Predictive Model as Screening Tool for Early Warning of Corporate Insolvency in Risk Management: Case Study from Slovak Republic

Abstract

Bankruptcy prediction in Slovakia’s industrial manufacturing sector is vital due to its significant role in the national economy. This study aims to develop a predictive model for forecasting corporate bankruptcy within the industrial manufacturing sector in Slovakia. The novelty of this study lies in developing a model tailored to crisis conditions, validated using COVID-19 data, and adapted to the Central European context for greater accuracy and relevance. The model is constructed using financial data extracted from the Orbis database, based on company financial statements from 2020 and 2021, and encompasses firms of various sizes. Employing backwards binary logistic regression, five statistically significant predictors were identified, enabling the model to forecast impending bankruptcy with a one-year lead time. The model was trained on a sample of 1305 companies and achieves an overall prediction accuracy of 83.78%, with an AUC (Area Under the Curve) value of 91.7%, indicating strong discriminative power. The resulting model demonstrates robust predictive capability and may serve as a practical decision-support tool for managers, investors, creditors, and other stakeholders assessing the financial health of firms.

1. Introduction

The contemporary business environment is undergoing constant transformation due to globalization and rapid technological advancement. Enterprises are increasingly exposed to a wide array of economic, social, and technological pressures. In such a dynamic and uncertain context, the early identification of financial distress and potential bankruptcy has become critically important.

Among the core areas of corporate risk management, bankruptcy risk occupies a central position. Early detection of this risk enables firms to implement preventive measures aimed at mitigating the adverse effects on their operations and long-term viability. As uncertainty in the market environment continues to grow, so too does the need for robust analytical tools, such as predictive models, that can help businesses and their stakeholders minimize potential losses.

Considering the current economic and market conditions in the Slovak Republic, the issue of corporate bankruptcy prediction is particularly relevant. Companies in the industrial manufacturing sector, as well as in many other industries, frequently face financial challenges arising from a variety of sources. High fixed costs, intense competition in both domestic and international markets, supply chain dependencies, excessive tax burdens, and other structural or macroeconomic factors all contribute to the erosion of financial stability. This underscores the importance of developing accurate and sector-specific tools for early risk assessment and strategic decision-making.

The primary aim is to develop a predictive model for forecasting corporate bankruptcy within the Slovak industrial manufacturing sector, using financial data from 2020–2021. We acknowledge that our data comes from the COVID-19 period. However, this is precisely the novelty of our work: the model is designed for crisis contexts, where traditional assumptions often break down. Using COVID-19 data allows us to validate the model under exceptional conditions, which we see as a strength rather than a limitation.

This paper contributes to the literature by

▪

Identifying five statistically significant predictors of bankruptcy through backwards binary logistic regression.

▪

Demonstrating strong predictive performance with an overall accuracy of 83.78% and an AUC of 91.7%, confirming the model’s discriminative power.

▪

Providing a practical decision-support tool for managers, investors, creditors, and other stakeholders to assess financial health and anticipate potential corporate distress.

▪

Offering insights into financial distress dynamics during the COVID-19 crisis, a period of exceptional economic conditions, thereby extending the applicability of classical insolvency prediction models to crisis contexts.

Focusing on the Slovak industrial sector is therefore justified by its significant role in the national economy and its sensitivity to economic shocks. Analyzing bankruptcy prediction within this sector provides valuable insights for managers, investors, and other stakeholders when assessing the financial stability and risk of insolvency of firms in this key industry.

2. Literature Review

Enterprise risk management has developed into a multidisciplinary field, evolving significantly over the past several decades. Its foundations were laid in the 1950s, when businesses in the United States began relying on insurance as the primary means of mitigating risk. However, the rising cost of insurance and its limited coverage soon highlighted the inadequacy of insurance alone in addressing the full spectrum of business risks. Organisations began to recognise the need for more initiative-taking and comprehensive strategies to protect their assets and personnel (Hopkins, 2017) [1].

This evolving perspective on risk management reached Europe in the 1970s. Companies began to focus more closely on the financial consequences of risk and acknowledged that many emerging threats could not be insured. In response, businesses started to adopt a wider range of tools and techniques aimed at managing risk more effectively. This marked the emergence of the modern risk management framework—an integrated approach that combines financial mechanisms with preventive measures to enhance organisational resilience in the face of diverse threats (Hopkins, 2017) [1].

In recent years, the development of bankruptcy prediction models has expanded significantly beyond traditional statistical methods to include a variety of advanced approaches, such as genetic algorithms, decision trees, neural networks, support vector machines, and artificial intelligence. Despite this progress, a substantial number of model developers continue to rely on simpler statistical techniques, as they are often regarded as more reliable tools for corporate bankruptcy prediction (Ashraf et al., 2019) [2].

We summarize selected bankruptcy prediction models, with a particular focus on those developed for the Visegrad Group countries and other European nations. Models based on data from firms operating in European countries tend to reflect the financial realities of Slovak enterprises more accurately, as they account for region-specific factors such as accounting standards, tax systems, legal frameworks, economic structure, business environment, and various macroeconomic conditions, including business cycles and access to financing. This review concentrates primarily on models built using post-2000 data.

The selection emphasizes models developed for industrial firms—e.g., Herman (2017) [3], Ježovita (2015) [4], and Voda et al. (2021) [5]—but also includes models spanning broader economic sectors, such as those by Svabova et al. (2020) [6] and Durica and Adamko (2016) [7]. Most of the reviewed models pertain to the manufacturing and processing industries.

Among the models reviewed, some of the largest datasets were utilized by Němec and Pavlík (2016) [8], Durica, Valaskova, and Janoskova (2019) [9] and Durica and Adamko (2016) [7]. Němec and Pavlík (2016) [8] constructed a logistic regression model using data from 175,556 Czech companies across various industries for the period 2005–2013. Their model predicts bankruptcy one year in advance with an accuracy of 84%. Durica and Adamko (2016) [7] employed multiple discriminant analysis to develop a model based on 109,550 Slovak firms from diverse sectors. Their five-variable model achieves a prediction accuracy of 82.2%, with firms classified as bankrupt when the model function value falls below −0.061.

Durica, Valaskova, and Janoskova (2019) [9] built a logistic regression model using data from more than 170,000 companies across the Visegrad countries. The model incorporates ten significant predictors, including financial indicators and a non-financial variable representing the country of origin. Slovakia serves as the reference category. The model achieves a classification accuracy exceeding 88%.

