Business Strategies and Financial Distress Risk in Seasoned Listed Companies: Extending Earnings Quality Role

Evaluating a company’s financial standing usually involves positioning it within two fundamentally contrasting states: financial health and FD. The boundary between them appears to lie in financial equilibrium, in which cash flows between revenues and costs, as well as inflows and outflows, are balanced. The two most critical areas for managerial attention are profitability and solvency, although equilibrium in earnings does not necessarily coincide with equilibrium in cash flows.

A company can be considered financially healthy as long as it maintains equilibrium within the financial system, securing the confidence of capital providers during periods of greatest need (Dainelli, Bet & Fabrizi, 2024). Temporary disturbances caused by managerial decisions do not necessarily pose a threat, whereas prolonged imbalance may trigger difficult-to-reverse changes, potentially resulting in failure, insolvency, bankruptcy or default (Altman, 1993). Some authors argue that FD should not be equated with an absolute state, such as bankruptcy (Sun et al., 2016), but rather understood as a path from its onset to the eventual final event (Balcaen & Ooghe, 2006).

The relationship between FD and financial health is not linear. At low levels of distress, even small increases can trigger disproportionately large declines in a company’s fiscal stability, whereas at high levels of distress, further increases generate only marginal additional effects. Thus, the sooner management detects early warning signs of financial vulnerability, the greater its ability to prepare for and mitigate the severity of a potential crisis, and even maintain the company’s financial health. The selection of remedial measures depends not only on the sources and severity of financial vulnerability but also on the stage of market and organisational development. An additional challenge lies in the risk of failing to recognise early symptoms of FD (Watkins & Bazerman, 2004).

Prior studies on the relationship between corporate business strategies and the risk of FD are largely grounded in Porter’s generic strategies (1985) (Hambrick, 1983; Agustia et al., 2020; Thu, 2023). These studies argue that the problem of choosing and successfully implementing a particular business strategy depends on the type of strategic advantage identified and the scope of competition. A firm may either benefit from a cost leadership position, resulting from economies of scale or technological excellence, or pursue a differentiation strategy by offering superior quality or uniqueness. This paper considers two strategies: low-cost and differentiation. The low-cost strategy reduces costs and improves efficiency, proxied by asset turnover of operations (ATO) (Hambrick, 1983; Wu, Gao & Gu, 2015), which reflects how efficiently a firm uses its resources to generate sales, making it a natural proxy for cost leadership. The differentiation strategy seeks competitive advantage through unique products or services, measured by PM (Agustia et al., 2020; Thu, 2023), capturing the premium achieved through product differentiation and value creation, serving as a proxy for a differentiation strategy.

The review of previous studies confirms that the choice of business strategy has a significant impact on FD risk in public companies (Table 1). Firms pursuing a differentiation strategy appear less exposed to distress, as higher PMs, greater innovation capacity and lower reliance on external financing enhance their resilience. In contrast, companies following a cost leadership strategy may achieve short-term efficiency gains, but over time, they are often more vulnerable to FD, particularly when external financing requirements are high or financial flexibility (FF) is limited. Similarly, asymmetric cost behaviour may affect FD risk. Firms that adjust costs downward slowly during sales declines can be more vulnerable, particularly if managerial decisions are overly optimistic or influenced by non-monetary factors such as morale or reputation (Hassanein & Younis, 2020; Hartlieb & Loy, 2022; Ahmed, Iqbal & Atif, 2025). However, cost stickiness may also help firms stabilise operations depending on expected sales trends (Costa & Habib, 2023; Lefebvre, 2025). Overall, we observe that the relationship between business strategy and FD risk is mixed, highlighting the need for further research in this area. In Table 1, we focus on key findings rather than the direction of the relationships, as our study does not aim to test causality or directional effects.

Based on the above observations, we formulated the following hypothesis, which will be tested with respect to companies listed on the WSE:

H1. Higher operational efficiency, measured by the asset turnover of operations and profit margin, significantly reduces the likelihood of financial distress in established issuers.

H2. Greater cost flexibility, reflected in lower stickiness of the cost of goods sold and discretionary expenses, significantly mitigates the risk of financial distress in seasoned listed companies.

