Research Article: 2020 Vol: 24 Issue: 6
Yehui Tong, Nanjing University of Finance and Economics
Zélia Serrasqueiro, University of Beira Interior
Using the traditional logistic regression model, this paper studies the impacts of some financial variables and the relative variance variables on the prediction of failure and financial distress of Portuguese small and medium-sized enterprises in high and medium-high technology manufacturing sectors. The results show that adding variance variables (standard deviations as the representative of variable stability) to the original variables does increase the classification accuracy of the models. As for the detailed impacts of financial factors, profitability as a positive factor is the most important indicator for both failure and financial distress, which is followed by debt structure and liquidity; on the other hand, different to leverage and firm size, intangible assets are more important in the distress prediction model compared to the failure prediction model. The differences of the statistically significant variables in the failure prediction model and the financial distress prediction model verify that it is necessary to separate failure from financial distress when doing predictions.
Financial Factors and Stability, Predictions of Failure and Financial Distress, Small and Medium-Sized Enterprises, High and Medium-High Technology Manufacturing Sectors.
This Since the recent world financial crisis, the unstable business environment has made the research on business failure and bankruptcy prediction more important for investors and creditors (Pervan & Kuvek, 2013; Alaminos et al., 2016). Business failure not only could cause large economic and social losses for stakeholders, but also may lead to severe domestic crisis because of inefficient allocation of domestic capital (Laitinen & Suvas, 2013); on the other hand, financial distress diagnosis and prediction can significantly influence the operation and related parties (such as, credit institutions and stockholders) of a firm and even the whole economy of a country at large (Doumpos & Zopounidis, 1999). In concrete, accurate business failure prediction models can increase people’s confidence in investment, lending and developing business relationships, and will promote the stability of economic growth (Gepp & Kumar, 2008).
As pointed out by Gepp & Kumar (2008), business failure prediction is the process to develop models in order to predict the financial failure of a business before it actually happens; and, because of its usefulness and value of in the real world, business failure prediction is widely studied from both industry and academia. However, although the research topics of financial distress and bankruptcy have been studied for several decades since mid-1960s, new challenges have constantly appeared regarding the factors and impacts on the success or failure of firms (Maripuu & Männasoo, 2014); and from the perspective of the literature regarding financial ratios, there is no consensus about the best accounting ratios to predict the likelihood of financial distress (Mossman et al., 1998).
Being unable to meet obligations and the value of a company’s liabilities beyond the value of its assets are two obvious hallmarks of financial distress, and the purpose of financial distress research is to identify the useful accounting information in predicting future financial distress (Omelka et al., 2013; Ward & Foster, 1997). Here, the difference between business failure and financial distress should be stressed: that is, financial distress means financial problems which not necessarily result in bankruptcy (Achim et al., 2016; Pozzoli & Paolone, 2016); by contrast, although there is no unique definition of failure (Fernández-Gámez et al., 2016), discontinuity of operation is a mutual feature of different definitions of failure (Dimitras et al., 1996).
As is shown in the study of Gupta et al. (2018), there exist differences in influential factors in predicting bankruptcy and financial distress. Besides, compared to large listed firms, small firms are less researched in the literature (Pompe & Bilderbeek, 2005). It is also necessary to develop country-specific prediction models and apply models in different economic sectors in order to reflect different country’s economic and business status (Šlefendorfas, 2016; Kanapickiene & Marcinkevicius, 2014). Therefore, this paper investigates the failure and financial distress of Portuguese small and medium-sized enterprises (SMEs) in high and medium-high technology manufacturing sectors.
As stated by Pacheco (2015), in 2012, 99.9% of Portuguese enterprises were SMEs, which contributed to 78% of the employees in the private sectors and 58% of the total turnover; according to the Caixa Bank Research reported by Pinheiro (2019) based on the data from Eurostat and the Bank of Portugal, in manufacturing industry, the firms in high and medium-high technology sectors contributed to about one fourth of the total sales in 2016. On the other hand, in Portugal SMEs have higher failure rate than large firms, and the failure rates of the Portuguese firms in high and medium-high technology sectors are higher than the failure rate in medium-low technology sectors (Succurro & Mannarino, 2014). So the first contribution of this paper is to help Portuguese high and medium-high technology SMEs to find significant financial indicators for predicting failure and financial distress.
The research of Dambolena & Khoury (1980) shows that: the stability of ratios can help to improve the ability to predict failure. Thus, in addition to the original financial ratios, this paper also takes the stability of ratios into account, which leads to the second contribution (that is, proffering empirical evidence to the usefulness of the stability of ratios in the prediction models). In particular, logistic regression analysis is used in the prediction models for the data of one, two and three years prior to the event; this three-year prediction method was used in the studies of (for instance) Mossman et al. (1998), Fernández-Gámez et al. (2016) and Alaminos et al. (2016). We also refer to the research method of Pompe & Bilderbeek (2005) in which the impacts of the stability of ratios (for example the standard deviation of three years) in three successive annual reports are explored in the prediction models. The followings of this paper are arranged in this order: literature review; data, variables, and research methodology; results and discussion; and conclusions.