Svabova et al. (2020) [6] developed a hybrid model tailored to the Slovak economic environment using a dataset comprising 75,652 small and medium-sized enterprises from various industries based in Slovakia. By combining discriminant analysis and logistic regression, the model utilises financial data from 2016 to 2018 and can identify bankrupt firms one year in advance with an accuracy of 93.4%. Střelec & Staňková (2024) [10], Vukčević et al. (2024) [11] and Kuster et al. (2025) [12] significantly contribute to the development of bankruptcy prediction models by applying logistic regression across various economic settings—the forestry sector in the EU, Montenegrin enterprises, and Serbian companies. Each highlights the importance of optimizing model thresholds and incorporating both financial and non-financial indicators to enhance prediction reliability. Collectively, they provide strong empirical support for the versatility and practical value of logistic regression in assessing corporate financial stability.

The most parsimonious model in terms of the number of predictors is the minimalist model by Pavličko and Mazanec (2022) [13], which was developed based on a literature review focusing on Central European models. It includes only two variables deemed essential for assessing bankruptcy risk: return on assets (ROA, EAT/total assets) and the debt ratio (total liabilities/total assets). This model successfully identifies bankrupt Slovak firms with an accuracy of up to 94.04% and achieves an average accuracy of 95.07% across the Visegrad countries.

Table 1. Summary.

Variable	Formula	Category	Authors	Number
Debt Ratio	Total Liabilities/Total Assets	Indebtedness	Ohlson (1980) [14], Zmijewski (1984) [15], Němec and Pavlík (2016) [8], Herman (2017) [3], Adamko, Klieštik, and Kováčová (2018) [16], Klieštik, Vrbka a Rowland (2018) [17], Alaminos, Del Castillo, and Fernández (2016) [18], Voda et al. (2021) [5], Svabova et al. (2020) [6], Pavličko a Mazanec (2022) [13], Ekes and Koloszar (2014) [19], Durica, Valaskova, and Janoskova (2019) [9]	12
Current Ratio	Current Assets/Current Liabilities	Liquidity	Zmijewski (1984) [15], Shirinkina a Valiullina (2015) [20], Ježovita (2015) [4], Mihalovič (2016) [21], Alaminos, Del Castillo and Fernández (2016) [18], Klieštik, Vrbka and Rowland (2018) [17], Voda et al. (2021) [5], Adamowicz and Noga (2018) [22], Němec and Pavlík (2016) [8], Ďurica and Adamko (2016) [7], Svabova et al. (2020) [6], Durica, Valaskova, and Janoskova (2019) [9]	12
Working Capital to Total Assets Ratio	Working Capital/Total Assets	Liquidity	Altman (1968) [23], Ohlson (1980) [14], Durica and Adamko (2016) [7], Almamy, Aston and Ngwa (2016) [24], Mihalovič (2016) [21], Adamko, Klieštik, and Kováčová (2018) [16], Jenčová, Štefko and Vašaničová (2020) [25], Jakubík and Teplý (2011) [26], Butkus, Žakarė, and Cibulskienė (2014) [27], Durica, Valaskova and Janoskova (2019) [9]	10
ROA (based on EAT)	EAT/Total Assets	Profitability	Ohlson (1980) [14], Zmijewski (1984) [15], Blums and College (2004) [28], Shirinkina and Valiullina (2015) [20], Alaminos, Del Castillo and Fernández (2016) [18], Mihalovič (2016) [21], Klieštik, Vrbka and Rowland (2018) [17], Pavličko and Mazanec (2022) [13], Butkus, Žakarė, and Cibulskienė (2014) [27], Durica, Valaskova and Janoskova (2019) [9]	10
ROTA (based on EBIT)	EBIT/Total Assets	Profitability	Altman (1968) [23], Ďurica and Adamko (2016) [7], Almamy, Aston, and Ngwa (2016) [24], Alaminos, Del Castillo, and Fernández (2016) [18], Klieštik, Vrbka, and Rowland (2018) [17], Voda et al. (2021) [5], Horváthová, Mokrišová, and Petruška (2021) [29], Adamko, Klieštik, and Kováčová (2018) [16]	8
Equity Ratio	Equity/Total Liabilities	Indebtedness	Němec and Pavlík (2016) [8], Valiullina (2015) [20] and Adamko (2016) [7], Almamy, Aston and Ngwa (2016) [24], Herman (2017) [3], Altman (1968) [23], Adamowicz and Noga (2018) [22], Ekes and Koloszar (2014) [19], Horváthová, Mokrišová, and Petruška (2021) [29], Butkus, Žakarė, and Cibulskienė (2014) [27]	8
Asset Turnover	Sales/Total Assets	Activity	Altman (1968) [23], Shirinkina a Valiullina (2015), Almamy, Aston, and Ngwa (2016) [24], Horváthová, Mokrišová, and Petruška (2021) [29], Svabova et al. (2020) [6], Jakubík and Teplý (2011) [26], Durica, Valaskova, and Janoskova (2019) [9]	7
ROE (based on EAT)	EAT/Equity	Profitability	Shirinkina and Valiullina (2015) [20], Ježovita (2015) [4], Klieštik, Vrbka, and Rowland (2018) [17], Svabova et al. (2020) [6], Jakubík and Teplý (2011) [26], Durica, Valaskova, and Janoskova (2019) [9]	6
Current Assets to Total Assets Ratio	Current Assets/Total Assets	Asset Structure	Shirinkina and Valiullina (2015) [20], Mihalovič (2016) [21], Alaminos, Del Castillo, and Fernández (2016) [18], Klieštik, Vrbka, and Rowland (2018) [17], Grünberg and Lukason (2014) [30]	5
Current Debt Ratio	Current Liabilities/Total Assets	Indebtedness	Blums and College (2003) [28], Mihalovič (2016) [21], Klieštik, Vrbka, and Rowland (2018) [17]	3
Quick Ratio	(Current Assets–Inventory)/Current Liabilities	Liquidity	Jenčová, Štefko, and Vašaničová (2020) [25], Sfakianakis (2021) [31], Durica, Valaskova, and Janoskova (2019) [9]	3
Retained Earnings to Total Assets Ratio	Retained Earnings/Total Assets	Indebtedness	Altman (1968) [23], Almamy, Aston and Ngwa (2016) [24]	2
EAT and Depreciation to Total Liabilities Ratio	(EAT + Depreciation)/Total Liabilities	Indebtedness	Adamko, Klieštik, and Kováčová (2018) [16], Voda et al. (2021) [5]	2
Financial Leverage	Total Assets/Equity	Indebtedness	Jenčová, Štefko, and Vašaničová (2020) [25], Němec and Pavlík (2016) [8]	2
Cash and Cash Equivalents to Total Assets	Cash and Cash Equivalents/Total Assets	Liquidity	Durica, Valaskova, and Janoskova (2019) [9], Klieštik, Vrbka, and Rowland (2018) [17]	2
Profit Margin	EAT/Sales	Profitability	Durica, Valaskova, and Janoskova (2019) [9], Svabova et al. (2020) [6]	2

presents the sixteen most frequently used financial indicators identified across the analysed models. The list includes all indicators that appeared in at least two models simultaneously. The indicators are primarily classified into four categories such as activity, liquidity, indebtedness, and profitability. The most frequently occurring variables are the debt ratio (total liabilities/total assets) and current ratio (current assets/short-term liabilities). Overall, indicators related to indebtedness are the most represented, with a total of six occurrences.

