A data science framework for profit health assessment: development and validation

Abstract

This study introduces a cutting-edge profit health assessment framework that merges signaling theory and agency theory within a data science context, leveraging both financial and nonfinancial indicators to provide a comprehensive, multidimensional evaluation of earnings quality in publicly traded companies. This model transcends conventional earnings management frameworks by focusing on the holistic earnings condition of firms and addressing challenges related to information asymmetry and managerial motives. The empirical analysis utilizes data from small and medium-sized private enterprises listed on China’s SME Board between 2018 and 2022, employing machine learning algorithms to rigorously validate the model’s effectiveness. The results reveal that 85.7% of the penalized firms by the China Securities Regulatory Commission from 2018 to 2023 exhibited low levels of profit health. This data science driven analysis deepens the understanding of corporate earnings health for regulators and investors. The proposed framework not only expands the theoretical underpinnings of earnings management but also serves as a novel evaluative tool for stakeholders, paving the way for broader application across diverse markets and sectors and rethinking earnings quality assessment methodologies through a data science lens.

1 Introduction

In the contemporary landscape of corporate governance, characterized by swift transformations in economic and regulatory contexts, traditional strategies centered on profit maximization have proven increasingly inadequate. This shift has necessitated a pivot towards more integrative measures of financial health, thereby catalyzing the development of the “profit health” concept. This study introduces a pioneering framework that enhances accounting theory with an advanced assessment system, integrating critical financial indicators with vital nonfinancial metrics closely associated with profit outcomes. This methodological innovation not only opens new vistas for assessing corporate financial health, but also significantly bolsters the applicability and impact of signaling theory [30] and agency theory [37] in modern corporate governance.

This research focuses on Private Listed Companies on China’s SME Board, which are pivotal within the Chinese capital markets due to their significant roles and are noted for their substantial growth potential and adaptability despite facing considerable market pressures and transparency issues. Through a detailed analysis of annual reports from 2018 to 2022, this investigation delves into how these firms manage and articulate their financial performance under fluctuating economic and market conditions. This comprehensive scrutiny not only clarifies their strategies for sustaining financial performance and transparency but also deepens our understanding of their profit health, thereby emphasizing the crucial interdependence between financial robustness and regulatory compliance.

Chinese President Xi Jinping introduced the groundbreaking concept of “new quality productivity,” redefining the research trajectory towards optimizing profit health. This concept underscores the imperative role of technological progression and innovation in strengthening an enterprise’s core competitiveness. It not only signifies the modernization of productivity but also represents a radical transformation of traditional productivity frameworks. Within this strategic context, the China Securities Regulatory Commission has specifically advocated for technology enterprises to drive their growth through innovative advancements, thereby catalyzing the superior development of the national economy. On the regulatory front, the “National Nine Articles” policy, enacted by the China Securities Regulatory Commission on April 12, 2024, highlights the critical importance of fortifying governance structures and enhancing transparency among publicly listed companies. It compels enterprises to proactively navigate the new challenges posed by economic restructuring and rapid technological evolution. This policy plays a pivotal role in stabilizing the capital market’s growth and elevating the transparency of market information, thereby ensuring the protection of investor interests and fostering enhanced market efficiency.

This paper contributes to the literature in several significant ways. First, it defines the concept of “profit health,” providing a novel perspective for evaluating corporate financial performance, thereby strengthening the applicability and practical effectiveness of signaling theory and agency theory within modern corporate governance. Second, by integrating the econometric models of Beneish [3] and Jansen et al. [23] with agency theory, it develops a robust framework for profit health assessment, thus expanding the scope of research on earnings management measurement. Third, the study introduces four diagnostic indicators with strong explanatory power, which can effectively detect and predict profit health issues in listed companies, offering investors and regulatory bodies more powerful tools for monitoring corporate performance. Lastly, by employing a distinctive penalty sample test, the study further substantiates the validity and scientific rigor of the profit health evaluation framework.

2 Literature review

Earnings management is the act of influencing financial information, especially short-term earnings, through the choice of different accounting policies and estimates. It can be achieved through legitimate accounting decisions or fraudulent means [19]. It aims to meet or exceed the earnings expectations of capital market analysts, improve borrowing conditions, increase corporate stock prices, or achieve management bonus targets. Research on earnings management has explored, to varying extents, its significant factors, identification models, key indicators, and the application of new technologies.

2.1 Exploration of important factors in earnings management

In terms of influencing factors, corporate manipulation of earnings can affect investors, creditors, loan conditions, market efficiency, government regulation, and contractual incentives. Through earnings management, companies signal their future growth and profitability, affecting stock prices and market valuations. When a company manipulates financial statements through earnings management, investors may be misled and unable to accurately assess the company’s financial situation and future performance, which impacts investment decisions [9]. Creditors may also struggle to accurately assess credit risk and repayment ability, reducing trust in the company and affecting loan conditions [46]. If such earnings management is revealed or if financial conditions deteriorate, both investors and creditors may face higher investment risks and losses [44]. Investors might misinterpret the company’s true financial condition, leading market prices to deviate from the company’s real value and affecting market effectiveness [18]. Earnings management during equity offerings can also mislead investors when they make decisions, causing an overvaluation of stock values and decreased capital allocation efficiency [55], thus adversely affecting the effective operation of capital markets.

In some cases, listed companies might engage in earnings management to meet capital market regulatory requirements, such as profitability requirements for share issuance [6]; maintain stock exchange listing qualifications; or avoid breaching financial ratio requirements in loan agreements [19]. Earnings management may also be related to internal reward mechanisms, such as management bonuses and incentive plans, which are often linked to a company’s financial performance. Management might actively implement earnings management to achieve rewards [16]. While earnings management may help companies achieve certain goals in the short term, its long-term effects can be negative, especially when the company’s true financial condition and operational performance are concealed as a result [9]. Understanding the motivations for, techniques of, and impacts of earnings management is crucial for investors, analysts, regulators, and scholars.

