Various theories explain earnings management behaviour during crises, including the agency, prospect and signalling theories. Agency theory highlights the separation of ownership and control within a firm. As a result, management possesses complete information about a firm’s actual economic performance, while shareholders and other outside stakeholders must rely on the information released by management. Jensen and Meckling (1976) argue that if both management (the agent) and the shareholders (the principal) are utility maximisers, then management will exploit the information asymmetries for their personal benefit. During crises, uncertainty and survival pressures increase management incentives to engage in earnings management (Liu & Sun 2022; Swai & Derefa 2024). This may occur either by inflating current-year performance to hide poor results or, when losses are too severe or investors are more tolerant of poor performance, by deliberately inflating current-year losses through a ‘big bath’, thereby enabling stronger reported performance in future years (Guo et al. 2025; Lassoued & Khanchel 2021; Liu & Sun 2022; Lizińska & Czapiewski 2023). Conversely, Ali et al. (2022) and Chia, Lapsley and Lee (2007) suggest that earnings management may decrease during crises. This can be attributed to increased regulatory scrutiny and closer monitoring by auditors, which constrain managerial opportunism. In addition, pressure on managers to manipulate results may decrease as investors may be more willing to accept poor performance during crises. Finally, the scale of losses incurred may be so severe that they cannot be reasonably concealed through any earnings management technique (Ali et al. 2022; Chia et al. 2007; Türegün 2020).

Prospect theory provides an alternative explanation of management behaviour during crises. This theory suggests that managers react asymmetrically to gains and losses. Managers are risk-averse when facing potential gains, but risk-seeking when facing potential losses (Kahneman & Tversky 1979). During crises such as COVID-19, the heightened threat of loss may incentivise management to engage in riskier accounting practices, including earnings management, to avoid reporting losses and to maintain investor confidence (Adejumo & Ogburie 2025). Alternatively, if management anticipates unavoidable losses, they may engage in income-decreasing earnings management (i.e. taking a ‘big bath’) to push earnings into future periods, as investors view losses beyond the cost of the crisis as inconsequential (Barton et al. 2025).

Finally, signalling theory focuses on reducing information asymmetries between management and stakeholders. Management has access to private information and must communicate it to external stakeholders who only have access to public information (Connelly et al. 2011). Signalling theory views earnings management as one mechanism through which management communicates their private information about a firm’s prospects to stakeholders (Connelly et al. 2011). In times of crises, such as the COVID-19 pandemic, information asymmetry increases, making stakeholders more sensitive to signals. Managers may, therefore, engage in earnings management to signal financial stability and reassure investors, thereby maintaining confidence in the firm (Hussein et al. 2025).

Empirical evidence supports the theoretical view that financial and economic crises influence management’s earnings management practices (Cimini 2015; Grimaldi 2019; Li et al. 2020; Saleh & Ahmed 2005; Türegün 2020). During such periods, earnings management practices may change in two material ways. Firstly, the magnitude of earnings management may vary, with managers either intensifying or reducing their use of earnings management (Aljughaiman et al. 2023). Secondly, the direction may change, as managers move between income-increasing and income-decreasing earnings management depending on their objectives (Aljughaiman et al. 2023). Despite extensive research, however, findings on earnings management during the COVID-19 pandemic remain inconclusive. In line with the objectives of this study, the review now focuses on the literature on AEM.

This study’s population comprised all firms listed on the JSE from 2016 to 2022, totalling 333 firms. Financial data and the variables were retrieved from Thomson Reuters Eikon. Firms with dual listings were excluded, as the regulatory environment in which they operate influences financial reporting quality and, consequently, the extent of earnings management (Ali et al. 2022). Financial services firms and real estate investment trusts were also excluded because of sector-specific financial reporting and regulatory characteristics (Chakroun, Ben Amar & Ben Amar 2022). Exchange Traded Funds (ETFs), certificates, portfolios, and retirement funds were also excluded because they are not operating firms. The final sample consisted of an unbalanced panel of 159 firms representing 1063 firm-year observations. The sample selection process is summarised in Table 1.