Existing studies often overlook the specific conditions of Central and Eastern European economies, focusing on U.S. or Western European contexts. Our study addresses this gap by developing and reviewing models based on data from the Visegrad countries, reflecting regional financial and economic realities during COVID-19. The selected models and authors were chosen for their methodological relevance, data coverage, and applicability to industrial firms, aligning directly with the objectives of our research.

3. Materials and Methods

3.1. Input Data

Data were extracted from the Orbis database, maintained by Moody’s Analytics–Bureau van Dijk. Orbis is a comprehensive database providing financial and business information on companies, primarily in Europe. Our dataset comprised financial statement data of companies operating in Slovakia for the period 2020 and 2021. The dataset included 4187 companies, all operating in the industrial sector (Moody’s, 2021) [32].

The Orbis database categorizes companies into four size classes: small, medium, large, and very large. This classification is based on three key indicators: sales, total assets, and number of employees. The company is classified as medium-sized if its total assets range from €2 million to €20 million, its operating revenue falls between €1 million and €10 million, and it employs between 15 and 150 individuals. The criteria for larger company classifications are detailed in

Table 2. Company classification by size.

Company Size	Sales (Million Euros)	Total Assets (Million Euros)	Number of Employees
Very large company	≥100	≥200	≥1000
Large company	≥10	≥20	≥150
Medium company	≥1	≥2	≥15
Small company	Other companies are not included in the remaining three categories.

. Companies not meeting the criteria for medium, large, or very large classifications are categorized as small enterprises (Moody’s, 2021) [32].

3.2. Sample

As a preliminary step in data filtering, 309 companies were excluded from the dataset due to missing data necessary for the calculation of financial ratios or the verification of insolvency criteria. Following this exclusion, 3878 companies were retained for further analysis.

Figure 1 shows that the final sample consists of 1866 companies, with the majority (55%) classified as medium-sized enterprises. The smallest proportion is represented by very large enterprises, accounting for only 5% of the sample. Overall, approximately 36% of the firms are identified as financially distressed. To ensure model validation, the dataset was divided into two subsets: a training sample and a testing sample, using a 70:30 ratio.

The training sample comprises 69.93% of the observations (1305 companies), while the testing sample consists of 30.07% (561 companies). This split was conducted randomly. The training subset is used to build the model, whereas the testing subset is employed to evaluate its performance. This type of division ensures objective validation of the model on previously unseen data, helping to prevent artificial inflation of accuracy.

The proportions of distress and non-distress companies are balanced across both subsets. Specifically, the share of distressed firms in the training sample is 35.17%, compared to 37.25% in the testing sample. Figure 2 presents both the absolute and relative frequencies of companies in the training and testing subsets.

3.3. Dependent Variable

For the binary logistic regression analysis, firms were classified into two categories based on their financial health: distress and non-distress companies. All statistical analyses, including the evaluation of model assumptions, were conducted using IBM SPSS Statistics 25. Following the guidelines proposed by Durica and Frnda (2021) [33], two criteria were used to identify distressed firms. Due to the lack of data regarding the number of creditors, the first criterion suggested by Durica and Frnda (2021) [33] was omitted. Instead, total liquidity was used as the first criterion, where firms with a total liquidity ratio below one were classified as distressed.

The second criterion was the ratio of equity to liabilities. Firms with a ratio below 0.014 were classified as distressed. This threshold was adjusted upward compared to the original value used by Durica and Frnda (2021) [33] to reflect the annual tightening of this criterion. Firms meeting at least one of the two criteria were coded as “1” (distress), while those that met neither were coded as “0” (non-distress) for the logistic regression.

Using financial data from 2021, 916 firms were identified as distressed under the first criterion and 1269 under the second. Altogether, from a total sample of 3878 firms, 1562 met at least one criterion for distressed companies. This number is lower than the sum of firms identified by each criterion due to the overlap between the two groups.

3.4. Independent Variables

The independent variables used in the analysis are financial ratios derived from the 2020 dataset. The selection process was guided by a review and aggregation of variables employed in established predictive models, primarily those based on logistic regression, with an emphasis on applications within the industrial sector across European countries.

An initial set of 16 financial ratios was identified from these models. However, due to a significant amount of missing data or the complete unavailability of accounting entries required for their calculation, two variables, the ratio of retained earnings to total assets and the proportion of short-term financial assets to total assets, were excluded from further analysis.

As a result,

Table 3. Input variables.

Variables	Formula	Category
Debt Ratio	Total Liabilities/Total Assets	Indebtedness
Current Ratio	Current Assets/Current Liabilities	Liquidity
Working Capital to Total Assets Ratio	Working Capital/Total Assets	Liquidity
ROA (based on EAT)	EAT/Total Assets	Profitability
ROTA (based on EBIT)	EBIT/Total Assets	Profitability
Equity Ratio	Equity/Total Liabilities	Indebtedness
Asset Turnover	Sales/Total Assets	Activity
ROE (based on EAT)	EAT/Equity	Profitability
Current Assets to Total Assets Ratio	Current Asset/Total Assets	Asset Structure
Current Debt Ratio	Current Liabilities/Total Assets	Indebtedness
Quick Ratio	(Current Assets–Inventory)/Current Liabilities	Liquidity
EAT and Depreciation to Total Liabilities	(EAT + Depreciation)/Total Liabilities	Indebtedness
Financial Leverage	Total Assets/Equity	Indebtedness
Profit Margin	EAT/Sales	Profitability

shows the final set of independent variables includes 14 financial ratios considered potentially statistically significant.

3.5. Hypothesis

We include a model hypothesis to confirm the statistical validity and predictive capability of the proposed framework.

H0: 

The logistic regression model does not significantly predict the likelihood of financial distress of companies.

H1: 

The logistic regression model significantly predicts the likelihood of financial distress of companies.