2.2 Exploration of earnings management identification models

The identification of earnings management has long been a key area of academic research, with models such as the Beneish M-score and the Modified Jones model standing out as widely adopted tools. These models, grounded in a variety of financial indicators and theoretical foundations, are indispensable for detecting potential earnings manipulation. The Beneish M-score model [3] computes the M-score by aggregating a series of financial ratios to signal the probability of earnings manipulation when the score surpasses a given threshold. This model is firmly rooted in signaling theory, where managers convey strategic signals to the market via financial disclosures. The M-score effectively acts as a decoder, enabling market participants to interpret these signals and assess the risk of managerial misconduct. A high M-score serves as a red flag, indicating potential earnings manipulation and aligning with the notion that market participants interpret signals to infer management’s strategic behavior. Thus, the M-score not only highlights financial anomalies but also reflects the broader dynamics of information asymmetry in financial reporting. Studies like those of Kaur et al. [26], Nguyen and Nguyen [41], and Hieu et al. [20] have successfully employed the M-score model to uncover determinants of earnings management in publicly listed firms, reinforcing the practical application of signaling theory in uncovering market anomalies.

By comparison, the Modified Jones model by Healy and Wahlen [19] offers a more comprehensive approach, enhancing the original Jones model by incorporating corporate governance variables, thus extending its utility within the framework of agency theory. Agency theory posits a fundamental conflict of interest between shareholders (principals) and managers (agents), with managers potentially incentivized to manipulate earnings to their advantage, often at shareholders’ expense. The Modified Jones model captures these managerial motivations by incorporating governance mechanisms—such as board independence and audit quality—that are designed to mitigate the risk of earnings manipulation. This incorporation makes the Modified Jones model more robust in contexts where governance plays a pivotal role, offering a dual-layered analysis that decodes both financial signals and the impact of corporate governance on managerial opportunism. Empirical studies by Supardi and Asmara [45], Marchellina and Firnanti [36], Harahap [17], Li and Cao [29], and Qiao and Tang [42] have utilized this model to investigate the extent to which governance structures and regulatory environments either curb or exacerbate earnings management, thereby highlighting the interaction between governance dynamics and agency conflicts.

The original Jones model [25], which focuses on nondiscretionary accruals to estimate the extent of earnings manipulation, remains foundational in the field. This model assumes that managers can influence reported earnings through the strategic use of accruals, with nondiscretionary accruals serving as a key mechanism for such manipulation. It draws directly from agency theory, emphasizing how managerial incentives are tied to short-term performance metrics and how such incentives may encourage earnings manipulation. Goel [15] expands on this framework with the DeAngelo model, which estimates discretionary accruals as a proxy for earnings manipulation, providing another valuable tool for researchers seeking to quantify the scope of earnings management.

Despite their widespread use, these models are not without limitations. A key concern is their reliance on assumptions about managerial behavior and the nature of financial signals, which can vary significantly across different institutional and regulatory contexts. For instance, assumptions regarding the predictability of discretionary accruals may not hold uniformly across diverse legal systems, leading to discrepancies in the models’ accuracy and reliability in detecting earnings manipulation. Additionally, governance structures themselves are not homogeneous across markets; the efficacy of corporate governance mechanisms, such as board independence or the role of external auditors, can differ based on regulatory regimes and cultural factors, which further complicates the application of these models. This underscores the pressing need for more refined models that can account for the dynamic and evolving nature of agency conflicts and the varied signaling mechanisms managers employ. Future research should aim to develop models that are more adaptable to different regulatory frameworks and governance systems, providing more context-sensitive tools for detecting earnings management. Such advancements will significantly enhance the precision and applicability of earnings management detection in a globalized corporate landscape, offering more reliable insights across varying market environments.

2.3 Exploration of selection criteria for earnings management indicators

In terms of indicator selection, a series of studies involving the Beneish M-score and Modified Jones models have adequately demonstrated the effectiveness of financial indicators such as sales revenue [34], accounts receivable [50], asset turnover ratio [26, 43], net cash flow from operating activities [22], net profit [1], and gross profit margin [2] in measuring earnings management in listed companies. Moreover, Jones [25] noted that total assets, gross income, and total fixed assets are determinants of nondiscretionary accruals. Binsaddig et al. [4] suggested that companies should pay more attention to inventory turnover and accounts receivable turnover rates to more effectively and efficiently optimize profits and improve financial performance. Du [13] also examined how certain *ST companies adjust total profits through nonrecurring earnings to achieve delisting, as well as exploring the role of financial indicators in detecting earnings management. Additionally, some scholars recommend adding qualitative performance indicators and using further mathematical models to measure earnings management [1]. Others suggest using types of audit opinions as evidence of the existence of earnings management because independent audit opinions can signal to the outside world the quality of a company’s accounting earnings [11, 48, 54]. These studies cover financial and nonfinancial indicators affecting earnings management, providing important references and insights that enable a deeper understanding and effective responses to earnings management behaviors.

2.4 Exploration of new theories and methods

When exploring new theories and methods, research on identifying earnings management has not only enhanced the capabilities of traditional models but also introduced new approaches such as machine learning and data analysis; these are capable of handling large datasets and learning to identify complex patterns of potential earnings management. Algorithms such as support vector machines (SVM), random forests, and neural networks have been used to predict the feasibility of earnings manipulation [27]. With the rapid development of artificial intelligence technology, text analysis techniques have been applied to financial reports and publicly disclosed information. By analyzing the language style and sentiment tendencies found in the management discussion and analysis sections, potential earnings management can be identified. Researchers have found that when engaging in earnings management, management teams might use more positive or vague language to mask or beautify financial conditions [21, 32]. Other academics have combined machine learning and text analysis techniques to analyze large amounts of financial reports and publicly disclosed texts, empirically demonstrating the relationships that earnings management has with digital operation levels, internal control levels, and other factors [31, 33]. Social network analysis has provided another new perspective by analyzing networks of relationships between companies and their auditors, board members, and other businesses to identify the likelihood of earnings management [8]. The emergence of these new theories and methods has not only broadened the horizons of earnings management research but also improved the precision and efficiency with which earnings management can be identified, enabling regulators, auditors, and investors to more effectively monitor and evaluate a company’s financial health.

These achievements provide researchers with valuable information to understand the existing methods of research on earnings management in listed companies, but some gaps remain in terms of considering the diversified influencing factors of earnings management, in-depth discussion on model limitations, and comparative analysis between methods. On this basis, therefore, this study first explores the intrinsic logic affecting profit health; secondly, it systematically and cumulatively scores the medium- and long-term change signals and nonfinancial indicator signals affecting profit health, thereby proposing an innovative profit health assessment model; finally, it empirically visualizes the model based on machine learning algorithms.

