Firm failure prediction has been extensively studied over the past decades, and the research area is in constant development, while the classical approach has been to determine the future status of a firm by using financial performance indicators, among them mainly financial ratios (Zhao et al., 2024). Still, recent literature reviews (e.g., Ciampi et al., 2021; Laitinen et al., 2023) have documented multiple drawbacks of using financial ratios in failure prediction, especially in the case of micro-, small- and medium-sized firms (MSMEs). Another issue emerges from information availability, as besides the usually long deadline to submit financial reports, distressed firms are historically known to delay their financial information over the legal deadline (Lawrence, 1983) or in the worst scenario, not to submit their financial reports at all (Andresson and Lukason, 2024).

Therefore, in recent studies the usage of non-financial information in failure prediction has gained popularity, with managerial characteristics among the most important of such predictors (Cheraghali and Molnár, 2024; Soukal et al., 2024). These variables might be especially topical for MSMEs, as there, the role of one or a few owner-managers running the firm is pivotal in comparison to larger entities (Brunninge et al., 2007). In relation to the latter, such owner-manager(s)’ general characteristics have been found to determine a firm’s failure risk (Süsi and Lukason, 2019). Despite the importance of managerial characteristics, the bulk of respective literature does not focus on their earlier behavior, but rather on more conventional corporate governance variables, while studies about MSMEs are also quite infrequent (Yousaf et al., 2024). The few specific managerial behavioral measures have, for instance, included financial performance (Altman et al., 2023) and payment behavior (Andresson and Lukason, 2024) in other firms ran by the managers earlier, managers’ personal payment defaults (Kallunki and Pyykkö, 2013), associations with other defaulted firms (Altman et al., 2020), and managers’ business networks (Kou et al., 2021), among others. In turn, the extant literature is relatively silent about the predictive value of (accounting) law violations of managers in forecasting future financial problems, and therefore, addressing the respective gap serves as the main novelty of this paper.

While the violation of accounting standards is rather common in the case of distressed firms (Habib et al., 2020), it is assumed that law violations are repetitive in the managerial career. The latter argument emerges from a multifaceted theoretical base, integrational of various streams of literature, including organizational misbehavior, accounting and financial research, being respectively discussed in the literature review section. Therefore, by relying on a diverse selection of earlier literature, this paper aims to find out how annual report submission delays (ARSDs) in managers’ earlier entrepreneurial careers can predict corporate default. Besides outlining the individual value of delays for forecasting purposes, a comparison with traditional financial ratios provided in this study helps to disclose whether the former can outrun the latter classical approach when forecasting a firm’s default. Therefore, besides introducing novel behavioral predictors of corporate default, the paper also contributes to the extant literature by disclosing their potential usefulness over a traditional approach.

The rest of the paper is structured as follows. The next section integrates various literature streams to build the foundation for the usage of ARSDs in corporate default prediction, while two hypotheses are formulated to guide the follow-up empirical strategy. The study design section outlines the characteristics of the Estonian firm population applied in the analysis, accompanied by the description of novel variables created and prediction methods implemented. The results section provides an empirical answer to the hypotheses set, being followed by theoretical implications. Besides summarizing the study, the final conclusive part also includes practical implications and ideas to develop this research domain further.

This section provides a track of the main theories that indicate why ARSDs might be valuable predictors of a firm’s default. This is achieved by integrating views from various streams of literature. Second, an array of results from earlier studies linking poor financial performance and submission delays is provided, while the settings of these studies include the former as a dependent and the latter as an independent variable, but also vice versa. This approach has a certain theoretical logic, as when considering the firm failure process, the respective events can occur sequentially. For instance, poor financial performance can be followed by a reporting delay, while the latter can in turn be followed by an onset of a payment default, which is in turn followed by even more severe delay(s), based on the conceptual scheme of events occurring in the firm failure process by Andresson and Lukason (2024).

There has been a lengthy debate on why firms operating in the same environment perform differently. A prominent explanation was provided by Hambrick and Mason (1984), who developed the upper echelons theory postulating that the previous entrepreneurial background and sociodemographic characteristics of decision makers (such as managers and board members) can be used to explain companies’ performance. The upper echelons theory has been widely used in the case of large and listed companies, while it is especially topical for MSMEs (Mayr et al., 2021), in the case of which corporate governance is very consolidated. Namely, in MSMEs, managers, board members and owners usually overlap, being often represented by a single individual (Brunninge et al., 2007). This is not merely a theoretical assumption, as population level studies have indicated that the most frequent case for MSMEs is an owner-managed firm with an unchanging board, while the same individual having taken multiple entrepreneurial endeavors during the business career (see, e.g., Süsi and Lukason, 2019). Therefore, the derivation from the upper echelons theory would be that in the case of MSMEs, managers’ earlier entrepreneurial background could play a pivotal role in explaining today’s performance.