3.6. Methods

To ensure the appropriate application of binary logistic regression, it is essential to adhere to a set of fundamental assumptions. These assumptions are critical for the reliability and validity of the logistic regression analysis. A thorough understanding of these conditions is necessary for proper model specification, accurate interpretation of results, and correct implementation of logistic regression in practice. Binary logistic regression relies on the following six key assumptions:

▪

Binary dependent variable. The outcome variable must be dichotomous, meaning it takes on two categorical values (e.g., success/failure, yes/no).

▪

Independence of Observations. The observations should be mutually independent, implying that the outcome of one observation does not influence the outcome or prediction of another. For instance, one respondent’s answer should not affect another’s. Ensuring independence is crucial for the predictive accuracy of the model.

▪

No or minimal multicollinearity. There should be no strong multicollinearity among the independent variables. In other words, the predictors should not be highly correlated with each other. High multicollinearity can distort the estimation of regression coefficients and reduce model interpretability.

▪

Linearity of independent variables with the logit. There should be a linear relationship between the continuous independent variables and the logit transformation of the dependent variable. This assumption can be evaluated using graphical methods, such as plotting individual predictors against the logit. However, this approach has limitations, particularly in models with multiple predictors, and should be interpreted cautiously.

▪

Absence of strong outliers. Logistic regression models are sensitive to outliers. Tools such as boxplots and histograms are commonly used to detect the presence of extreme or influential data points, which can adversely affect model performance. Before model estimation, outliers were removed to prevent distortion of the results, and companies with missing financial data were excluded from the analysis. This preprocessing ensured that the dataset used for model estimation was complete and reliable, supporting the robustness and validity of the predictive model. All outliers identified using the standard IQR method (values below Q1 − 1.5 × IQR or above Q3 + 1.5 × IQR) were removed from the dataset to ensure the robustness of the results.

▪

Large sample size. A sufficiently large dataset is necessary to ensure the stability and generalizability of the model. Several rules exist for determining the minimum sample size, most of which are based on the ratio of observations to explanatory variables. One widely used guideline is the Events per Variable (EPV) rule, which recommends a minimum of 10 events (i.e., occurrences of the outcome of interest) for each independent variable included in the model.

Meeting these assumptions substantially enhances the accuracy and reliability of the logistic regression model, especially when applied to future datasets (Tranmer a kol., 2024) [34].

4. Results

Table 4. Descriptive statistics for the initial sample.

	Mean	Median	Standard Deviation	Kurtosis	Skewness	Interquartile Range	Variance	Min	Max
Debt Ratio	0.81	0.60	1.39	1.93	234.7	12.2	36.9	−0.2	36.7
Current Ratio	2.56	1.34	4.60	21.14	116.9	8.2	122.3	−14.3	108.0
Working Capital to Total Assets Ratio	0.15	0.11	0.86	0.74	181.8	1.1	40.3	−18.8	21.5
ROA	0.03	0.02	0.21	0.04	159.7	0.7	9.4	−4.2	5.2
ROTA	0.05	0.04	0.21	0.05	77.1	−2.9	6.4	−4.1	2.3
Equity Ratio	2.22	0.66	7.77	60.39	688.9	21.0	336.6	−30.0	306.6
Asset Turnover	1.60	1.25	1.55	2.40	37.0	4.5	23.3	0.0	23.3
ROE	−0.08	0.07	5.88	34.53	787.5	−21.5	307.4	−202.7	104.6
Current Assets to Total Assets Ratio	0.70	0.57	0.94	0.88	245.2	12.0	25.7	0.0	25.7
Current Debt Ratio	0.55	0.38	0.90	0.81	336.0	13.4	30.4	−0.2	30.1
Quick Ratio	2.14	0.89	7.35	54.03	872.5	24.0	314.6	−2.2	312.4
EAT and Depreciation to Total Liabilities Ratio	0.35	0.15	1.00	1.01	279.8	12.6	35.9	−6.1	29.8
Financial Leverage	10.36	2.19	150.11	22,532.04	1777.1	38.0	9354.9	−1834.3	7520.6
Profit Margin	−0.31	0.02	15.17	230.20	3385.9	−57.0	920.0	−912.5	7.5

presents the basic statistical characteristics of individual variables in absolute values, based on a sample of 3878 companies. For several indicators, such as financial leverage, profit margin, equity ratio, current ratio, and profitability measures, very wide interquartile ranges are observed. Notably, the minimum value of total liquidity reaches as low as −14.3, which is highly atypical and may signal data errors or inconsistencies in companies’ balance sheets.

The arithmetic mean of return on equity (ROE) stands at −8%, whereas the median is positive. This discrepancy suggests the mean is highly sensitive to outliers and extreme values, due to some companies operating at significant losses or possessing highly irregular capital structures. Similarly, the mean financial leverage exceeds a value of 10, which is also uncharacteristically high and may indicate either accounting inaccuracies or extreme levels of indebtedness within certain firms.

Such unnatural values and statistical anomalies must be addressed before further data processing. This is typically achieved through the exclusion of outliers and extreme values, ensuring a cleaner and more representative dataset for robust analysis.

Table 5. Descriptive statistics for the final sample.

	Mean	Median	Standard Deviation	Kurtosis	Skewness	Interquartile Range	Variance	Min	Max
Debt Ratio	0.62	0.60	0.30	0.09	1.07	0.76	1.76	0.05	1.81
Current Ratio	1.50	1.28	0.90	0.81	1.29	1.16	4.99	0.00	4.99
Working Capital to Total Assets Ratio	0.11	0.11	0.27	0.08	0.14	−0.12	1.85	−0.79	1.05
ROA	0.03	0.02	0.05	0.00	0.94	0.65	0.31	−0.11	0.19
ROTA	0.05	0.04	0.06	0.00	0.84	0.80	0.36	−0.11	0.25
Equity Ratio	0.97	0.67	0.84	0.70	1.57	1.40	4.75	−0.60	4.15
Asset Turnover	1.42	1.29	0.76	0.58	0.28	0.77	3.86	0.03	3.89
ROE	0.08	0.06	0.12	0.02	0.84	0.47	0.78	−0.29	0.49
Current Assets to Total Assets Ratio	0.54	0.53	0.27	0.07	0.19	0.49	1.58	0.00	1.58
Current Debt Ratio	0.43	0.40	0.22	0.05	0.58	0.74	1.31	0.02	1.33
Quick Ratio	1.00	0.81	0.71	0.50	1.75	1.33	3.97	0.02	3.72
EAT and Depreciation to Total Liabilities Ratio	0.20	0.15	0.17	0.03	1.39	1.16	1.08	−0.26	0.82
Financial Leverage	3.14	2.49	1.98	3.93	1.12	1.13	12.83	−3.31	9.52
Profit Margin	0.02	0.02	0.04	0.00	1.07	0.28	0.24	−0.09	0.15

shows the basic statistical characteristics of the independent variables after removing outliers and extreme values. Compared to the initial descriptive statistics, the resulting dataset is significantly more representative. The arithmetic mean for all indicators is very close to the median, indicating that the data are symmetric and evenly distributed around the centre. This suggests the absence of extreme or outlier values that could negatively influence the analytical outcomes.