3 Theoretical model and research hypotheses

3.1 Theoretical model

In exploring the construction of a theoretical framework for assessing corporate profit health, this study delves into deeper factors such as the rationality and stability of profit growth. Typically, the continuous enhancement or weakening of a company’s profitability is a dynamic process formed by the interweaving and accumulation of multiple factors; this is not achievable overnight. For instance, the sustained growth of a company’s profits might result from the management’s efforts over many years to implement comprehensive measures including, but not limited to, cost control, such as refined cost management, supply chain optimization, automation, and technological innovation; revenue growth [52] such as market expansion, product and service diversification, pricing strategy optimization, and marketing strategy innovation; tax strategies [56] such as tax planning, utilizing losses for tax deductions, international tax planning; and financial strategies such as asset restructuring, financial leverage adjustment, cash flow management, and mergers and acquisitions. Conversely, a continuous decline in profits might be caused by a variety of complex factors such as a deterioration of the macroeconomic environment and poor operational management. Management teams of listed companies often do not want to face the reality of a significant profit decline. Concerned about stock price performance, regulatory risk warnings, impaired financing capabilities, and the implementation of option incentive plans, they may consider a series of “legal” or “seemingly reasonable” measures to adjust their company’s profit level. Some companies may even disregard the risk of punishment by the Securities Regulatory Commission and engage in false actions to forge or tamper with profit data.

Over the past 30 years, the academic community’s research on earnings manipulation behavior has continued to deepen and innovate. The research findings have primarily focused on false revenue recognition [57], improper expense deferral or capitalization [5], malicious accounting estimates and judgments [47], profit manipulation through reserves [12], unrealistic asset and liability evaluations [35], related-party transaction manipulations [51], and other nefarious means. Through theoretical analysis, case studies, as well as empirical research on the motivations, implementation methods, and impacts of these behaviors, this study not only deepens the understanding of earnings manipulation but also provides practical guidance for identifying and addressing manipulative behaviors in financial reporting.

In their seminal work, Jansen et al. [23] introduced a methodological innovation for diagnosing earnings management, focusing on variations in asset turnover and profit margins. This approach eschews the traditional reliance on correlations, favoring instead a robust reconciliation between accounting equations and financial statements. Signaling theory assumes that in a market environment characterized by information asymmetry, company management can convey valuable information to external investors through financial reports and other disclosures [38]. In this study, signaling theory is used to explain how companies utilize financial indicators, such as profit growth rate and accounts receivable growth rate, to signal their internal operational conditions and profitability. On the other hand, agency theory focuses on the interest asymmetry and potential conflicts between agents (e.g., company management) and principals (e.g., shareholders) [24]. The application of agency theory in this study is primarily reflected in how financial decision-making can be optimized to balance the conflicting interests between management and shareholders, thereby maximizing corporate profits and ensuring sustainable growth. Building on a foundation of rigorous academic research, this study employs a carefully curated set of financial indicators, including operating revenue, accounts receivable, asset turnover, net cash flow from operating activities, net profit, gross profit margin, inventory turnover rate, gross income, and total profit. Collectively, these indicators provide a multifaceted view of the sources of profits, the efficiency of operational management, and liquidity capabilities.

In the model presented in this study, the intersection between signaling and agency theories lies in the evaluation of corporate financial health. Signaling theory helps explain how management manipulates financial indicators, such as profit health, to convey positive signals to the market. However, agency theory further suggests that these signals may be influenced by conflicts of interest between management and shareholders. For instance, management might employ short-term financial strategies to improve company performance for their own benefit, thereby sending misleading positive signals to external stakeholders [14]. Consequently, this study’s model integrates the market response provided by signaling theory and the behavioral insights offered by agency theory, aiming to construct a more comprehensive and dynamic framework for evaluating corporate financial health. As a result, this research incorporates changes in accounting firms (NCAF) and audit opinions (AO) as nonfinancial indicators to delve into the nuances of information asymmetry between corporate management and financial information users. According to signaling theory, audit opinions not only mitigate this asymmetry but also enhance the transparency of the financial conditions, affirming the authenticity and compliance of the enterprises. Frequent changes in accounting firms, conversely, may suggest an attempt by management to circumvent rigorous audit scrutiny. Moreover, by analytically decomposing seven financial indicators into four paired groups and scoring them based on their annual divergent directions, this methodology provides a nuanced and granular assessment of profit health, enabling stakeholders to gauge the varying degrees of financial robustness and transparency across listed companies.

Based on the concept of profit health classification and building upon the discussion in the Introduction, this study theoretically defines profit health as the continuous, stable, and reasonable growth in profits that enterprises achieve by adhering to market principles and legal regulations while leveraging technological innovation, optimizing resource allocation, improving total factor productivity, and promoting industrial transformation and upgrading. This profit growth should genuinely reflect the enterprise’s operational performance, support sustainable development, and create value for shareholders. The concept of new-quality productivity emphasizes revolutionary technological breakthroughs, innovative allocation of production factors, and profound industrial transformation. It fundamentally involves qualitative changes in labor, tools of production, and objects of labor, as well as their optimized combination, with the improvement in total factor productivity as the core indicator. Thus, healthy profits are the direct outcome of new-quality productivity development rather than the result of short-term financial maneuvers. Furthermore, third-party audit opinions provide an additional layer of assurance regarding the authenticity.

Inspired by the research findings of Jansen et al. [23], Maslow’s Hierarchy of Needs Theory, and the Balanced Scorecard Theory developed by Kaplan et al., an innovative profit health assessment theoretical model is proposed in this study. This model divides profit health into four levels, whereby level four represents the highest degree of profit health and level one indicates the lowest degree of profit health. The model includes scores from four financial indicator signal change detectors and two nonfinancial indicator score, summing up the profit health scores over a five-year span. The constructed theoretical model is depicted in Fig. 1.

Figure 1

Profit Health Assessment Theoretical Model

As shown in Fig. 1, from the perspective of agency theory, management may change accounting firms multiple times within five years to obtain more favorable audit opinions in response to the board’s performance evaluation and to maintain a high market value. This is done to send positive signals about the company’s performance to the external environment, thereby influencing external investors’ decision-making behavior.