The significance of earnings stems from their multiple functions within a company, including financial, motivational, safeguarding and goal-oriented functions – the latter indirectly supporting the company’s primary objective of value maximisation. Despite their importance, the concept of EQ remains complex and lacks a single, unified definition. Dechow et al. (2010) highlight three key aspects: characteristics of earnings over time, investor and market reactions, and external stakeholders’ assessment. According to Francis, Olsson & Schipper (2006), inherent sources of EQ derive from the business model, operational environment and firm-specific characteristics, while reporting quality depends on accounting standards, supervisory systems, information technology (IT) infrastructure and managerial choices in preparing financial statements.

The literature generally finds that lower EQ is associated with higher FD, although results are not fully conclusive (Table 2). Most studies focus on EM and FD risk, while fewer examine income smoothing – deliberate efforts to reduce fluctuations in reported earnings – and conditional conservatism, which entails faster recognition of losses than gains. Table 2 summarises the key findings from previous studies, highlighting evidence on EM, conditional conservatism and income smoothing, without implying causality.

Based on the literature summarised above, we propose the following hypotheses for companies listed on the WSE:

H3. The extent of accrual-based earnings management is positively associated with the risk of financial distress in continuously listed companies.

H4. Earnings quality, measured by the asymmetry in recognising good and bad news, is significantly associated with the risk of financial distress in long-standing issuers.

H5. An increasing extent of income smoothing is positively associated with the risk of financial distress in seasoned companies.

The study covers 217 long-standing issuers on the Main Market of the WSE, all of which maintained uninterrupted trading of their shares during 2014–2023. A total of 2,170 observations were analysed. The sample is diversified by industry, following the WSE’s sectoral classification and includes 30 companies from the financial sector (excluding banks and insurers), 11 from fuel and energy, 25 from chemicals and raw materials, 72 from industrial production and construction, 32 from consumer goods, 19 from trade and services, 10 from healthcare and 18 from technologies. The financial sector was included to capture strategic reporting and cost behaviour in capital-intensive firms, while banks and insurers were excluded due to their distinct regulatory frameworks and, above all, their different financial statement structure.

The choice of research timeframe involves distinct macroeconomic regimes, including pre-pandemic growth, the COVID-19 crisis and the subsequent phase of heightened uncertainty. Specifically, pre-pandemic growth covers 2014–2019, the COVID-19 crisis spans 2020–2021 and the post-pandemic period extends from 2022 to 2023. In this way, we capture factors that could identify asymmetric cost behaviour and strategic earnings reporting under financial pressure. The period 2014–2023 is particularly suitable for examining FD risk in seasoned listed firms, as it encompasses both normal and stressed economic conditions, allowing for observation of patterns in cost management, EM and strategic decision-making across different macroeconomic contexts. Beyond the theoretical arguments, we also considered methodological requirements. The calculation of several EQ proxies requires a minimum of 10 consecutive firm-year observations, ensuring statistical reliability and comparability across firms (Dechow, Sloan & Sweeney, 1995).

When designing our research sample, we were guided by the conviction that business strategies and EQ are shaped at the level of individual firms rather than being simple replications of sector-wide patterns. This assumption is consistent with several theoretical perspectives. According to agency theory (Jensen & Meckling, 1976), EM and reporting practices are the result of firm-specific incentive structures and conflicts of interest between managers and shareholders. Upper echelons theory (Hambrick & Mason, 1984) stresses that strategic choices mirror the values, experiences and cognitive frames of top executives, which leads to heterogeneity in business strategies even within the same industry. Similarly, the resource-based view (Barney, 1991) points to the uniqueness of firm-level resources and capabilities – including internal reporting systems, IT infrastructure and governance mechanisms – which directly affect the quality of reported earnings. At the same time, institutional theory (DiMaggio & Powell, 1983) acknowledges that although firms in one sector may experience isomorphic pressures, substantial variation remains at the individual level.