The research on business failure can be traced back to 1930s (Pervan & Kuvek, 2013); however, it is since 1960s that statistical and mathematical models have been built for business failure prediction (Gepp & Kumar, 2008). As stated by Mossman et al. (1998), the models of using financial ratios to predict bankruptcy are firstly developed by Beaver (1968) and Altman (1968). Haber (2005) points out the difference between the research of Beaver (1966) and the research of Altman’s (1968): that is, the previous one focuses on financial distress and insolvency and latter one pays attention to bankruptcy; and it is the application of sophisticated statistical technique (namely multiple discriminant analysis) that makes Altman’s (1968) research as one milestone of the bankruptcy study, which is different to the univariate technique introduced by Beaver (1966) for classifying firms in two groups by using financial ratios (Dimitras et al., 1996).
On the basis of companies’ financial characteristics (financial ratios), multivariate discriminant analysis calculates the discriminant score to classify companies into healthy and bankrupt categories (Fejér-Király, 2015). However, multiple discriminate analysis requires for normality of predictors and the same variance-covariance matrices for both groups (Pervan & Kuvek, 2013). In order to overcome the limitations of the linear discriminant analysis approach, Ohlson (1980) begun to use logistic regression in the prediction of failure (Charitou et al., 2004); after that, data mining techniques (such as, neural networks, case-based reasoning, and decision trees) are applied in bankruptcy prediction models (Mihalovič, 2016).
Logit model is one of the most commonly used methods in bankruptcy prediction (Bauweraerts, 2016). Arnis et al. (2018, p.118) state that, “The Logit model is a nonlinear regression model specifically designed to assess dependent binary variables. It gives the probability that the dependent variable will get the value 1, given the values of the independent variables, by adopting techniques that lead the values being assessed to move in the range (0,1).” It is further explained by Kanapickiene & Marcinkevicius (2014) that: “In logistic regression models the bankruptcy probability is calculated by the following formula: P(Z)=1/(1+e-Z), where P is bankruptcy probability (from 0 to 1), and Z is Z value of the analyzed model. When P > 50%, there is a bankruptcy probability; when P ≤ 50%, there isn’t any bankruptcy threat to a company.” Logit analysis does not require to fulfill the requirements of linear discriminant analysis, such as, the multivariate normal distribution of the variables and the equivalence of the variance and covariance matrices of the variables for the non-failed and failed firms; notwithstanding that, logit models still have some limitations including multicollinearity problem and the problems of outliers and missing values (Giacosa et al., 2016).
The Portuguese small and medium-sized enterprises (SMEs) in the high technology and medium-high technology manufacturing sectors are chosen from the Iberian Balance Sheet Analysis System (SABI; developed by Bureau Van Dijk) database for building the sample. According to the criteria of European Union, here SME is defined as: number of employees less than 250; and turnover less than or equal to 50 million Euros or balance sheet total less than or equal to 43 million Euros. Based on the classification of NACE Rev. 2 2-digit level (from Eurostat), high technology manufacturing sectors include manufacture of basic pharmaceutical products and pharmaceutical preparations and manufacture of computer, electronic and optical products while medium-high technology manufacturing sectors contain manufacture of chemicals and chemical products, manufacture of electrical equipment, manufacture of machinery and equipment n.e.c., manufacture of motor vehicles, trailers and semi-trailers, and manufacture of other transport equipment.
It is required that all the candidate SMEs must report operating revenues in 2013, 2014, and 2015 to SABI database (in the observed five years from 2013 to 2017); and the data in 2016 and 2017 are used to identify the failed firms, financially distressed firms, and financially healthy firms. In concrete, the identifying method of Quintiliani (2017) is referred to for differentiating the financially distressed firms to the financially healthy firms: that is, “we consider as financial distress companies those that meet some of the following conditions: (i) its earnings before interest and taxes depreciation and amortization (EBITDA) are lower than its financial expenses for two consecutive years; and/or (ii) increase in the debt-to-net worth formula for two consecutive periods with concomitant decrease of the denominator.” Following the above criteria, here the financially healthy firms are the “active” firms with higher EBITDA (compared to financial expenses), decrease in debt ratio, and increase in net worth. As for the failed firms, we follow both the identifying methods of Pacheco (2015) and Mata and Portugal (1994): failed firms are the firms that are not labeled as “active” firms in SABI database (discriminating from the firms labeled as “active”) and do not report operating revenues in 2016 and 2017 two consecutive years.
In order to find the significant financial factors separately in the failure and financial distress prediction models, we classify the total sample into two groups: the failure group (for comparing the financially healthy firms with the failed firms) and the financial distress group (for comparing the financially healthy firms with the financially distressed firms). Binary logit regression model is employed here, as it is suitable for the dichotomous dependent variable and the explanatory variables can be quantitative or qualitative (Brédart, 2014). In particular, referring to the research method of Pompe & Bilderbeek (2005) where the independent variables are identified into four groups (the ratios in year 1; the ratios and standard deviations in year 1; the ratios in year 3; and the ratios and standard deviations in year 3), we also run regression four times respectively with only the 2015 original data, only the 2013 original data, the 2015 original data together with the variance variables, and the 2013 original data together with the variance variables. The variance variables are the standard deviations of the 2013, 2014 and 2015 original data.