On average, industrial enterprises in Slovakia exhibit a debt ratio of 62%, implying that they finance their assets through external sources. The average value of the quick ratio falls within the optimal interval of <0.65; 1>, as recommended by Costea and Hostiuc (2009) [35]. This suggests that, on average, firms can fully cover their short-term liabilities with short-term financial assets and receivables. Overall liquidity is also situated near the ideal benchmark, both in terms of the mean and median.

The profit margin indicator remains low. On average, Slovak industrial firms generate only €0.02 of net profit per €1 of revenue. Such low profitability implies limited internal resources for innovation and growth and restricts the ability to build sufficient reserves. Consequently, this increases firms’ vulnerability to market fluctuations, unexpected expenses and contributes to higher levels of indebtedness. In terms of return on assets (ROA) and return on equity (ROE), performance is suboptimal; however, this is to be expected in capital-intensive industries, where lower returns are common.

The financial leverage ratio illustrates the extent to which firms rely on external financing. In our sample, the average value is 3.14, meaning that for every €1 of equity, there are €3.14 of liabilities. This confirms a high level of indebtedness among Slovak enterprises. Nevertheless, such values are typical in the industrial sector, where substantial initial investments in equipment and infrastructure are often required. Leverage ratios between 3 and 5 are considered standard in this context.

Table A1. Correlation matrix.

Variable		Debt Ratio	Current Ratio	Working Capital to Total Assets Ratio	ROA	ROTA	Equity Ratio	Assets Turnover	ROE	Current Assets to Total Assets Ratio	Current Debt Ratio	Quick Ratio	EAT and Depreciation to Total Assets	Financial Leverage	Profit Margin
Debt Ratio	Pearson r	1	−0.406 **	−0.326 **	−0.106 **	−0.043	−0.697 **	0.173 **	0.016	0.158 **	0.604 **	−0.346 **	−0.450 **	0.216 **	−0.140 **
	Sig. (2-tailed)		0.000	0.000	0.000	0.066	0.000	0.000	0.499	0.000	0.000	0.000	0.000	0.000	0.000
Current Ratio	Pearson r		1	0.844 **	0.16	0.149 **	0.608 **	0.055*	−0.001	0.422 **	−0.537 **	0.648 **	0.281 **	−0.354 **	0.144 **
	Sig. (2-tailed)			0.000	0.000	0.000	0.000	0.018	0.965	0.000	0.000	0.000	0.000	0.000	0.000
Working Capital to Total Assets Ratio	Pearson r			1	0.227 **	0.216 **	0.494 **	0.206 **	0.037	0.678 **	−0.417 **	0.582 **	0.215 **	−0.427 **	0.143 **
	Sig. (2-tailed)				0.000	0.000	0.000	0.000	0.113	0.000	0.000	0.000	0.000	0.000	0.000
ROA	Pearson r				1	0.972 **	0.153 **	0.269 **	0.826 **	0.159 **	−0.089 **	0.233 **	0.660 **	−0.207 **	0.829 **
	Sig. (2-tailed)					0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000
ROTA	Pearson r					1	0.114 **	0.315 **	0.806 **	0.176 **	−0.054 *	0.220 **	0.633 **	−0.205 **	0.786 **
	Sig. (2-tailed)						0.000	0.000	0.000	0.000	0.020	0.000	0.000	0.000	0.000
Equity Ratio	Pearson r						1	−0.022	−0.082 **	0.090 **	−0.509 **	0.460 **	0.541 **	−0.515 **	0.155 **
	Sig. (2-tailed)							0.346	0.000	0.000	0.000	0.000	0.000	0.000	0.000
Assets Turnover	Pearson r							1	0.202 **	0.430 **	0.273 **	0.019	0.151 **	−0.172	0.034
	Sig. (2-tailed)								0.000	0.000	0.000	0.408	0.000	0.000	0.140
ROE	Pearson r								1	0.041	0.004	0.089 **	0.439*	0.071 **	0.744 **
	Sig. (2-tailed)									0.079	0.857	0.000	0.000	0.002	0.000
Current Assets to Total Assets Ratio	Pearson r									1	0.385 **	0.280 **	−0.052*	−0.272	0.014
	Sig. (2-tailed)										0.000	0.000	0.025	0.000	0.557
Current Debt Ratio	Pearson r										1	−0.385 **	−0.334 **	0.200 **	−0.163 **
	Sig. (2-tailed)											0.000	0.000	0.000	0.000
Quick Ratio	Pearson r											1	0.351 **	−0.273 **	0.221 **
	Sig. (2-tailed)												0.000	0.000	0.000
EAT and Depreciation to Total Assets	Pearson r												1	−0.303 **	0.587 **
	Sig. (2-tailed)													0.000	0.000
Financial Leverage	Pearson r													1	−0.130 **
	Sig. (2-tailed)														0.000
Profit Margin	Pearson r														1
	Sig. (2-tailed)

The correlation is statistically significant at the 0.01 level (**) (two-tailed). The correlation is statistically significant at the 0.05 level (*) (two-tailed).

(see Appendix A) demonstrates that the strongest correlations were observed between ROA (calculated from EAT) and ROE (r = 0.826), ROA calculated from EAT and EBIT (r = 0.972), profit margin and both measures of ROA (r = 0.829 and r = 0.786), as well as ROE (r = 0.744), and finally, between the ratio of working capital to total assets and current liquidity (r = 0.844). Most of the calculated correlation coefficients are statistically significant at the 0.01 or 0.05 alpha levels. Only six correlations were found to be statistically insignificant, all of which exhibit weak correlation strength.

Based on the results of the correlation analysis, four variables were excluded from further modeling: profit margin, return on total assets based on EBIT (ROTA), return on equity (ROE), and the ratio of working capital to total assets.

Table A2. Correlation coefficients after eliminating variables with high correlations.