To assess the level of financial bankruptcy risk, we employed the Altman Z″-score model adapted for emerging markets (Altman & Hotchkiss, 2006). This model was chosen for two main reasons: first, it allows for meaningful international comparisons, and second, it has demonstrated high predictive accuracy in identifying FD among publicly listed companies in the Polish capital market (Lisicki & Dąbrowska, 2024). The resulting Z″-score values were subsequently translated into financial risk categories as follows: Z″-score ≥5.85 – safe zone; 4.15 ≤ Z″-score < 5.85 – uncertain zone; Z″-score < 4.15 – distress zone. The Altman Z″-score is calculated as follows (1): (1) Z″−scoret=3.25+6.25⋅X1t ++ 3.26⋅X2t+6.72⋅X3t+1.05⋅X4t \matrix{ {Z'' - {\rm{scor}}{{\rm{e}}_t} = 3.25 + 6.25 \cdot X{1_t}\ + } \hfill \cr { + \;3.26 \cdot X{2_t} + 6.72 \cdot X{3_t} + 1.05 \cdot X{4_t}} \hfill \cr } where Z-score_t – the Altman Z²-score in year t; X1_t – the ratio of current assets minus current liabilities to total assets in year t; X2_t – the ratio of retained earnings to total assets in year t; X3_t – the ratio of operating incomes to total assets in year t; X4_t – the ratio of the book value of equity to the book value of debt in year t.

As a second tool for estimating the risk of FD, we employed the ‘G’ model developed by INE PAN. This model was constructed based on the data from Polish companies and was specifically tailored to the characteristics of the Polish economy. Therefore, we used it as an instrument for robustness checks. Its interpretation is binary: a function value ≥0 indicates a sound financial condition, whereas a value < 0 signals a prediction of bankruptcy (Jagiełło, 2013). The model is expressed by the following formula (2): (2) ZINE PAN t=−1.498+9.489⋅Z1t ++ 3.556⋅Z2t+2.903⋅Z3t+0.452⋅Z4t \matrix{ {{Z_{INE\;PAN\;t}} = - 1.498 + 9.489 \cdot Z{1_t}\ + } \hfill \cr { + \;3.556 \cdot Z{2_t} + 2.903 \cdot Z{3_t} + 0.452 \cdot Z{4_t}} \hfill \cr } where Z INE PAN t – the financial distress score (Z-score) calculated according to the G INE PAN model in year t; Z1t – operating profit to total assets ratio in year t; Z2t – equity-to-total-assets ratio in year t; Z3t – the ratio of net income plus depreciation to total liabilities in year t; Z4t – current assets to current liabilities ratio in year t.

In this study, the Porterian business strategies are reflected through the low-cost and the differentiation strategies. The first one is implemented to achieve operational efficiency and cost minimisation, while maintaining competitive performance. Prior research (e.g., Hambrick, 1983; David et al., 2002; Thu, 2023) describes companies’ tendency towards a low-cost strategy using the ATO indicator. We interpret a higher output-to-input ratio as evidence that the company uses its resources more efficiently to strengthen its cost leadership position (Agustia et al., 2020). Similar to Wu et al. (2015), we calculate ATO using the following Eq. (3): (3) ATOt=REVtOAt {\rm{AT}}{{\rm{O}}_t} = {{{\rm{RE}}{{\rm{V}}_t}} \over {O{A_t}}} where ATO_t – assets turnover of operations in year t; REV_t – revenues generated from the sale of goods, products, and services related to the company’s core operations in year t; OA_t – operating assets in year t (computed as total assets minus cash and short-term investments).

In turn, the differentiation strategy requires companies to create a competitive advantage within the sector by offering unique and distinctive features in their products and services. The differentiation strategy is proxied by the PM ratio (Wu et al., 2015), as it captures a company’s focus on achieving superior profitability through distinctive products and an emphasis on innovative activities. In practical terms, a higher PM suggests that the company prioritises differentiation, maintains relatively high profit levels and commits comparatively more resources to R&D than other firms in the industry. The PM indicator is specified as Eq. (4): (4) PMt=OPt+R&DtREVt P{M_t} = {{O{P_t} + R\& {D_t}} \over {{\rm{RE}}{{\rm{V}}_t}}} where PM_t – profit margins in year t; OP_t – operating profit in year t; R&D_t – research and development expenditures in year t; other designations – as above.