As MSMEs are characterized by consolidated decision making, there might be practically no additional control over the daily activities, and moreover, as these firms are likely not to be audited, the same applies to the strategic management context. For instance, Woods and Joyce (2003) found that owner-managed firms are characterized by low knowledge and application of different managerial tools (among them risk and cost-benefit analysis tools) in comparison with their larger counterparts, which means that unjustified risks can be taken with a higher likelihood than in firms employing hired managers with specific experience. Large and listed firms have different levels of governance, such as a supervisory board controlling the management board in many European countries (Albert-Roulhac and Breen, 2005), but also more procedures and rules to mitigate risks. This can often slow down the decision-making process, but since decisions have been approved and questioned by different individuals, the risks related to those decisions are more likely to be reduced. Such risk reduction falls under the standard setting of agency theory, developed by Jensen and Meckling (1976) and Fama and Jensen (1983). According to the agency theory, principals’ (shareholders’) and agents’ (management’s) priorities and interests do not always align. Because of the latter, shareholders and managers could often act differently in a given situation, as for instance, managers might want to take more risks to achieve short-term goals, while owners want to protect their investment and look more into the future. Therefore, in the case a manager has a track record of risky business endeavors with a high likelihood of failure, in a multilayered governance system it is likely that business strategies with too uncertain outcomes will be rejected. In turn, owner-managed firms can often be subject to non-learning from failure due to different reasons (see, e.g., Dahlin et al., 2018).

In order to achieve desirable outcomes, managers as upper echelons make decisions and take risks. Hoskisson et al. (2017) summarized different theories, which help to explain managerial risk taking. One of the most prominent ones explaining the connection between ARSDs and firm default is the behavioral theory of the firm (BTF). Namely, the central postulate of BTF is that when a gap exists between a firm’s performance and its aspirations, it becomes more risk taking, while in the extreme, this can be in the form of illegal behavior (Hoskisson et al., 2017). Greve et al. (2010) summarized different theories of organizational misconduct, based on which it can be concluded that according to the rational choice theory, managers need to be controlled, because otherwise they are likely to choose actions which are beneficial for them and might harm third parties. Similarly, according to strain theory, not achieving goals is likely to lead to illegal activities (Greve et al., 2010). Empirical research has also indicated that corporate wrongdoing tends to be repetitive (Baucus and Near, 1991), especially in the circumstances when illegal behavior remains undiscovered or not sufficiently penalized.

One type of illegal behavior that firms engage in, is delaying publishing their annual reports over the deadline set in law, while when enough lengthy, in many countries such firms can be subject to a forced closure by the regulator (Lukason and Kantšukov, 2024). Although rules for publishing annual reports differ through countries, researchers have found that firms in a bad financial situation tend to delay their financial reports or even fail to submit reports at all rather systematically (e.g., Lukason and Camacho-Miñano, 2021; Selleslagh et al., 2021; Nguyen et al., 2022). This provides evidence of an extreme form of obfuscation theory (Courtis, 1998), as firms potentially unable to mask their bad performance with accounting techniques are engaged in holding back bad news as long as possible by not disclosing financial information at all.

While reporting delays can be associated with financial distress, the subsequent question is their potential repetitiveness in the managerial careers, for which only a few empirical examples are available. Lukason and Camacho-Miñano (2021) showed that lengthy delays in the recent past over all companies the managers had ran obtained the highest explanatory power for the same wrongdoing in the future. The study by Lukason and Kantšukov (2024) extended this finding by outlining that a firm’s forced closure because of lengthily delaying annual reports was highly determined by the same behavior in the near past. Both of those studies indicated the lower usefulness of shorter delays in explaining the respective dependent variables, while the benefit of delays as predictors also diminished when they originated from a more distant past. Besides proving the argument that managerial misbehavior can be repetitive (Baucus and Near, 1991), these findings more broadly link to the (non-)learning from failure theories. Specifically, by relying on the theoretical concept developed by Nielsen and Sarasvathy (2016), the non-learning from failure in habitual entrepreneurship potentially occurs due to managerial overconfidence. The latter can in turn be linked with the behavioral consistency theory (Funder and Colvin, 1991), implying in the current context that when financial distress occurs, mangers systematically opt for hiding financial information throughout their careers.