Variable		Debt Ratio	Current Ratio	ROA	Equity Ratio	Assets Turnover	Current Assets to Total Assets Ratio	Current Debt Ratio	Quick Ratio	EAT and Depreciation to Total Assets	Financial Leverage
Debt Ratio	Pearsonovo r	1	−0.406 **	−0.106 **	−0.697 **	0.173 **	0.158 **	0.604 **	−0.346 **	−0.450 **	0.216 **
	Sig. (2-tailed)		0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000
Current Ratio	Pearsonovo r		1	0.166 **	0.608 **	0.055*	0.422 **	−0.537 **	0.648 **	0.281 **	−0.354 **
	Sig. (2-tailed)			0.000	0.000	0.018	0.000	0.000	0.000	0.000	0.000
ROA	Pearsonovo r			1	0.153 **	0.269 **	0.159 **	−0.089 **	0.233 **	0.660 **	−0.207 **
	Sig. (2-tailed)				0.000	0.000	0.000	0.000	0.000	0.000	0.000
Equity Ratio	Pearsonovo r				1	−0.022 **	0.090 **	−0.509 **	0.460 **	0.541 **	−0.515 **
	Sig.(2-tailed)					0.000	0.000	0.000	0.000	0.000	0.000
Assets Turnover	Pearsonovo r					1	0.430 **	0.273 **	0.019	0.151 **	−0.172 **
	Sig. (2-tailed)						0.000	0.000	0.408	0.000	0.000
Current Assets to Total Assets Ratio	Pearsonovo r						1	0.385 **	0.280 *	−0.052 *	−0.272 **
	Sig. (2-tailed)							0.000	0.025	0.000	0.000
Current Debt Ratio	Pearsonovo r							1	−0.385 **	−0.334 **	0.200 **
	Sig. (2-tailed)								0.000	0.000	0.000
Quick Ratio	Pearsonovo r								1	0.351 **	−0.273 **
	Sig. (2-tailed)									0.000	0.000
EAT and Depreciation to Total Assets	Pearsonovo r									1	−0.303 **
	Sig. (2-tailed)										0.000
Financial Leverage	Pearsonovo r										1
	Sig. (2-tailed)

The correlation is statistically significant at the 0.01 level (**) (two-tailed). The correlation is statistically significant at the 0.05 level (*) (two-tailed).

(see Appendix A) presents the revised correlation matrix after removing these variables. All remaining correlation coefficients fall below the threshold value of 0.7, thereby meeting the defined multicollinearity criteria. Only two correlations—between asset turnover and both equity ratio and current ratio—are statistically insignificant; however, both display a low level of correlation.

In general, acceptable values of the Variance Inflation Factor (VIF) are around one, while values below five typically indicate low multicollinearity. VIF values exceeding 10 suggest strong multicollinearity, whereas values between 5 and 10 are indicative of moderate multicollinearity Daoud (2017) [36], Alin (2010) [37], and Yoo et al. (2014) [38]. In this study, a threshold value of five was adopted. Based on this criterion, the variable current ratio was excluded due to its VIF exceeding the cutoff. All remaining variables exhibited VIF values below five (see

Table 6. Multicollinearity analysis using VIF.

Variable	VIF
Debt Ratio	2.596
ROA	2.279
Equity Ratio	3.935
Asset Turnover	1.403
Current Assets to Total Assets Ratio	4.288
Current Debt Ratio	4.738
Quick Ratio	1.858
EAT and Depreciation to Total Liabilities	3.232
Financial Leverage	1.618
Current Ratio	5.223

).

Table A3. Multicollinearity analysis using Condition Index.

Dimension	Eigenvalue	Variance Proportions
		Condition Index	Debt Ratio	ROA	Equity Ratio	Assets Turnover	Current Assets to Total Assets Ratio	Current Debt Ratio	Quick Ratio	EAT and Depreciation to Total Liabilities	Financial Leverage	Current Ratio
1	8.161	1.000	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00
2	2.662	1.753	0.01	0.26	0.00	0.00	0.03	0.00	0.02	0.00	0.01	0.00
3	1.566	2.293	0.00	0.25	0.00	0.03	0.00	0.03	0.00	0.00	0.01	0.01
4	1.434	2.378	0.00	0.00	0.05	0.00	0.00	0.00	0.03	0.00	0.00	0.00
5	0.733	3.321	0.00	0.04	0.67	0.00	0.05	0.01	0.00	0.00	0.00	0.00
6	0.605	3.677	0.04	0.00	0.00	0.05	0.00	0.00	0.03	0.01	0.00	0.00
7	0.334	4.942	0.00	0.00	0.00	0.67	0.01	0.00	0.00	0.00	0.00	0.00
8	0.163	7.048	0.00	0.00	0.00	0.00	0.00	0.05	0.00	0.00	0.00	0.00
9	0.069	10.966	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00
10	0.065	11.206	0.42	0.00	0.00	0.00	0.46	0.04	0.16	0.39	0.27	0.01
11	0.018	21.544	0.53	0.00	0.03	0.00	0.65	0.86	0.76	0.59	0.70	0.95

(see Appendix B) shows that all values above 0.5 are considered unacceptable. For multicollinearity to be confirmed within a single dimension, all three variables must exceed this threshold simultaneously. As shown in Table A2, signs of increased multicollinearity are observed only in Dimension 11, where both the current debt ratio and the ratio of current assets to total assets display substantial variance proportions within the same dimension. However, the condition index for Dimension 11 remains below the critical value of 30, indicating that multicollinearity is not present between these variables. Finally, we excluded the variable representing the ratio of EAT and depreciation to total liabilities, which is a modified version of the operating cash flow to total liabilities indicator.

Prediction model. The Exp(B) value for this variable was 8.8, suggesting that a one-unit increase in this ratio would result in an 8.8-fold increase in the probability of bankruptcy. From a practical standpoint, this interpretation is counterintuitive, as an increase in cash flow should, in principle, contribute to a company’s financial health. This discrepancy may stem from several factors. One explanation is that some firms report artificially high depreciation expenses to reduce taxable income by lowering net profit. If such companies go bankrupt, the model may then associate this variable with a higher risk of failure due to the observed statistical correlation.

Table 7. Prediction model.

Variables	B	S.E.	Wald	df	Sig.	Exp(B)	95% C.I. for EXP(B)
							Lower	Upper
Debt Ratio	4.226	0.410	106.306	1	0.000	68.462	30.658	152.882
ROA	−4.180	2.001	4.365	1	0.037	0.015	0.000	0.772
Current Assets to Total Assets Ratio	−6.414	0.548	136.891	1	0.000	0.002	0.001	0.005
Current Debt Ratio	4.499	0.622	52.251	1	0.000	89.939	26.555	304.614
Current Ratio	−2.088	0.257	65.785	1	0.000	0.124	0.075	0.205
Constant	−0.462	0.309	2.231	1	0.135	0.630

shows that the final model consists of five predictors, all of which are statistically significant at the 0.05 significance level. Two variables fall into the category of indebtedness, and one variable is drawn from each of the following categories such as profitability, liquidity, and asset structure. Moreover, Table 7 also includes the results of the Wald test, among other diagnostic statistics. Independent variables were selected for inclusion in the model using the backward stepwise method of binary logistic regression. This approach starts with a full set of potential predictors identified through a comprehensive literature review and iteratively removes variables that do not contribute significantly to the model, ensuring that the final model includes only statistically meaningful predictors. Appendix C shows VIF for this model.