Finally, to measure cost stickiness (STICKY_COGS; STICKY_SG&A), we used the model by Anderson et al. (2003), which examines the linear relationship between changes in the natural logarithm of costs and changes in the natural logarithm of sales revenue. According to the authors, the coefficient shows how costs respond when sales decrease. Therefore, when a company follows a cost stickiness (or anti-stickiness) strategy, a negative (or positive) α₂ parameter is expected. Simultaneously, a higher α₂ reflects a lower degree of cost stickiness in the company. This coefficient also provides insight into managerial responses to sales fluctuations, as it captures the degree to which managers adjust costs asymmetrically in reaction to declining revenues, reflecting their approach to operational risk and strategic decision-making (Anderson et al., 2003). We analysed cost stickiness separately for both selling, general and administrative costs (SG&A) and cost of goods sold (COGS). Eq. (5) presents the general form of the model: (5) ΔlnSG&AtΔlnCOGSt=α0+α1ΔlnREVt+α2DECt×ΔlnREVt+εt \left[ {\matrix{ {\Delta lnSG\& {A_t}} \hfill \cr {\Delta lnCOG{S_t}} \hfill \cr } } \right] = {\alpha _0} + {\alpha _1}\Delta lnRE{V_t} + {\alpha _2}DE{C_t} \times \Delta lnRE{V_t} + {\varepsilon _t} where SG&At – selling, general and administrative costs in year t; COGS_t – cost of goods sold in year t; DEC_t – dummy variable equal to 1 when ΔlnREV_t < 0, and 0 otherwise; α₁, α₂, α₃ – a firm-specific parameter; ε_t – a random error; other designations – as above.

As previously indicated, EQ was measured using three variables: income smoothing (SMOOTH), conditional conservatism (CONS) and DACC. Income smoothing is calculated as the ratio of the standard deviation of net income to the standard deviation of operating cash flows (6). A higher value of this measure indicates a lower degree of earnings smoothing, meaning that net results more closely reflect the actual fluctuations in cash flows (Leuz, Nanda & Wysocki, 2003). From an economic perspective, we interpret higher SMOOTH values as a signal that the company’s reported results are more aligned with its real operational performance, whereas lower values suggest greater smoothing of reported earnings, potentially reflecting managerial interventions. (6) SMOOTH=σEATtσOCFt {\rm{SMOOTH}} = {{\sigma \left( {EA{T_t}} \right)} \over {\sigma \left( {OC{F_t}} \right)}} where SMOOTH_t – income smoothing indicator; EAT_t – earnings after taxes in year t; OCF_t operating cash flows in year t.

To assess the scope of conditional conservatism (CONS), we applied the regression model by Ball and Shivakumar (2005), where conservatism is captured by the parameter α₄ (7). A higher α₄ indicates a stronger tendency of the company to recognise negative earnings news faster than the positive news. This approach relies on the asymmetry between total accruals (TACC) and operating cash flows, based on the assumption that economic losses are recognised more promptly than economic gains. The formula for the described model is as follows: (7) TACCtTAt−1=α11TAt−1+α2NOCFtTAt−1++ α3OCFtTAt−1+α4NOCFtTAt−1×OCFtTAt−1+εt \matrix{ {{{TAC{C_t}} \over {T{A_{t - 1}}}} = {\alpha _1}\left( {{1 \over {T{A_{t - 1}}}}} \right) + {\alpha _2}\left( {{{NOC{F_t}} \over {T{A_{t - 1}}}}} \right) + } \cr { + \;{\alpha _3}\left( {{{OC{F_t}} \over {T{A_{t - 1}}}}} \right) + {\alpha _4}\left( {{{NOC{F_t}} \over {T{A_{t - 1}}}}} \right) \times \left( {{{OC{F_t}} \over {T{A_{t - 1}}}}} \right) + {\varepsilon _t}} \cr } where TACCt – total accruals in period t (determined as the difference between earnings after taxes and operating cash flows); TAt – total assets in year t; NOCF_t – dummy variable (equal to 1 when operating cash flows in year t were negative and 0 in other cases); other designations – as above.