Detector I₁ focuses on whether the change direction between Operating Revenue Growth Rate (RGR) and Accounts Receivable Growth Rate (ARGR) is consistent, the aim being to identify potential anomalies in revenue growth. According to basic accounting principles, if revenue growth does not accompany a corresponding increase in accounts receivable, it is marked as an anomaly, scoring 1 point; if both grow in sync, 0 points are scored.

Detector I₂ focuses on whether the change direction between RGR and Net Cash Flow Growth Rate from operating activities (NCFGR) is consistent; this is used to assess the real value of revenue. If the growth in revenue is inconsistent with the growth in net cash flow from operating activities, it is deemed an anomaly, scoring 1 point; if this growth is consistent, 0 points are scored.

Detector I₃ focuses on whether the change direction between Net Profit Growth Rate (NPGR) and Inventory Turnover Growth Rate (ITR) is consistent in order to judge the rationality of net profit growth. According to accounting theory, an increase in the inventory turnover rate should promote company profit growth; therefore, if the inventory turnover rate is inconsistent with net profit growth, it is viewed as an anomaly, scoring 1 point; if it is consistent, 0 points are scored.

Detector I₄ focuses on whether the change direction between Gross Profit Growth Rate (GPGR) and Gross Revenue Growth Rate (GRGR) is consistent, the aim being to detect signs of manipulation in total profit. If gross revenue grows while gross profit decreases or if gross profit grows while gross revenue decreases, both are deemed abnormal phenomena, scoring 1 point; otherwise, 0 points are scored.

Although these four detectors score independently, they form a closely related logical chain. In detector I₁, a growth of accounts receivable relates directly to the authenticity of revenue growth, which further affects the increase in gross revenue in detector I₄, thereby promoting an increase in total profit. Meanwhile, an increase in total profit will affect the growth of net profit in detector I₃. To verify the authenticity of net profit growth, it is necessary to check whether the direction of net profit growth is consistent with the growth direction of the inventory turnover rate in I₃. The consistency between the growth direction of net cash flow from operating activities and the growth direction of revenue in detector I₂ is used to assess the liquidity of revenue. This series of detections not only interlock with each other but also collectively form the basis for a deep analysis of a company’s financial health condition.

Although financial indicators derived from publicly accessible annual reports are objective, they are not inherently comprehensive. Consequently, this model incorporates a scoring system based on the annual AO issued by third-party accounting firms over the last five years. The scoring assigns 0 for a “standard unqualified opinion,” 1 for a “qualified opinion with emphasis of matter,” 3 for a “qualified opinion,” and 5 for either a “disclaimer of opinion” or an “adverse opinion.” Additionally, a score of 1 is attributed to the noncompliance with auditor firm change (Number of Accounting Firm Changes, NCAF) if there is a change in the accounting firm compared to the previous year; otherwise, a score of 0 is maintained. By synthesizing data from five years across four financial indicators and two nonfinancial indicators, the company’s Profit Health (PH) score is computed, leading to the proposition of hypothesis H0.

3.2 Research hypotheses

In Fig. 1, the Profit Health Assessment Theoretical Model includes seven financial indicators and two nonfinancial indicators that are closely related to PH. The key question is whether these nine variables indeed have a significant impact on PH and to what extent. To address this, the following hypothesis is proposed:

H0: RGR, ARGR, NCFGR, NPGR, ITR, GRGR, GPGR, NCAF, and AO have a significant impact on PH.

To rigorously validate and clearly delineate the visual progression of how four specific detectors impact profit health, four hypotheses have been crafted based on detailed discussions:

Firstly, consistent with [10], a substantial relationship exists between revenue and accounts receivable. Typically, an upsurge in revenue triggers a corresponding rise in accounts receivable, encapsulating revenue within receivables before their conversion to actual cash flows. This parallel movement generally signals a company’s adeptness at converting sales into receivables, thereby fortifying its operational and financial stability. Conversely, a simultaneous increase in revenue with a reduction in accounts receivable might indicate a boost in cash sales or expedited payments from customers, which points to an improved receivables management. Alternatively, it could hint at an attempt to enhance the company’s perceived financial health by accelerating revenue recognition or minimizing provisions for doubtful debts [44]. Such divergences in directional trends might uncover potential issues in the company’s collection practices, warranting a prudent evaluation of looming risks to profit health. Hence, hypothesis H1 is proposed.

H1: Discrepancies in the growth rates between operating revenue and accounts receivable are indicated as potential signals of unhealthy profits, with the frequency of changes in accounting firms and the nature of audit opinions also impacting profit health.

Secondly, substantial empirical evidence demonstrates a significant linkage between operating revenue and net operational cash flows, as evidenced by Mohammad [39]. An increase in operating revenue generally correlates with enhanced net cash flows, suggesting robust profit health, where both indicators typically exhibit synchronous movements. However, Lee et al. [28] identified pronounced disparities between operating revenue and net cash flows in entities engaged in fraudulent activities. These discrepancies might suggest the use of aggressive accounting techniques designed to artificially inflate reported net profits, adversely affecting profit health. Consequently, hypothesis H2 is proposed.

H2: Variations between the growth rates of operating revenue and net cash flows from operating activities could potentially signal issues with profit health. Moreover, the frequency of changes in accounting firms along with the nature of audit opinions exert considerable influence on the financial well-being of the company.

Thirdly, findings by Winda and Nasution [49] substantiate a significant and positive correlation between inventory turnover and net profit. A heightened inventory turnover rate is generally associated with increased net profits, stemming from more effective sales conversions, whereas a diminished turnover rate often corresponds with decreased profitability. Hence, an elevated turnover rate is instrumental in bolstering net profits. Concurrently, increased profits might compel firms to expand investments and scale up production capabilities, which can reciprocally influence inventory turnover. Typically, inventory turnover and net profit changes exhibit parallel trajectories. Conversely, if inventory turnover escalates while net profit declines, this could indicate practices of earnings management, such as postponing cost recognition or manipulating inventory valuations—suggesting that discrepancies in the trends between inventory turnover and net profits might be indicative of manipulative financial reporting practices [19]. Accordingly, hypothesis H3 is proposed.