The Wald test provides a straightforward method for evaluating the contribution of individual predictors; however, its application in small samples remains a subject of debate. As noted by Bewick et al. (2005) [39], the greater the Wald statistic, the more significant the predictor’s role in forecasting the target variable. In the presented model, the most influential variables, based on Wald statistics exceeding 100, are the debt ratio and the ratio of current assets to total assets. These indicators, both classified under the domain of indebtedness, contribute most significantly to the prediction of bankruptcy in Slovak industrial firms. The third most important variable is current liquidity, with a Wald value of 65.8, followed by the short-term debt ratio at 52.3.

In terms of odds ratios (Exp(B)), the results indicate that, ceteris paribus, a one-unit increase in the debt ratio multiplies the odds of bankruptcy by a factor of 68. Similarly, a one-unit increase in the short-term debt ratio raises the odds of bankruptcy nearly 90-fold. However, this interpretation may be misleading due to the small range of observed values for these variables, which inflates the odds ratios. A more realistic interpretation suggests that a 10-percentage-point increase in the short-term debt ratio would increase the likelihood of bankruptcy by approximately 1.55 times.

For current liquidity, the odds ratio is less than one, indicating a protective effect. Specifically, a one-unit increase in current liquidity, holding all other variables constant, reduces the probability of bankruptcy by approximately 87.6%.

The remaining two variables—Return on Assets (ROA) and the current-to-total assets ratio—exhibit similar patterns to the debt indicators. Although a one-unit increase in these variables appears to eliminate the probability of bankruptcy due to the narrow data range, a more accurate interpretation is again percentage-based. An increase of 10 percentage points in ROA would reduce the probability of bankruptcy by 34.3%, while a similar increase in the current-to-total assets ratio would lower it by 46.3%.

Table 8. Coefficients of Determination.

Initial −2 Log Likelihood	Final −2 Log Likelihood	Cox & Snell R Square	Nagelkerke R Square
1692.604	838.752	0.480	0.661

presents the output of two coefficients of determination: the Cox & Snell R² and the Nagelkerke R². The Nagelkerke coefficient is an adjusted version of the Cox & Snell R², designed to scale the value to a maximum of 1. While the Cox & Snell R² is constrained to values below 1, the Nagelkerke R² ranges from 0 to 1, making it the more commonly preferred metric for interpreting model fit. These coefficients indicate the extent to which the explanatory variables contribute to the predictive accuracy of the model (Bewick et al., 2005) [39].

Nagelkerke R² reaches a value of 0.661, suggesting that the model improves predictive performance by 66.1% compared to the null model (i.e., a model that includes only the intercept). In other words, the model accounts for 66.1% of the variance in the dependent variable, namely, the financial condition of the enterprise. This value indicates that the independent variables play a significant role in predicting the phenomenon under study.

Furthermore, the difference between the initial value of the −2 log-likelihood (in the model with only the constant) and the final value after including the predictors is substantial. Specifically, the −2 log-likelihood statistic decreased by 50.5%, demonstrating that the included variables strongly contribute to explaining the probability of corporate bankruptcy.

After constructing the model, we assess the fourth assumption of binary logistic regression (see the relationship between the explanatory variables and the model’s logit values). This is examined through graphical analysis. Figure 3 illustrates the relationships for all five statistically significant variables. All variables exhibit at least a moderate level of association with the logit (ranging from 0.3 to 0.7). The weakest association is observed for return on assets (0.318), while the strongest is found for current liquidity (0.837).

Table 9. Classification table for the training subsample (threshold: 0.50).

Threshold = 0.5		Predicted		Accuracy (%)
		Yes (1)	No (0)
Observed	Yes (1)	154	55	73.68%
	No (0)	36	316	89.77%
		Total Accuracy		83.78%

shows that the model achieved a Type I error rate of 10.23% based on the test sample. The Type II error rate was 26.32%. From the confusion matrix, we also calculated the overall accuracy (83.78%).

A threshold of 0.4 provides a more balanced and effective classification performance, representing a better compromise between sensitivity and specificity than the 0.5 threshold (see

Table 10. Threshold testing.

Probability Threshold	Youden’s Index (J)	F1 Score	Overall Accuracy
0.1	0.551	0.724	0.733
0.2	0.645	0.772	0.800
0.3	0.685	0.798	0.838
0.4	0.682	0.800	0.847
0.5	0.635	0.772	0.838
0.6	0.610	0.757	0.840
0.7	0.581	0.736	0.836
0.8	0.471	0.644	0.799
0.9	0.318	0.484	0.745

). Table 10 shows that the model achieves its highest predictive accuracy according to Youden’s Index at a probability threshold of 0.3. In contrast, based on the F1-score and overall accuracy, the model performs best at a threshold of 0.4.

Table 11. Classification table for the training subsample (threshold: 0.40).

Threshold = 0.4		Predicted		Accuracy (%)
		Yes (1)	No (0)
Observed	Yes (1)	157	52	75.12%
	No (0)	33	319	90.63%
		Total accuracy		84.75%

shows the classification results of the model on the test subsample using a threshold of 0.4. The model correctly classified 157 positive and 319 negative cases, while 52 positives and 33 negatives were misclassified. The accuracy for positive cases (Yes) is 75.12%, and for negative cases (No) it is 90.63%. The total accuracy of 84.75% indicates solid overall performance.

Nevertheless, we prefer using a threshold of 0.5, as it assumes equal importance in both classes and helps maintain a balanced trade-off between sensitivity (true positive rate) and specificity (true negative rate). Additionally, using 0.5 ensures comparability with other studies and standard benchmarks, simplifies interpretation, and provides a neutral starting point when no strong class imbalance or preference is assumed.

Table 12. Classification table for the test subsample.

Threshold = 0.5		Predicted		Accuracy (%)
		Yes (1)	No (0)
Observed	Yes (1)	369	90	80.40%
	No (0)	76	770	91.00%
		Total accuracy		87.30%

shows that the predictive model achieves an overall accuracy approximately four percentage points higher on the training subsample compared to the test sample. The Type I error rate is 9%. The Type II error is lower on the training data than on the test data.

The ROC curves of the presented model, generated from both the testing and training subsets, are illustrated in Figure 4. On the testing subset, the model achieves an AUC of 91.7%. The performance is slightly higher on the training subset, where the predictive accuracy reaches 93%.