We capture accrual-based EM using DACC, estimated according to the classical Jones model (Jones, 1991). In this framework, abnormal accruals represent the portion of TACC not explained by the model. DACC are computed by subtracting non-discretionary accruals (NDACC) from TACC (8). The difference between the theoretical (expected) and observed TACC is assumed to reflect the firm’s engagement in accounting discretion. From a methodological perspective, the extraction of individual accrual subcategories was carried out using the cash-based approach, in which TACC are defined as the difference between earnings after taxes and operating cash flows (9). (8) DACCt=TACCt−NDAC^Ct {DACC_t} = TAC{C_t} - N\widehat {DAC}{C_t} where DACC_t – discretionary accruals in year t; TACC_t – total accruals in year t; NDAC^Ct N\widehat {DAC}{C_t} – normal (expected) accruals, i.e., the portion of total accruals predicted by the regression model in year t. (9) TACCt=EATt−OCFt {TACC_t} = EA{T_t} - OC{F_t} The Jones model is specified in Eq. (10): (10) TACCtTAt−1=α11TAt−1+α2ΔREVtTAt−1+α3PPEtTAt−1+εt {{TAC{C_t}} \over {T{A_{t - 1}}}} = {\alpha _1}\left( {{1 \over {T{A_{t - 1}}}}} \right) + {\alpha _2}\left( {{{\Delta RE{V_t}} \over {T{A_{t - 1}}}}} \right) + {\alpha _3}\left( {{{PP{E_t}} \over {T{A_{t - 1}}}}} \right) + {\varepsilon _t} where PPE_t – gross property, plant and equipment in year t; other designations – as above.

We employ the original Jones model to estimate DACC because it remains widely used in empirical research and allows for consistent comparison across firms and over time. Although more recent models exist, prior studies suggest that for large samples, the choice of model usually does not qualitatively change the findings (Dechow et al., 1995; Kothari, Leone & Wasley, 2005). In our research, DACC are analysed in both signed and absolute form. Signed accruals show the direction of EM (i.e., income-increasing vs. income-decreasing), while absolute |DACC| show the overall size of managerial discretion in financial reporting, regardless of direction. Looking at both measures gives a fuller picture of how managers use accounting choices, making it easier to see links between EM and FD.

In our study, we also included several control variables. First, asset structure (tangibility [TANG]) is measured as the proportion of property, plant and equipment in total assets in year t. Previous studies suggest that asset TANG affects a firm’s risk of FD because tangible assets improve collateral value and debt capacity, reducing default risk and easing access to external finance (Camisón, Clemente & Camisón-Haba, 2022). Leverage (LEV) is defined as the ratio of the firm’s interest-bearing debt to total assets in year t. The literature indicates that LEV increases fixed debt obligations and limits FF, which directly raises the probability of FD, especially during periods of economic instability (Uğur, Solomon & Zeynalov, 2022). Short-term liabilities (STL) represents the proportion of short-term liabilities in total assets in year t. A higher share of short-term debt makes firms more exposed to refinancing and liquidity problems, which increases the risk of FD when credit conditions become tighter (Draief & Chouaya, 2022). Additionally, we account for the impact of the COVID-19 pandemic in Poland by including a dummy variable (COVID), which takes the value of 1 for observations corresponding to the years 2020–2021 and 0 otherwise, reflecting the period of the greatest pandemic-related effects (Hsu & Jan, 2023). These control variables are relevant as they capture key aspects of a firm’s financial structure and external shocks, which may influence both EQ and the risk of FD.

In the initial stage of our research, we examined the distribution of business strategy and EQ proxies across different Z-Score zones. As expected, we found that most of the tested variables changed significantly according to the estimated FD zone. Based on the data presented in Table 3, we notice several findings worthy of comment. We observe that companies under FD show greater variability and less predictable cost behaviour, which we assume results from difficulties in maintaining efficient cost structures under pressure. As financial standing improves, the asymmetry between costs and revenues decreases. Considering the ATO and PM indicators, we find that companies’ economic efficiency rises with financial stability, whereas distress brings volatility and weaker performance in both operational efficiency and profitability. Finally, EQ measures reveal that distressed firms are more prone to income smoothing and accrual-based management.

However, the results of the Kruskal–Wallis test brought slightly different insights (Table 4). The null hypothesis of the Kruskal–Wallis test assumes that the distribution of the analysed proxy is identical across all Z-score zones. We found that the differences across financial zones are statistically significant for ATO and PM, indicating that both asset efficiency and profitability vary meaningfully with financial condition. In the case of STICKY_COGS, the result is marginally significant at the 0.1 level, which allows us to cautiously assume that cost stickiness related to COGS may also differ across zones. In contrast, the STICKY_SG&A measure did not show significant differences, suggesting that this type of cost rigidity may not directly depend on the company’s financial standing. When turning our attention to EQ, the outcomes are consistent across all proxies. We observed significant variations in SMOOTH, CONS, DACC and |DACC| which indicates that reporting behaviour changes systematically with the company’s financial health. These findings suggest that as financial stability improves, earnings become less managed and more reflective of underlying performance.