The JSE uses the Industry Classification Benchmark (ICB) method to categorise firms. The sector sample distribution is detailed in Table 2.

Observations from 2016 to 2019 were classified as the pre-crisis period, while 2020 to 2022 constituted the crisis period. The crisis period was informed by lockdown protocols (levels 1–5) that spanned from March 2020 until April 2022.

As earnings management cannot be explicitly observed, discretionary accruals were used as a proxy. Following studies by Kumar and Vij (2017), Sellami and Slimi (2016) and Türegün (2020), the cross-sectional Kothari et al. (2005) model was used to estimate discretionary accruals. By including return on assets as a performance term, this model controls for abnormal performance observed during the COVID-19 crisis. The model is as Equation 1: TACCit=α1(1/Assetsit-1)+α2(ΔSalesit−ΔTRit)/Assetsit-1     +α3PPEit/Assetsit-1+α4(ROAit)+εt[Eqn 1]

Where:

Equation (1) was estimated for each ICB sector, and the coefficients α₁, α₂, α₃ and α₄ were calculated by running panel ordinary least squares (OLS) regression (Türegün 2020). The residual resulting from Equation (1), ε_t, measured discretionary accruals (as a percentage of total assets). Further, cluster-robust standard errors were applied to address potential heteroskedasticity or serial correlation (Hoechle 2007).

Both the absolute and signed values of discretionary accruals were used. The absolute values indicate the magnitude of AEM, with a high absolute value indicating intensified AEM leading to a lower quality of earnings, and vice versa. In contrast, the signed values indicate the direction of AEM (i.e. income-increasing or -decreasing) to achieve specific objectives (Lizińska & Czapiewski 2023).

A two-sample t-test was performed to assess whether the change in discretionary accruals pre- and post-crisis is significantly different.

To assess the robustness of the t-test result, the significance of the relationship between AEM and the COVID-19 pandemic was assessed. The following fixed effects panel regression with panel corrected standard errors (Equation 2) was used to address heteroskedasticity, serial correlation, and cross-sectional correlation (Hoechle 2007): AEMit=β0+β1Pandemicit+β2ROAit+β2LnTAit-1        +αi+μit[Eqn 2]

Where AEM_it represents the discretionary accruals for firm i in year t. Pandemic_it is a dummy variable indicating whether the observation fell within the pre-crisis (0) or crisis (1) periods. Return on assets, a proxy for profitability, and the natural log of total assets, representing company size, were used as control variables. Finally, α_i is the unobserved individual effect while µ_it is the error term.

This article followed all ethical standards for research without direct contact with human or animal subjects.

Table 3 presents the mean discretionary accruals in total and by year, as well as the means for the pre-crisis and crisis periods, for the JSE as a whole and by sector. Panel A presents the absolute discretionary accruals, while Panel B presents the signed discretionary accruals.

Considering the overall sample, the magnitude of AEM (measured by absolute discretionary accruals) decreased between 2016 and 2017. This was accompanied by a shift from income-increasing AEM in 2016 to income-decreasing AEM in 2017. From 2017 to 2019, AEM levels remained stable. However, in 2020 (the start of the pandemic), AEM activity increased, as indicated by higher absolute discretionary accruals (mean = 0.0921) and stronger income-decreasing practices (mean = –0.0636), suggesting a deterioration in earnings quality. This trend persisted into 2021 (absolute mean = 0.0810, mean = –0.0543), before earnings quality improved in 2022, as signified by the decrease in both the absolute and signed discretionary accruals (absolute mean = 0.0718, mean = –0.0229). By 2022, AEM levels had returned to pre-crisis levels.