5. Discussion

A major advantage of the developed predictive model lies in its narrow focus on a specific industrial sector and country of operation, which enhances its accuracy within the targeted domain. Due to this specialization, the model more accurately reflects characteristics typical of industrial enterprises operating in Slovakia. As a result, its predictive precision and practical applicability for businesses within this sector are significantly improved.

On the other hand, a key limitation of the model is the absence of qualitative variables such as firm size, region of operation, or legal form. These factors, along with numerous others, are known to have a substantial impact on the likelihood of business failure. However, they were not included in the present model due to the necessity of excluding many observations to maintain a sufficiently robust sample size, or due to the unavailability of the required data for constructing such qualitative predictors. Incorporating these complementary characteristics in future research would allow for a more comprehensive understanding of business risk and further increase the explanatory power of the model.

Summary. One of the key strengths of the proposed model lies in its high predictive performance across multiple evaluation metrics. The model achieves an overall accuracy of 83.78%, reflecting its strong classification capability. When evaluated using metrics that better account for class imbalance, such as the Area Under the Curve (AUC) and the F1-score, the model also delivers good to excellent results. On the test dataset, it reaches an AUC of 91.7% and an F1-score of 80% at a probability threshold of 0.4.

Table 13. Model performance comparison.

Authors	Country	Industry	AUC (%)	Overall Accuracy (%)	Common Variables
Mihalovič (2016) [21]	SK	Various	77.20	93.90	ROA, Current Debt Ratio, Current Assets to Total Assets Ratio
Adamko, Klieštik, and Kováčová (2018) [16]	SK	Various	73.73	88.85	Debt Ratio
Jenčová, Štefko, and Vašaničová (2020) [25]	SK	Electrotechnical and Mechanical Industry	92.02	94.00	Debt ratio, Current Ratio, ROA
Svabova et al. (2020) [6]	SK	Various	95.35	93.80	-
Durica, Valaskova, and Janoskova (2019) [9]	SK	Various	93.40	88.10	-
Pavličko and Mazanec (2022) [13]	V4	Various	95.07	93.91	Debt Ratio, ROA
Horváthová, Mokrišová, and Petruška (2021) [29]	SK	Heating Industry	n/a	84.00	-
Durica and Adamko (2016) [7]	SK	Industrial Sector	n/a	82.20	-
New model	SK	Industrial Sector	91.70	83.78	Debt Ratio, ROA

presents a comparison of prediction models developed for the V4 countries and Slovakia. Some models target a mix of industries, while others are tailored to specific sectors. The performance of the proposed model is comparable to that of other models within the region. In terms of AUC, the presented model outperforms the model by Mihalovič (2016) [21] and yields results similar to those of Adamko, Klieštik, and Kováčová (2018) [16]. For the F1-score, the model performs comparably to those developed by Mihalovič (2016) [21], Durica and Adamko (2016) [7], and Horváthová, Mokrišová, and Petruška (2021) [29]. The highest number of shared predictors with the presented logistic regression model is observed in the studies by Mihalovič (2016) [21] and Durica, Valaskova and Janoskova (2019) [9], each incorporating three identical variables.

Figure 5 presents a comparison of several models that we were able to evaluate using an identical testing sample. We compared their ROC curves along with the corresponding AUC values. The model developed by Grünberg and Lukason (2014) [30], which was created for manufacturing companies operating in Estonia, achieved an AUC of 59.4%. The model by Pavličko and Mazanec (2022) [13], designed for various industries across the V4 countries, performed quite well on our testing sample, reaching an AUC of 77.8%. The most accurate bankruptcy prediction was achieved by the model from Mihalovič (2016) [21], which focuses on Slovak companies, with an AUC of 84.7%.

As part of this comparison, the constructed model achieves the best results, which confirms that a specific focus on Slovak industrial enterprises proves beneficial in developing the prediction model. However, it is essential to note that the AUC values of the other models remain high, considering they are designed for use across multiple industries or different countries. The developed model adapts well to specific conditions and shows higher predictive performance. This result indicates that specialization can be an effective approach to enhancing the practical usability of a model.

Limitations. Despite these promising outcomes, certain limitations must be acknowledged. First, the dataset is limited to 3878 Slovak industrial manufacturing firms, which may restrict the generalizability of results to other industries or countries. Due to missing or erroneous financial indicators, a considerable number of companies had to be excluded from the original dataset. As a result, the model was developed solely using quantitative variables and trained on a sample of 1305 firms. Second, the industry focus, while justified, narrows the scope of applicability. These limitations provide opportunities for future research to extend the analysis across larger datasets, multiple sectors, and international contexts.

Future research. Future research should aim to increase the sample size and consider the integration of non-financial indicators, which may provide additional explanatory power in predicting corporate failure. While our current model focuses on quantitative financial indicators, we recognize that qualitative variables—such as corporate governance, management experience, ownership structure, or market reputation—can provide additional insights into a firm’s risk profile. Integrating such qualitative factors in future research could potentially enhance predictive power, capture aspects of financial distress not reflected in numerical data and improve the model’s applicability for managerial decision-making and early-warning systems.

6. Conclusions

The primary contribution is a bankruptcy prediction model specifically designed for crisis contexts. Our model is thus designed to capture financial distress in exceptional contexts, offering insights that are highly relevant in times of crisis. A predictive model was successfully developed to estimate the probability of corporate bankruptcy one year in advance. The model is based on five predictors, with the most significant being the proportion of current assets to total assets, along with the levels of total and short-term debt. The model specifically targets industrial enterprises operating in Slovakia.

The validity and robustness of the model were confirmed through testing on an independent dataset not used during model training. Performance evaluation metrics demonstrate the model’s strong ability to distinguish between financially healthy and distressed firms. The model achieved an overall classification accuracy of 83.78%, meaning it correctly classified approximately four out of five observations. In terms of discriminatory power, the model yielded an Area Under the Curve (AUC) value of 91.7%, indicating excellent predictive capability.

These results suggest that the model holds substantial practical potential for various stakeholders, including businesses themselves, as well as creditors, investors, banks, and other relevant parties. We emphasise that the model can be applied in various ways in business practice.

▪

The company management can monitor financial health and identify early warning signs of financial distress, enabling timely corrective actions such as cost optimization or debt restructuring.

▪

The investors can evaluate the financial stability of potential investment targets and support risk-adjusted decision-making.

▪

The banks and creditors assess the creditworthiness of clients and predict default risk when evaluating loan applications.

▪

The auditors and consultants can incorporate the model into financial risk assessment tools for regular company evaluations.