Because the results of the Kruskal–Wallis test required multiple comparisons, we conducted Dunn’s tests, including the Bonferroni-corrected version. The statistical significance of the pairwise comparisons is presented in Table 5. We observed the largest number of significant differences for the ATO and PM indicators, with nearly all pairwise comparisons between Z-score zones remaining statistically significant after Bonferroni correction. Regarding the income smoothing proxy (SMOOTH), all three pairwise comparisons were also found to be statistically significant.

Although the descriptive statistics presented in Table 1 suggested that the values of the STICKY_SG&A proxy varied across the compared subpopulations, the Kruskal–Wallis test did not confirm these differences as statistically significant, even at the 0.10 significance level. We assume that this apparent inconsistency arises because descriptive measures (such as means or medians) may indicate numerical variation that is not strong enough to reach statistical significance once the within-group variability is considered. Based on these results, we decided to exclude the STICKY_SG&A variable from further analyses performed using the PLUM regression model.

The Kruskal–Wallis test applied in the previous section allowed us to verify whether the distributions of selected proxies differed across the predefined Altman’s Z-score zones. However, it did not capture the direction or intensity of these relationships. Therefore, the next stage of our scientific exploration leads to examine how specific variables influence the probability of a firm being in a higher or lower distress category. For this purpose, we applied the PLUM regression model, which treats FD risk as an ordered outcome. Because the dependent variable is ordinal and observations are unevenly distributed across financial condition zones (21% in the distress zone, 20% in the uncertain zone and 59% in the safe zone), we rely on the Cauchit link function. This specification is well-suited to asymmetric distributions and enables a clear analysis of movements between ordered FD levels under the proportional odds framework.

We ran two models: one including the variable DACC, which captures the magnitude and direction of accrual-based EM in the investigated companies, and another excluding it. This approach is motivated by the fact that EQ proxies, such as conditional conservatism and, in particular, income smoothing, may be influenced by earnings-altering practices. Importantly, we do not expect CONS and SMOOTH to be inherently correlated; however, including DACC allows us to control for the potential effects of accrual-based EM on these measures, ensuring the robustness of our results.

In the first step, we ran the PLUM regression model without including the accrual-based EM variable (DACC) (Table 6). Among the business-related variables, ATO and PM were positively associated with financial health, indicating that firms with higher operational efficiency and greater profitability are more likely to be in safer financial zones. In contrast, we found that STICKY_COGS variable was negatively associated with financial health, suggesting that firms with more rigid cost structures are more prone to FD. Regarding EQ, SMOOTH showed a negative association with financial stability, implying that firms engaging in stronger smoothing practices tend to be in higher distress categories. Conversely, CONS exhibited a small protective effect, suggesting that more conservative accounting practices slightly reduce the likelihood of FD. We verified that the assumptions of the PLUM regression model were adequately met, including the proportional odds assumption, ensuring the validity of the estimated coefficients. Regarding multicollinearity, the VIF (Variance Inflation Factor) values for all predictors were well below commonly accepted thresholds, confirming that collinearity does not pose a concern in this model. Overall, the model demonstrates a satisfactory fit to the empirical data.

Following this, we ran the PLUM regression model including the accrual-based EM variable (DACC) to control for potential effects of accrual manipulations on FD risk (Table 7). Compared with the previous model, the inclusion of DACC does not substantially alter the direction or significance of the main business-related variables (ATO, PM and STICKY_COGS) or EQ proxies (SMOOTH and CONS), suggesting that our initial findings are robust. The DACC variable itself exhibits a negative but statistically insignificant association with financial health, indicating that accrual-based EM does not have a strong direct effect on the likelihood of a firm being in a higher distress category within investigated sample.

Considering the tested control variables, we found that asset TANG strongly reduces the probability of FD in our sample, which confirms previous evidence by Camisón et al. (2022). What is interesting is that a higher share of short-term liabilities (STL) reduces distress risk, and LEV also reduces the likelihood of being in a higher distress category, which is opposite to what the literature usually reports (Draief & Chouaya, 2022; Uğur et al., 2022). The COVID dummy is not statistically significant, suggesting no clear direct impact of the pandemic in our sample. These results are consistent across models with and without DACC, confirming that accrual-based EM does not drive the observed relationships.