Based on Laitinen (2011) and Andresson and Lukason (2024), it is assumable that many firms ran by managers in the past, which had terminal delays, also witnessed payment defaults or went bankrupt. Therefore, terminal delays could serve as a valuable proxy of earlier business failures, especially when the latter remain undocumented due to different reasons. Still, based on Lukason and Kantšukov (2024) studying the determinants of firms’ forced closures, it is assumable that terminal violations do not have long-horizon predictive abilities, which is likely due to the fact that managers’ illegal behavior cannot be persistent for a lengthy period. These findings more broadly link with the extant literature, where various issues with managerial control and certain entrepreneurial characteristics have been noted as pivotal in shaping the destiny of a firm (Dieperink et al., 2025).

Several earlier studies have applied late filing for firm failure prediction, while the specific variables have portrayed the respective behavior in the same firm, not during the manager’s whole entrepreneurial career. The findings of these studies have been consolidated into Table 1, and they clearly indicate the usefulness of reporting delays as predictors of firm failure. Moreover, the studies applying categories of delays based on their length or the continuous form of the respective variable have indicated the greater benefits of longer delays in predicting failure.

Based on the theoretical narrative and extant empirical evidence, we propose that managers of defaulted firms have exhibited more ARSDs. The specific first testable hypothesis is therefore:

Hypothesis 1. With the severity of annual report submission delays in managers’ earlier entrepreneurial careers, their usefulness for corporate default prediction increases.

The probable superiority of past filing behavior over the past financial performance in explaining the onset of financial distress has roots in the theories of firm failure processes. It has been long established that firm failure can follow different pathways, ranging from a sudden deterioration to a gradual downfall over a longer period, while MSMEs are especially subject to the former, as they are more influenced by abrupt external events and have no slack resources (i.e., financial buffers) to cope with the decline (Lukason and Laitinen, 2019). Therefore, in the case of MSMEs, the predictive capabilities of financial ratios to forecast default could be modest, especially when up to date financial information is absent (Andresson and Lukason, 2024), while the accuracy of financial ratios usually drops when the prediction horizon becomes longer (Altman et al., 2020). In the case of MSMEs, another issue can emerge: for instance, Laitinen and Laitinen (2009) and Ciampi et al. (2020) have noted that MSMEs in a weak financial status are more open to manipulating financial statements. It is important to add that MSMEs’ annual reports are often unaudited, making it potentially easier for them to misreport without subsequent consequences. In addition, Ciampi (2015) concluded that financial ratios used for predicting large firms’ failure are ineffective in the case of MSMEs, which might be caused by their high volatility. Therefore, it has been suggested that only the post-default annual reports, if available, could potentially be the ones signaling financial problems with sufficient accuracy (Andresson and Lukason, 2024). The latter means that financial ratios calculated based on the information available in the annual report at the time of default might indicate performance not distinguishable (well) from that of non-defaulted counterparts. The second testable hypothesis is therefore:

Hypothesis 2. The corporate default prediction model including only annual report submission delays in managers’ earlier entrepreneurial careers is more accurate than the model including only financial ratios.

The results of logistic regressions to test Hypothesis 1 are provided in Table 3 (descriptive statistics of the variables can be followed in Appendix Table 5). The size of marginal effects (ME) for ARSDs reduces as follows: ME of SUMT_1-10 $>$ ME SUML_1-10 $>$ ME SUMM_1-10 $>$ ME SUMS_1-10. Exactly the same tendency can be followed, when either accuracy or AUC of the respective models is considered. Therefore, Hypothesis 1 can be accepted, i.e., the more severe the annual report submission delay is, the more likely is the occurrence of default after its marginal increase and the higher the accuracy and AUC of the respective prediction model.

While with the severity of ARSD the size of the marginal effect clearly increases, then concerning the accuracy and AUC, the model inclusive of the most severe violations (i.e., SUMT_1-10) is the most distinguishable from others. For instance, this is exemplified by its AUC of 0.868 being much higher than the others, which are in the range of 0.776–0.786. Concerning control variables, being an exporting or older firm, having a larger board or being functional in the service industry instead of other sectors, decrease the likelihood of default. In turn, being a micro firm increases it. The additional calculations with all ARSDs originating from either recent or distant past reveal additional important information (see Table 4). Namely, the predictive potential of ARSD variables from the recent five years is higher when compared with the more distant five years. Therefore, for practical model building purposes, delays occurring in the more recent past stand out as more beneficial predictors.