For the overall sample, mean absolute discretionary accruals increased from 0.0669 in the pre-crisis period to 0.0820 during the COVID-19 crisis. The independent-samples t-test confirmed that the increase of 0.0151 was statistically significant (p < 0.05). Similarly, mean discretionary accruals decreased from –0.0134 in the pre-crisis period to –0.0476 during the crisis, with the difference of –0.0341 also being statistically significant (p < 0.01). These results provide preliminary support for hypotheses 1 and 2. This highlights possible ethical and governance concerns, which may indicate a lack of investor protection during a period of high economic turbulence.

When examining absolute discretionary accruals by sector, Table 3 panel A indicates that the basic materials and telecommunications sectors had the highest AEM levels. Except for the energy and healthcare sectors, all sectors experienced an increase in absolute discretionary accruals during the pandemic, indicating a decline in earnings quality. Post the onset of the pandemic, basic materials continued to display the highest level of AEM (mean = 0.1462) and showed the largest increase in AEM (change = 0.0353). This was followed by the consumer discretionary sector (change = 0.0258). However, the increase was only statistically significant for the consumer discretionary sector (p < 0.01).

Table 3 panel B reveals that, on average, apart from the basic material and industrial sectors, South African firms were engaged in income-decreasing AEM pre-crisis. Post the onset of the pandemic, the basic materials and industrials sectors changed strategies from income-increasing to income-decreasing AEM. The change in discretionary accruals was statistically significant only for the basic materials (p < 0.05), consumer discretionary (p < 0.01) and technology (p < 0.10) sectors. These findings provide preliminary support for hypothesis 3. The sector analysis also shows that a few sectors influenced the overall sample, and the results should not necessarily be extrapolated to the whole market. This outlines that behavioural change in AEM practices is not driven by economic turbulence in isolation, but rather by the interaction between economic turbulence and sector-specific factors that shape the incentives and opportunities faced by management.

Table 4 presents the regression results from formula (2), with panel A reporting absolute discretionary accruals as the dependent variable, and panel B reporting signed discretionary accruals. The results indicate that for the full sample (p < 0.01), as well as for the consumer discretionary (p < 0.05), basic materials (p < 0.05), and technology sectors (p < 0.1), the pandemic was a significant factor in the determination of discretionary accruals. For absolute discretionary accruals, the overall sample (p < 0.05), consumer discretionary (p < 0.05) and industrials (p < 0.05) sectors, the pandemic was found to be a significant determinant.

Considering the JSE firms collectively, similar to the results in Table 3, the regression results show that during the crisis period, firms engaged in more AEM activities, as reflected in the positive, statistically significant coefficient under absolute discretionary accruals (β₁ = 0.0138, p < 0.01). Panel B (signed discretionary accruals) further indicates that this took the form of more income-decreasing AEM (β₁ = –0.0354, p < 0.01). These findings are consistent with Liu and Sun (2022), who documented an increase in absolute discretionary accruals alongside a significant decrease in discretionary accruals, suggesting that US firms increased their use of income-decreasing AEM during the COVID-19 pandemic. Similarly, Chia et al. (2007) also observed income-decreasing AEM during a crisis.

From an agency and prospect theory perspective, these results suggest that managers used the crisis as an opportunity to engage in self-interested behaviour by using ‘big bath’ accounting. By deliberately manipulating earnings downwards, capitalising on investors’ expectations of poor performance, managers could reduce current period earnings expectations and subsequently inflate future performance once the crisis had passed (Liu & Sun 2022; Lizińska & Czapiewski 2023). Thus, while there was no benefit to upward earnings management during the pandemic, firms chose to manage earnings downward, consistent with the ‘big bath’ explanation. Consequently, these findings support hypotheses 1 and 2, which state that the COVID-19 pandemic affected both the magnitude and direction of AEM in South Africa.