We additionally estimated the PLUM model using the absolute value of |DACC| as an alternative measure of EM intensity. However, the proportional odds (parallel lines) assumption was violated, as indicated by a significant test of parallel lines (p < 0.05). Therefore, we decided to excluded this model version from the main analysis.

To check the robustness of our results, we apply a two-step procedure. First, we estimate the ordinal regression (PLUM) model using the standard Logit link function. This procedure allows us to verify whether the previously obtained results depend on the choice of the link function. Second, we repeat the whole empirical analysis using the INE PAN ‘G’ model as an alternative measure of FD. In this step, we reclassify firm-year observations according to the ‘G’ model, while keeping the same set of explanatory variables and the same ordinal regression framework.

As expected, the PLUM regression results obtained using the Logit link function do not alter the overall pattern of the findings (Table 8). The key explanatory variables – ATO, PM, STICKY_COGS, SMOOTH and CONS – remain statistically significant determinants of FD risk in seasoned firms from the WSE. In turn, DACC do not have a statistically significant effect on the financial health of these firms. In contrast to some earlier studies, the coefficient on the CONS variable is positive, indicating that higher accounting conservatism is associated with a higher probability of being classified in safer financial condition categories rather than in the distress zone.

Next, we performed several robustness tests based on the G INE PAN model. The aim was to check if our results were stable. We tested different model specifications and variable inclusions. The main relationships between business-related variables, EQ proxies and FD risk remained consistent.

First, in Table 9, we present descriptive statistics for the business strategy and EQ variables calculated for the two zones defined in the ‘G’ INE PAN model, namely the distress and safe zones. Compared with the statistics reported earlier in Table 1, we found the results to be generally consistent. We observed that most business-related variables (STICKY_COGS, STICKY_SG&A, ATO and PM) and EQ proxies (SMOOTH, DACC and |DACC|) show similar trends across the distress and safe zones. The only notable exception is CONS, where we found differences in values and sign, suggesting higher variability and possible sensitivity to the sample or model specification. Overall, we conclude that the patterns observed in the data are largely robust across the two analyses.

Subsequently, we performed the non-parametric Mann–Whitney U test to examine distributional differences of the tested business strategy and EQ proxies across the considered subpopulations (Table 10). Its null hypothesis states that the distributions of the examined variable are the same in both subpopulations under consideration. Relative to the previous analyses (Table 4), we found that the results indicate acceptance of the null hypothesis for both STICKY_COGS and STICKY_SG&A. It is worth noting that for STICKY_COGS, we had previously rejected the null hypothesis at the 0.1 significance level in the Kruskal–Wallis test. This difference may result from the distinct nature of the two tests, the reduction to two subpopulations in the current analysis or the sensitivity of the results to the significance level. In the remaining cases, we did not observe statistically significant differences compared with our earlier observations. Overall, we confirmed that the main patterns identified in the data are largely robust.

Finally, in the last stage of our research procedure, we re-estimated the PLUM regression model as part of the robustness verification (Table 11). The sample included 326 observations classified in the distress zone and 1,844 observations in the safe zone. When comparing these results with those obtained for the Altman Z-score model (Table 7), we observed that the overall pattern of relationships remained consistent, although some differences in the strength and significance of individual coefficients appeared. The signs of the key business-related variables (ATO, PM and STICKY_COGS) as well as EQ proxies (SMOOTH) stayed in line with our earlier results, which confirms the robustness of the main directions of the relationships. Importantly, the CONS variable changed its sign from negative to positive, which we interpret because of lower model fit or differences in sample composition rather than a genuine reversal of its economic meaning. Moreover, the COVID variable, which was insignificant in the previous model, became statistically significant at the 0.05 level. It can indicate that external shocks might play a more pronounced role under the robustness framework. For the sake of clarity, we emphasise that the test of parallel lines was not performed in this model because the dependent variable includes only two zones. In such a case, the model works as a standard binary logistic regression, and the proportional odds assumption is automatically met.

In the robustness checks, we found that the control variables mostly behave as in the main analysis. Although asset TANG turned out to be non-significant, both short-term liabilities (STL) and LEV tend to reduce the likelihood of being in a higher distress category. The COVID dummy became statistically significant only in the Cauchit model, suggesting that its impact may depend on the model specification.