Concerning Hypothesis 2, ARSDs from the full ten-year period with control variables also included lead to accuracies in the range of 76.5%–79.3% and AUC in the range of 0.843–0.866 for the three methods applied. Out of the three methods, neural networks results in the best predictive abilities. The usage of only the recent five-year period for ARSD variables possesses slightly better predictive capabilities, the ranges being 77.4%–80.6% for accuracy and 0.857–0.878 for AUC. As the ranges of accuracy and AUC for the financial ratios are 68.3%–71.7% and 0.755–0.787, Hypothesis 2 about the superiority of ARSDs over financial ratios can be accepted in the case of all three methods applied. The benefit of using only ARSDs from the recent period over financial ratios is 8.9 percentage points of accuracy and 0.091 units of AUC. When ARSDs from the distant period are applied, their benefit over pre-default financial ratios in prediction disappears. Finally, the predictive potential of using ARSDs with financial ratios has been assessed. The best result is obtained with neural networks, leading to 82.1% accuracy and 0.895 AUC, while the increment from the best model including only ARSDs is small (the respective figures being 80.6% and 0.878).

As neural networks possesses the best predictive capabilities, the variables useful for the respective prediction are commented by using their normalized importance (NI). Table 4 lists the variables having at least 50% NI in each of the models. All those models that include ARSDs from either the full ten-year period or from the recent five years are characterized by SUMT being the most important variable. In those models the other important variables mainly include different control variables, concerning which it can be generalized that firm age, board size and being a micro firm are important predictors. In turn, being an exporter and sectoral status never appear among the important variables. In the same models also inclusive of financial ratios, total equity to total assets ratio has certain importance, followed by a ratio of net income to total assets reflecting the ceteris paribus change of the former equity ratio with a borderline importance. In turn, in the models which include ARSDs from the more distant period, the latter have only borderline importance and being represented only by SUMT. In those models, the controls (or the financial ratios noted earlier, when included) have the greatest relevance, while the predictive capabilities of such models remain modest.

Finally, an additional robustness check was conducted with neural networks in a way that defaults occurring during 2017–2019 and the corresponding proportion of non-defaulted firms were separated into a control group. Thereafter, three neural networks models were created and validated on the respective control group. The difference of the control group’s accuracy from that of the test group ranged from –0.3 to +0.9 percentage points. This indicates a relative homogeneity of the observations from different time periods.

Based on the research, the following scientific implications can be highlighted. This research provides support to the upper echelons, non-learning from failure and organizational misconduct theories. Managers’ entrepreneurial history can be used to predict firm failure, supporting the central idea in the upper echelons theory by Hambrick and Mason (1984) that firm outcomes can be dependent on managerial behavior. In addition, these managers tend not to learn from their previous mistakes and continuously fail in entrepreneurship. In this study, it is specifically manifested in the link between default and leaving annual reports terminally non-submitted in the earlier entrepreneurial career, which because of its permanency leads to cessation of the firm’s activities. Therefore, a non-learning route in habitual entrepreneurship (Nielsen and Sarasvathy, 2016) is clearly visible. The latter also links with the general repetitive illegal behavior idea in Baucus and Near (1991), and more specifically, with the repetitive non-submission behavior (Lukason and Camacho-Miñano, 2021; Lukason and Kantšukov, 2024), especially as defaulted firms had several non-submitted annual reports at the moment of deletion. When linking serial violations with the fact that managers might want to obfuscate business failures (Courtis, 1998), being operationalized as poor financial performance in many available empirical studies, the managers of defaulted firms have potentially witnessed either defaults or poor financial performance in their earlier careers as well.

As a novel finding into the strand of literature linking entrepreneurial misconduct and firm failure, this study indicated that from the delays ranked by means of their severity, the most severe violations have strong corporate default prediction capabilities. Leaving annual reports terminally non-submitted results in a predictive power exceeding remarkably that of four classical financial ratios. The models inclusive of either short-, medium- or long-term delays have all weaker predictive properties, not differing considerably from each other, but still exceeding the 70% threshold of accuracy. Still, the marginal effects clearly indicate the ascending order of delays’ predictive abilities based on their severity. That leads to a theoretical proposition that there is a pecking order between the future corporate default and the severity of past annual reports’ submission delays in the managerial career. In the extant literature, the latter fact has so far been mainly known for a single company (e.g., Keasey and Watson, 1988; Laitinen, 2011).