H3: Disparities in the growth rates of net profits versus inventory turnover rates could signal underlying disturbances in profit health, a situation further compounded by the rate of changes in accounting firms and the characteristics of audit opinions.

Fourthly, changes in gross income invariably influence total profits, and a parallel trajectory between these metrics signifies effective financial management, enabling the transformation of revenue into tangible profits. In contrast, divergent movements between these metrics might suggest that companies engage in unorthodox practices to manipulate their declared profits through complex strategies in expense, indirect cost, and tax management, thereby detrimentally impacting the health of their profits. Accordingly, hypothesis H4 is proposed.

H4: A divergence between the growth trajectories of total profits and gross income could be indicative of potential fiscal health issues, with the dynamics surrounding changes in accounting practices and audit opinions exerting significant influence.

If the empirical validation of these hypotheses demonstrates a statistically significant impact on PH, it affirms the robustness of the theoretical model proposed in this study for quantifying profit health, particularly suited for automated calculations by computer systems. In support of the automated analysis by such systems, this study delineates four distinct gradational intervals of profit health based on comprehensive data analytics, thereby augmenting the financial acumen of accountants, auditors, and investors in recognizing and addressing financial manipulations [7, 40].

4 Research design

4.1 Data source

For this investigation, small and medium-sized private enterprises listed on the A-share main board in China as of December 31, 2018, were selected from the Wind database as the research subjects. This study compiled a panel dataset incorporating essential financial and nonfinancial indicators from 2018 to 2022. To ensure the robustness of the data, companies designated as ST or *ST were excluded, resulting in a dataset comprising 580 companies with a total of 2900 records (5 years × 580 companies). For the panel data cleaning process, this study adopts the following approach: First, when calculating growth rate indicators such as RGR, ARGR, NCFGR, NPGR, ITR, GRGR, and GPGR, if the base value of any indicator is missing, the indicator is assigned a value of zero. Second, for each indicator, if its value deviates significantly (by more than 50 times) from the average value of that indicator, the corresponding record is removed (for example, the RGR for 2018_002798.SZ is 7.0731, which is 53.4 times the average RGR value of 0.1324). Based on these criteria, 24 abnormal records were deleted: 1 for RGR, 1 for ARGR, 2 for NCFGR, 2 for NPGR, 4 for ITR, 12 for GPGR, and 2 for GRGR, resulting in a final dataset of 2876 valid records.

4.2 Variable description

Guided by the theoretical model depicted in Fig. 1 and hypotheses H0 to H4, this study developed four detectors (I₁ to I₄) to assess profit health along with a comprehensive profit health indicator. The functionality and rationale behind each detector are detailed in Table 1. For a detailed breakdown of the scoring methods applied to I₁ and AO from 2018 to 2022 for selected companies, refer to Appendix A.

Table 1 Detectors and Explanations of Variables

• ARGR: Accounts Receivable Growth Rate, used to measure the efficiency of a company’s sales growth and accounts receivable management.

• RGR: Operating Revenue Growth Rate, indicating the percentage growth in the company’s annual revenue.

• NCFGR: Net Cash Flow from Operating Activities Growth Rate, measuring the growth in cash generated from the company’s operating activities.

• NPGR: Net Profit Growth Rate, showing the change in the company’s annual net profit.

• ITR: Inventory Turnover Growth Rate, reflecting the efficiency of the company’s inventory management.

• GRGR: Gross Revenue Growth Rate, assessing the changes in the company’s gross income.

• GPGR: Gross Profit Growth Rate, representing the changes in the company’s gross profit.

• NCAF: Number of Accounting Firm Changes, used to assess the continuity and credibility of the company’s financial reporting.

• AO: Audit Opinion, based on the rating of different audit opinions.

• PH: Profit Health, an overall score calculated by aggregating the above indicators.

4.3 Descriptive statistics

According to Table 2, the standard deviations of variables RGR, ITR, GRGR, NCAF, and AO are all less than 1, indicating that these four variables exhibited relatively low volatility from 2018 to 2022; the standard deviations of ARGR and PH are respectively 1.583 and 1.264, which are also relatively low. However, the volatility of NCFGR, NPGR, and GPGR between 2018 and 2022 was considerably higher, with minimum values ranging from −69.434, −100.647, and −108.904 to maximum values of 312.859, 96.833, and 238.881.

Table 2 Detailed Statistical Descriptions of All 10 Variables

From the perspective of fundamental accounting theory, ARGR is closely related to RGR, RGR to NCFGR, NPGR to ITR, and GRGR to GPGR, hence using multiple linear regression modeling would result in severe multicollinearity. Therefore, this study considers employing a nonlinear regression model for empirical hypothesis testing. The random forest algorithm, which combines random sampling with decision trees through multiple rounds of randomly selecting sample sizes and feature values to build an ensemble of decision trees, applies the principle of the majority rule to make decisions, ensuring the effective reduction of decision errors. Additionally, the random forest algorithm is a machine learning method that performs iterative calculations on existing models, potentially enhancing the accuracy of model predictions [53].

Although the random forest algorithm does not have a clear mathematical analytical expression, it can be represented as follows:

$$ f \left ( x \right ) = \frac{1}{M} \sum _{m=1}^{M} T_{m} (x) $$

(1)

In Eq. (1), $f ( x )$ represents the predicted result, and $T_{m} ( x )$ denotes the prediction result of the mth decision tree. This study has chosen the random forest algorithm as the modeling tool and the gradient boosting machine algorithm for robustness testing comparisons based on the following three considerations:

• Observations of the sample revealed that the data obtained in this study are not normally distributed, making them unsuitable for multiple linear regression. The random forest and gradient boosting machine algorithms do not require specific data distributions or data types and are tolerant of noise and outliers, making them highly compatible with the data structure of this study [58].

• The random forest and gradient boosting machine algorithms can uncover variables with interactive effects and various complex nonlinear relationships. They allow for the assessment of the importance of variables influencing profit health, measuring the impact of different variables, which aligns well with the research content of this study.

• This study needs to investigate the explanatory power of each detector on profit health. The partial dependence plots generated by the random forest algorithm facilitate researchers in visually observing the dynamic evolution of how various variables affect profit health.

