan A, Vasarhelyi MA. Putting continuous auditing theory into practice: lessons from two pilot implementations. Journal of Information Systems. 2008;22(2):195–214.
- 33. Ezekiel Onyekachukwu U, Prisca A, Kudirat Bukola A, Anwulika Ogechukwu S. The role of IoT in boosting supply chain transparency and efficiency. Magna Sci Adv Res Rev. 2024;12(1):178–97.
- 34. Appelbaum D, Kogan A, Vasarhelyi MA. Big Data and Analytics in the Modern Audit Engagement: Research Needs. Auditing: A Journal of Practice & Theory. 2017;36(4):1–27.
- 35. Cheng Q, Lee J, Shevlin T. Internal Governance and Real Earnings Management. The Accounting Review. 2015;91(4):1051–85.
- 36. Ghasemaghaei M, Calic G. Can big data improve firm decision quality? The role of data quality and data diagnosticity. Decision Support Systems. 2019;120:38–49.
- 37. Ma J, Yang X, Li M. Are data assets useful? Evidence based on corporate cost stickiness. Emerging Markets Finance and Trade. 2024;60(15):3734–52.
- 38. Yu F (Frank). Analyst coverage and earnings management. Journal of Financial Economics. 2008;88(2):245–71.
- 39. Huang Q, Yu Y, Zhang S. Internet development and productivity growth in manufacturing industry: Internal mechanism and China experiences. China Industrial Economics. 2019;(8):5–23.
- 40. Nunn N, Qian N. US Food aid and civil conflict. American Economic Review. 2014;104(6):1630–66.
- 41. Dallas LL. Short-termism, the financial crisis, and corporate governance. Journal of Corporation Law. 2011;37:265.
- 42. Lin X, Wang L. How do enterprise big data applications mitigate asset mispricing? Finance Research Letters. 2025;79:107256.
- 43. Francis J, LaFond R, Olsson P, Schipper K. The market pricing of accruals quality. Journal of Accounting and Economics. 2005;39(2):295–327.
- 44. Wamba SF, Gunasekaran A, Akter S, Ren SJ, Dubey R, Childe SJ. Big data analytics and firm performance: effects of dynamic capabilities. Journal of Business Research. 2017;70:356–65.
- 45. Brochet F, Loumioti M, Serafeim G. Speaking of the short-term: disclosure horizon and managerial myopia. Rev Account Stud. 2015;20(3):1122–63.
- 46. Demir F. Financial liberalization, private investment and portfolio choice: financialization of real sectors in emerging markets. Journal of Development Economics. 2009;88(2):314–24.
- 47. Bushee BJ, Noe CF. Corporate disclosure practices, institutional investors, and stock return volatility. Journal of Accounting Research. 2000;38:171.
- 48. Chen J, Cumming D, Hou W, Lee E. Does the external monitoring effect of financial analysts Deter Corporate Fraud in China?. J Bus Ethics. 2014;134(4):727–42.
- 49. Bartov E, Givoly D, Hayn C. The rewards to meeting or beating earnings expectations. Journal of Accounting and Economics. 2002;33(2):173–204.