It was also found that the timing of the misconduct matters, which extends the available theoretical narrative further. Specifically, unlike the earlier research, this paper showed that failure to publish financial report(s) during the recent five years before the occurrence of default is more meaningful than the same phenomenon from the more distant past. Therefore, the repetitive misconduct behavior outlined by Baucus and Near (1991) is in the payment default prediction context applicable in a short-range setting. Another important implication derived from the latter is that non-learning from failure might not be persistent over a lengthy time horizon. This would extend the concept by Nielsen and Sarasvathy (2016) of (non-)learning from failure pathways in habitual entrepreneurship by introducing its specific applicability timeframe in the MSMEs’ illegal behavior context.

An important finding for the literature of firm failure prediction is that for the early warning of a forthcoming default, misbehavior by means of terminal non-submission of annual reports can be applied separately without including financial ratios in the respective models. This directly answers the recent call by Ciampi et al. (2021) to find novel non-financial failure predictors. The latter is especially topical in the case of young firms, which have not yet submitted financial reports, or when the submitted financial reports for the majority of firms indicate poor performance, as they have not yet surpassed the breakeven level to be profitable. In support of the latter argument, the accuracies of financial ratios for predicting the failure of young MSMEs have often been modest (e.g., Laitinen, 1992; Fuertes-Callén et al., 2022).

This paper studied the abilities of annual report submission delays (ARSDs) in managers’ earlier careers to predict the default of a firm currently managed by them. For that purpose, the population of Estonian (non-)defaulted firms was used. Novel variables reflected ARSDs with varying severity, while their emergence time by means of recent and distant history was considered as well. To exemplify the behavior of individual ARSDs, average marginal effect, accuracy and AUC from the logistic regressions were used, while the multivariate prediction models inclusive of all ARSDs were composed with three different methods. The resulting multivariate models based on ARSDs were compared against models including classical financial ratios.

The results indicated that a specific type of delay, namely leaving annual reports terminally non-submitted, obtains the highest accuracy in the prediction models. Moreover, these terminal delays are useful when they emerge from the recent managerial history. Through marginal effects it is visible that with the severity of delays, the likelihood of default increases. Prediction models including ARSDs can outrun those including financial ratios by around nine percentage points, validating the usefulness of these non-financial variables in corporate default prediction.

Concerning the practical implications, novel non-financial variables useful for increasing the accuracy of corporate default prediction models were introduced. Therefore, the inclusion of terminal annual report non-submissions the manager has conducted during recent entrepreneurial career is suggested in the respective prediction models. The latter predictor, when supplemented with certain variables reflecting firm type, would lead to acceptably accurate prediction of corporate default. This is especially important when the firm has no up to date financial information or in the worst scenario, has not submitted any financial reports. In such circumstances it is inevitable to find valuable non-financial predictors. Moreover, the findings would allow to propose that potentially various other severe violations could fit for that purpose, especially when derived from the behavioral logic “when you did it once, you will do it again”. As unlike in many other studies, the start of a payment default was predicted, the proposed approach provides a reasonable early warning possibility, which in the case of MSMEs might not be possible with classical variables such as financial ratios, being potentially more suited for factual insolvency (e.g., bankruptcy declaration) prediction and for the segment of larger firms. For instance, the usage of the created models could be suited for certain lending decisions, such as the provision of trade credit to newly founded companies missing a financial history.

Several limitations of the paper have to be accounted. Although the non-submission of annual reports is usual in other countries as well, especially in the case of firms in a poor financial status, the respective laws and their implementation can still be to a certain extent country-specific. Therefore, before drawing finite conclusions about the universal applicability of the respective variables, revalidation of the same research setting in other environments is of essence. Potentially in countries where delays are more strictly penalized, managers are less likely to conduct them, while in turn in a liberal legal setting, probably the opposite occurs.

The paper can be developed further in multiple ways, besides the inter-country comparison noted earlier. For instance, the prediction accuracy might potentially be enhanced with the usage of a larger variety of variables portraying different law violations. These could include different accounting manipulations, non-submission of other compulsory reports such as tax declarations, but also tax fraud instances, among others. In addition, a larger variety of financial variables and in a lengthier timeframe (if available for micro firms) could be included in future studies.

Data cannot be shared, as it was obtained from a third party, who does not allow redistribution.

AA and OL designed the research of the study. AA obtained the data. AA and OL performed the data analysis. AA composed the first draft of the manuscript, which was then modified by OL. Both authors contributed to the critical revision of the manuscript for important intellectual content. Both authors reviewed and approved the final version of the paper. Both authors have participated sufficiently in the work and agreed to be accountable for all aspects of the work.

The authors thank three anonymous reviewers for useful comments.

This work was supported by the Estonian Research Council under grant PRG1418 “Export(ers’) Performance in VUCA and Non-VUCA Environments”.

The authors declare no conflict of interest.

See Table 5.