4.4 Model construction and robustness tests

4.4.1 Overall model construction and robustness testing

The primary parameters were fine-tuned to the following settings: n_estimators = 300, max_depth = 20, min_samples_split = 2, min_samples_leaf = 1, and random_state = 42. To ensure the model’s robustness, both the gradient boosting and random forest algorithms were developed under the same parameter settings, as shown in Table 3. (The Python code for the random forest model can be found in Appendix B; the code for the gradient boosting machine is structurally similar to that of the random forest and is therefore omitted.)

Table 3 Empirical Results and Robustness Tests for Hypothesis H0

As delineated in Table 3, both the Random Forest and Gradient Boosting Machine models show nearly perfect $R^{2}$ values close to 1 on the training sets and maintain low Mean Squared Errors (MSE). The Random Forest model exhibited an $R^{2}$ value of 0.9425 and an MSE of 0.0971 on the test set, indicating that the model possesses strong predictive ability. Although the rankings of feature importance differ between the gradient boosting and random forest models, the features with an importance exceeding 15% are consistent across both, with the top-ranked feature, AO, showing marginal variation in its relative importance. Cross-validation between the two models reveals that the Random Forest model marginally outperforms the Gradient Boosting Machine model. Consequently, adopting the Random Forest algorithm for modeling proves to be both feasible and robust. Furthermore, the Random Forest model’s feature importance rankings indicate that all nine independent variables contribute more than 5% to PH thereby substantiating Hypothesis H0.

4.4.2 Development and robustness assessment of submodels

To maintain consistency and ensure the robustness of our approach, the principal parameters of the random forest models for the four detectors were aligned with those of the overarching model. The outcomes are detailed in Table 4. Analysis from Table 4 indicates that on the training set, all four detectors achieved $R^{2}$ values around 0.9, with the Mean Squared Errors (MSE) not exceeding 0.2, suggesting effective model fits. Concerning the test sets, while the random forest models for Detectors I₁ and I₂ show mild overfitting (MSEs greater than 0.9 but less than 1), Detectors I₃ and I₄ exhibit strong generalization abilities with MSEs below 0.72. Notably, across all four models, the importance ratings for AO oscillate between 25% and 26%, and NCAF fluctuate from 6.61% to 6.89%. Additionally, the importance attributed to each pair of financial indicators within the detectors ranges from 67.20% to 68.00%, indicating a consistently high explanatory power for financial metrics across the board.

Table 4 Random Forest Models and Tests for Four Detectors

To ascertain the robustness of the random forest models, the same principal parameters were utilized to develop gradient boosting machine models for the four detectors, as depicted in Table 5. The results in Table 5 show that all four gradient boosting models achieved nearly perfect $R^{2}$ values close to 1 and MSEs approaching zero on the training sets. However, on the test sets, $R^{2}$ values fell below 0.4, and MSEs exceeded 1, indicating significant underperformance, although there were marginal improvements in the models for Detectors I₃ and I₄. Consequently, given the unchanged parameters and the poor generalization capabilities observed, this study prefers the random forest algorithm for modeling. The results from the random forest models for the four detectors, as presented in Table 4, preliminarily confirm Hypotheses H1 to H4.

Table 5 Gradient Boosting Machine Models and Tests for Four Detectors

5 Empirical visual analysis

5.1 Visual empirical analysis of detector I ₁

Figures 2(a) and 2(b) demonstrate that when both the RGR and ARGR are negative, Profit Health (PH) remains at a relatively elevated level. As these financial indicators gradually shift from negative towards zero, there is an observable increase in PH, suggesting a growing risk to profit health. In contrast, when both RGR and ARGR are positive and increasing, an escalation in RGR leads to a rapid decrease in PH, indicating that as revenue growth accelerates, profit health improves. However, a continual rise in ARGR slightly enhances PH, hinting that the efficiency of accounts receivable management may not be keeping pace with the expansion of sales operations, thus potentially heightening the risk to profit health. From Figs. 2(c) and 2(d), it is evident that more frequent changes in accounting firms and poorer audit opinions correlate with higher PH, signaling deteriorating profit health.

Figure 2

Partial Dependency Plots for Detector I₁’s Random Forest

Therefore, when the RGR and ARGR transition from negative to positive, it is crucial for information users to meticulously scrutinize whether there is a significant divergence between the company’s disclosed customer payment terms, discounts, and promotions, and the actual circumstances, particularly when the AO is considerably adverse. This discrepancy necessitates a cautious approach in evaluating the company’s investment strategies. As such, Hypothesis H1 is validated, indicating that discrepancies in the impacts of RGR and ARGR on PH may signal potential risks to financial health, further exacerbated by changes in accounting firms and the severity of audit opinions.

5.2 Visual empirical analysis of detector I ₂

Figures 3(a) and 3(b) reveal that when both the RGR and the NCFGR are negative, there is an accumulation of risk to profit health. As illustrated in Fig. 3(a), when RGR swiftly shifts towards positive values, there is a corresponding decrease in profit risk, which is mirrored by an increase in NCFGR. From Fig. 3(b), as NCFGR progressively increases in positive territory, PH experiences slight fluctuations within a constrained range, likely influenced by variations in AO. When both RGR and NCFGR approach zero from negative values, especially with a heightened AO, stakeholders can assess the company’s profit health through the lens of changes in sales policies, efficiency of accounts receivable management, and the efficacy of fund utilization. Therefore, Hypothesis H2 is substantiated, suggesting that inconsistencies in the effects of RGR and NCFGR on PH might be indicative of underlying profit health issues, which are further influenced by the frequency of changes in accounting practices and audit opinions.

Figure 3

Partial Dependency Plots for Detector I₂’s Random Forest

5.3 Visual empirical analysis of detector I ₃

From Fig. 4(a), as the NPGR transitions from negative to positive, there is an observed increase in PH, indicating an escalating risk to profit health potentially influenced by a lower ITR. As NPGR continues its ascent toward positive values, significant fluctuations in PH suggest possible earnings management activities by the management. Moreover, as depicted in Fig. 4(b), when ITR consistently rises in positive territory, PH generally declines, indicating healthier profit levels. Additionally, the trajectories of the NCAF and AO depicted in Figs. 4(c) and 4(d), which align with those observed in Figs. 2(c) and 2(d), and Figs. 3(c) and 3(d), indicate that increased frequency of changes in accounting firms or more adverse audit opinions correlate with heightened risks to profit health. When both NPGR and ITR are negative and the AO is significantly high, information users should scrutinize the company’s inventory management, stock, or sales efficiency as disclosed, to gauge profit health risks. Thus, Hypothesis H3 is supported, suggesting that discrepancies in the impacts of ITR and NPGR on PH may signal potential profit health risks, and that changes in accounting practices and audit opinions influence profit health.