Considering the sector results, the consumer discretionary (β₁ = 0.0216, p < 0.05) and industrial (β₁ = 0.0084, p < 0.05) sectors showed a significant increase in AEM magnitude (panel A). Concerning direction, the basic materials (β₁ = –0.1420, p < 0.05), consumer discretionary (β₁ = –0.0245, p < 0.05) and technology (β₁ = –0.0394, p < 0.10) sectors shifted towards more income-decreasing AEM (panel B). In contrast, the consumer staples, energy, healthcare, and telecommunications sectors showed no statistically significant change in their earnings management behaviour. Overall, these results are consistent with the descriptive evidence presented in Table 3, reinforcing the observation that COVID-19’s impact on AEM was not uniform across sectors. Thus, these results highlight the importance of considering how sector-specific factors influence AEM behaviour, which may explain the inconsistent results in prior literature that have considered the individual sectors collectively.

The sector results show that sector-specific exposure to COVID-19 influenced earnings management behaviour, consistent with agency, prospect and signalling theories. Sectors that were more severely affected by the pandemic (such as basic materials, consumer discretionary and industrials) effectively came to a halt as they were deemed non-essential. These sectors experienced the biggest shifts in AEM behaviour, consistent with the findings of Chia et al. (2007) and Yan et al. (2022). From an agency and prospect theory perspective, managers in more severely affected sectors may have employed a ‘big bath’ strategy by deliberately reducing earnings so that they could inflate future performance (Liu & Sun 2022). This strategy would have been facilitated by investor acceptance of poor performance during the pandemic.

In contrast, sectors less severely affected by the pandemic (such as consumer staples, healthcare, energy, and telecommunications) displayed no meaningful change in AEM behaviour. From an agency perspective, firms in these sectors would have faced less crisis-related pressure and, hence, weaker management incentives to manipulate earnings. From a prospect theory perspective, managers in these sectors faced limited operational disruptions and a lower threat of reduced performance. Consequently, these managers would have been less likely to engage in risky accounting practices, such as AEM (Kahneman & Tversky 1979). Finally, applying signalling theory, these managers would have had limited need to manipulate earnings through AEM to signal stability or reassure investors (Connelly et al. 2011).

Interestingly, however, is the finding that the technology sector engaged in greater income-decreasing earnings management during the pandemic (although this was only at the 10% significance level). This was unexpected given the minimal impact of the pandemic on advanced technology sectors (He et al. 2020; Shen et al. 2020). In fact, He et al. (2020) and Mazur, Dang and Vega (2021) noted that the pandemic had a positive impact on the technology sector’s performance in China and the United States, respectively. However, despite arguments that the technology sector should not have been severely affected by the pandemic, these findings are consistent with those of Lizińska and Czapiewski (2023). One possible explanation for this increase in income-decreasing AEM is that firms’ behaviour remained conservative as a result of the uncertainty surrounding the pandemic. From an agency perspective, Lizińska and Czapiewski (2023) argued that technology firms engaged in a ‘big bath’ strategy, using positive earnings to absorb the income-decreasing AEM, thereby allowing better results in future years when the accruals are reversed. Alternatively, from a signalling theory perspective, the firms may have used income-decreasing AEM to signal caution to investors amid the pandemic’s uncertainty.

Consequently, hypothesis 3, that earnings management practices varied across the JSE sectors because of the differing impact of the COVID-19 pandemic, was accepted.

Various additional tests were conducted to assess the robustness of the results. First of all, Kothari et al. (2005) suggest at least 10 observations per sector per year. This criterion was not met for the energy, telecommunications, and healthcare sectors. Consequently, these sectors were consolidated and tested together. As in the initial tests, no statistical difference was found between pre- and post-pandemic discretionary and absolute discretionary accruals for the combined sectors.

Initially, a normal OLS with no adjustments to standard errors was used to determine the coefficients, and significance was found in the same sectors. With firm and market data, there are potential issues relating to endogeneity, serial and autocorrelation and heteroscedasticity. Without wanting to change the established model, a generalised method of moments (GMM) with White two-way cluster-robust standard errors was also used to estimate the coefficients in the Kothari et al. (2005) regression equation. While endogeneity can never be fully addressed, a GMM with White two-way cluster-robust error adjustments makes the model more robust against potential regression issues. Based on the revised coefficients, the residual errors were determined. The results, using GMM to determine the betas, were consistent with those documented in Table 3.