Figure 4

Partial Dependency Plots for Detector I₃’s Random Forest

5.4 Visual empirical analysis of detector I ₄

From Figs. 5(a) and 5(b), when both the GRGR and the GPGR are negative, PH is observed to range between 1.5 and 2.2, indicative of a relatively robust level of profit health. As demonstrated in Fig. 5(b), when GPGR shifts into positive territory, there is a rapid alleviation of profit health risks, likely attributed to substantial improvements in GRGR. This correlation is evident in Fig. 5(a), where PH experiences a steep ascent as GRGR transitions from negative to zero. This rapid surge in gross income may result from strategic pricing adjustments or significant reductions in operational costs by the management. Moreover, as GRGR progressively increases, a continuous enhancement in the company’s profit health is evident. Additionally, as shown in Figs. 5(c) and 5(d), the frequency of changes in accounting firms or the severity of audit opinions has a concurrent rising impact on profit health risks. Therefore, in scenarios where both GPGR and GRGR are negative and either the NCAF or AO is high, it is imperative for information users to meticulously assess the company’s disclosures on cost control, sales figures, and adjustments in revenue recognition practices, cost measurement methods, or asset valuations. This analysis is crucial for evaluating potential earnings management. Consequently, Hypothesis H4 is substantiated, revealing that inconsistencies in the impacts of GPGR and GRGR on PH may signify underlying issues with profit health, influenced by changes in accounting practices and audit opinions.

Figure 5

Partial Dependency Plots for Detector I₄’s Random Forest

6 Further validation

The panel dataset employed in this study is restricted to a continuous five-year period of data from main board-listed companies in China’s A-share market, which introduces the limitation of incomplete data coverage across other market segments. In future research, we intend to expand the sample to include all market segments. Furthermore, identifying whether more effective indicators exist for assessing the profit health of listed companies remains an area for further investigation in future studies.

To enhance the consolidation and verification of the four submodels’ reliability and utility, this study aggregated the panel data from 580 sample companies for the years 2018 to 2022. Each company’s total PH score over five years was calculated and subsequently grouped into intervals based on their PH scores. The formula for calculating the total PH score for company i (where j represents the year and i represents the company) is provided in Eq. (2):

$$\begin{aligned}& \quad \qquad PH_{i} = \sum _{j =2018}^{2022} ( RGR_{ij} \& \ ARGR_{ij} + RGR_{ij} \&\ NCFGR_{ij} + NPGR_{ij} \ \& \ ITR_{ij} +\\& GPGR_{ij} \ \& \ GRGR_{ij} + NCAF_{ij} + AO_{ij} ), i =1,2,\dots ,580, j =2018,\ 2019,\dots ,\ 2022. \end{aligned}$$

(2)

In Eq. (2), $PH_{i}$ represents the sum of the PH scores for the ith company from 2018 to 2022; $RGR_{ij}$ and $ARGR_{ij}$ denote the Operating Revenue Growth Rate and Accounts Receivable Growth Rate, respectively, for company i in year j. If both variables share the same sign, the value is set to 1; otherwise, it is set to 0; $NCFGR_{ij}$ represents Net Cash Flow from Operating Activities Growth Rate for company i in year j. Similarly, if the signs of $RGR_{ij}$ and $NCFGR_{ij}$ are identical, the value is set to 1; otherwise, 0. $NPGR_{ij}$ and $ITR_{ij}$ stand for the Net Profit Growth Rate and Inventory Turnover Growth Rate, respectively, for company i in year j, and their matching signs also result in a value of 1; otherwise, 0; $GPGR_{ij}$ and $GRGR_{ij}$ refer to the Gross Profit Growth Rate and Gross Revenue Growth Rate, respectively, for company i in year j, with the same condition: a value of 1 if their signs align, and 0 if not. Lastly, $NCAF_{ij}$ indicates whether company i changed its accounting firm in year j, assigning a value of 1 if the firm was changed, and 0 if it remained the same; $AO_{ij}$ represents the Audit Opinion score for company i in year j (the scoring criteria are detailed in the last paragraph of Sect. 3.1).

Based on these calculations, the study divided the 580 companies into four groups according to their total PH scores: [0, 4], [5, 9], [10, 14], and [15, +∞). Each group corresponds to a different level of profit health, with four stars denoting the highest level and one star the lowest. The distribution of these PH score groups is illustrated in Table 6.

Table 6 Profit Health Grouping Statistics

Table 6 reveals that from 2018 to 2022, the distribution of the total PH scores among the 580 small and medium-sized private listed companies generally shows smaller values at the extremes and larger values in the middle, corroborating the theoretical model depicted in Fig. 1. Specifically, companies with three and two stars collectively represent 86.38% of the total; those with the healthiest profits comprise 9.31%, exceeding the 4.31% of companies at the lowest health level. This indicates that nearly 10% of these companies possess high investment value, whereas nearly 5% warrant cautious avoidance.

Additionally, Table 6 visually demonstrates that as the profit health level decreases, the AO notably increases. The proportions of companies with higher AO scores range from 1.9% in four-star companies to 60.0% in one-star companies, further validating the hypotheses H1 to H4, which propose that poorer audit opinions are associated with lower levels of profit health.

• The first group has a total PH score ranging from 0 to 4, classified as four-star, comprising 54 companies, accounting for 9.31% of the total.

• The second group has a total PH score between 5 and 9, classified as three-star, comprising 360 companies, accounting for 62.07%.

• The third group has a total PH score between 10 and 14, classified as two-star, comprising 141 companies, accounting for 24.31%.

• The fourth group has a total PH score ranging from 15 to 28, classified as one-star, comprising 25 companies, accounting for 4.31%.

Additionally, to further validate the insights of this study into profit health, we reviewed administrative penalty records issued by the China Securities Regulatory Commission from 2018 to 2023 for violations such as improper disclosure of information, related-party transactions, and falsified reporting. We found that of the penalized private listed companies on the SME board, 20 were identified, 6 of which had been delisted and were not included in this study. Of the remaining 14 companies, 12 were identified by the profit health assessment model as high risk, representing 85.7%, as shown in Table 7.

Table 7 2018 to 2023 Penalties and Profit Health Levels of SME Board Private Enterprises

According to the results from Table 7, firstly, the profit health assessment model identified 85.7% of the penalized entities, demonstrating high assessment efficiency given the complexity and obscurity of identifying manipulative financial information practices. Secondly, the lower the profit health level, the higher the likelihood of a company engaging in financial information manipulation. From Table 4, as profit health decreases from three-star to one-star, the proportion of penalized companies in their respective groups jumps from 0.56% to 20%, indicating an exponential increase, further validating the utility of the profit health assessment model. Therefore, it is advised that information users pay close attention to the company’s profit health ratings. If a company is found to have a low profit health rating or a decline relative to the previous period, it indicates a very high risk to profit health.

7 Discussion & conclusions

This study presents a novel profit health assessment framework that integrates financial and nonfinancial indicators, offering a comprehensive evaluation of listed companies’ earnings quality. By incorporating signaling and agency theories, the model allows stakeholders to gauge both the financial soundness of companies and the potential managerial incentives behind financial manipulations. The model has been rigorously tested using panel data from Chinese private enterprises listed on the SME board between 2018 and 2022, and the findings provide strong empirical support for its effectiveness in identifying earnings management and detecting profit health risks.

7.1 Discussion

A key finding of this research is that inconsistencies between core financial indicators, such as operating revenue growth and accounts receivable growth, are indicative of underlying profit health risks. This observation aligns with signaling theory, where managers, driven by short-term incentives, may attempt to signal financial stability to investors through accounting adjustments. However, these signals may be misleading, particularly in cases where audit scrutiny is weaker, evidenced by frequent changes in accounting firms. The results strongly support the hypothesis that companies with frequent auditor switches exhibit greater financial risk, reinforcing the need for regulatory attention in such cases.

Additionally, the analysis revealed that discrepancies between net profit growth and inventory turnover are often symptomatic of earnings management, particularly through cost recognition or inventory valuation manipulations. This finding echoes agency theory, which emphasizes the divergence of interests between managers and shareholders. Management may prioritize actions that temporarily inflate financial metrics to meet performance targets or secure bonuses, at the expense of long-term corporate health. This insight is crucial for investors and auditors, as it underscores the importance of closely monitoring operational metrics, especially when financial results appear overly optimistic.

The inclusion of nonfinancial metrics, such as audit opinions, provides an additional layer of granularity in assessing profit health. The empirical evidence demonstrates that adverse audit opinions are closely associated with declining profit health, further validating the integration of such qualitative factors into the model. This is particularly relevant in the context of China’s evolving regulatory landscape, exemplified by the “National Nine Articles” policy, which emphasizes the need for improved transparency and governance among listed companies. The findings of this study align well with these regulatory goals, offering practical tools for enhancing corporate oversight and investor protection.

7.2 Conclusions

In conclusion, this research contributes to the literature on earnings management and corporate financial health by developing a robust and versatile profit health assessment framework. The model successfully integrates both financial metrics and non-financial indicators, offering a more holistic view of corporate health than traditional financial models alone. The findings underscore the significance of audit opinions and accounting firm changes as critical signals for external stakeholders, complementing conventional financial metrics in assessing earnings quality and corporate governance.

The proposed model not only enhances the theoretical understanding of profit health but also provides a practical tool for investors, regulators, and auditors. As corporate governance and financial transparency become increasingly prioritized in global markets, the adaptability of this model makes it a valuable asset for evaluating profit health across various industries and regulatory contexts. Future research should aim to further refine the model by incorporating additional non-financial metrics, particularly in international settings where regulatory environments and market structures differ. Moreover, advancements in machine learning and AI present opportunities to enhance the model’s predictive capabilities, allowing for real-time assessments and more dynamic decision-making processes.

Ultimately, this study highlights the critical role that a comprehensive, multidimensional assessment of profit health can play in strengthening corporate governance and protecting investor interests, particularly in an era of increasing regulatory scrutiny and global market integration.

Appendices

Appendix A: Examples of I1 and AO scores for certain listed companies between 2018 and 2022

See Table 8.

Table 8 Examples of I1 and AO scores for certain listed companies between 2018 and 2022

Appendix B: Complete random forest model Python code

import pandas as pd

from sklearn.model_selection import train_test_split

from sklearn.ensemble import RandomForestRegressor

from sklearn.metrics import r2_score, mean_squared_error

# Load data

data = pd.read_csv(‘D:/zwdata.csv’)   # Ensure to replace with the correct path

Y = data[‘PH’]

X = data.drop(‘PH’, axis = 1)

# Split the dataset into training and testing sets

X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size = 0.2, random_state = 42)

# Initialize the Random Forest regressor with determined optimal parameters

rfr = RandomForestRegressor(

  n_estimators = 300,

  max_depth = 20,

  min_samples_split = 2,

  min_samples_leaf = 1,

  random_state = 42

)

# Train the model

rfr.fit(X_train, Y_train)

# Perform predictions on both the training and testing sets

Y_train_pred = rfr.predict(X_train)

Y_test_pred = rfr.predict(X_test)

# Calculate R² and MSE for both training and testing sets

train_r2 = r2_score(Y_train, Y_train_pred)

train_mse = mean_squared_error(Y_train, Y_train_pred)

test_r2 = r2_score(Y_test, Y_test_pred)

test_mse = mean_squared_error(Y_test, Y_test_pred)

# Print R² and MSE

print(“Train R^2 Score:”, train_r2)

print(“Train Mean Squared Error:”, train_mse)

print(“Test R^2 Score:”, test_r2)

print(“Test Mean Squared Error:”, test_mse)

# Feature importance

feature_importances = rfr.feature_importances_

features = pd.DataFrame({

  ‘Feature’: X.columns,

  ‘Importance’: feature_importances

}).sort_values(by =’Importance’, ascending = False)

print(“Feature Importances: ∖n”, features)