Since logit model is employed here, it is necessary to avoid multi-linearity problem when choosing independent variables. Financial ratios, however, usually are internally related. So, instead of grabbing a bunch of financial ratios to describe one category of financial characteristic, we choose the most commonly used financial ratios (or variables) to represent financial characteristics, which would reduce the number of independent variables. As pointed out by Blanco-Oliver et al. (2015), traditionally, leverage and debt related ratios are strong predictors related to bankruptcy and financial risk, and heavy liabilities may cause financial problems; in addition, profitability ratios represent the ability of firms to accumulate reserves and are widely used in the prediction of bankruptcy. Low liquidity and being difficult to meet the commitments are the common features of distressed firms, so liquidity-related variables are also necessary for measuring the capacity of a firm to pay its debts and to continue its activity (Brédart, 2014). Thus, indebtedness, the ratio of current liabilities to total liabilities, return on assets (ROA), and general liquidity are used in this paper for respectively representing leverage, debt structure, profitability, and liquidity.
Because we focus on the SMEs in high and medium-high technology sectors, it is necessary to consider the influence of intangible assets which play an important role (Elston & Audretsch, 2011); considering that many firms in the sample do not report intangible assets, a dummy variable is created. Firm size (total assets) and assets structure (tangible fixed assets) are also included in this study. The definitions of the variables and the statistics of the sample are shown in Table 1, 2 and 3. On average, compared to the financially distressed firms, the financially healthy firms show higher total assets, higher ROA, higher proportion of tangible fixed assets, higher proportion of firms with intangible assets, and lower liquidity. In the failure group, generally it shows similar situation, aside from financially healthy firms showing obviously lower leverage and higher proportion of current liabilities.
|Table 1 Variable Definitions|
|Dependent variable (1): failure or financial health||Failed firms are the “inactive” firms that report operating revenues in 2013, 2014, and 2015 but not report operating revenues in 2016 and 2017.
Financially healthy firms are the “active” firms with higher EBITDA (compared to financial expenses), continuous decrease in debt ratio, and continuous increase in net worth from 2013 to 2017.
Note: the failed firms only report operating revenues in 2013, 2014, and 2015, while the financially healthy firms report operating revenues in all the five years from 2013 to 2017.
|Dependent variable (2): financial distress or financial health||Financially distressed firms are those that meet some of the following conditions: (i) its earnings before interest and taxes depreciation and amortization (EBITDA) are lower than its financial expenses in both 2016 and 2017; and/or (ii) increase in the debt-to-net worth formula from 2016 to 2017 with concomitant decrease of the denominator; and these conditions do not appear in 2013, 2014, and 2015 (based on the classifying method of Quintiliani (2017, pp. 71-72)).
Financially healthy firms are the “active” firms with higher EBITDA (compared to financial expenses), continuous decrease in debt ratio, and continuous increase in net worth from 2013 to 2017.
Note: both the financially distressed firms and the financially healthy firms must report operating revenues in all the five years from 2013 to 2017.
|Firm size||Natural logarithm of total assets (in thousands of Euros); Ln assets|
|Profitability||ROA (return on assets): Profits before tax/Total assets|
|Liquidity||General liquidity (current ratio): the ratio of current assets to current liabilities|
|Solvency (leverage)||Indebtedness: (Total shareholders funds and liabilities?Shareholders equity)/Total shareholders funds and liabilities|
|Intangibles||Dummy variable of intangible assets (if firm’s intangible assets are more than zero, it equals 1; if not, it equals 0)|
|Tangibles (assets structure)||The ratio of tangible fixed assets to total assets|
|Liability structure||The ratio of current liabilities to total liabilities|
|Table 2 The Statistics of the Sample Data in the Financial Distress Group (250 Cases)|
|Variables||Mean||Standard Deviation||Min||Max||The mean of financially distressed firms (60 cases)||The mean of financially healthy firms (190 cases)|
|Ln assets 2015||6.691||1.602||2.381||10.585||6.471||6.760|
|Ln assets 2013||6.568||1.619||2.226||10.777||6.370||6.631|
|Standard deviation of in assets||0.134||0.141||0.000||0.928||0.146||0.131|
|Standard deviation of ROA||0.054||0.071||0.000||0.603||0.073||0.047|
|Standard deviation of indebtedness||0.088||0.181||0.000||2.466||0.090||0.087|
|Standard deviation of tangibles||0.040||0.048||0.000||0.366||0.049||0.037|
|Dummy intangibles 2015||0.332||0.472||0.000||1.000||0.233||0.363|
|Dummy intangibles 2013||0.352||0.479||0.000||1.000||0.283||0.374|
|Current liabilities to total liabilities 2015||0.699||0.272||0.054||1.000||0.692||0.701|
|Current liabilities to total liabilities 2013||0.724||0.270||0.005||1.000||0.775||0.709|
|Standard deviation of current liabilities to total liabilities||0.082||0.100||0.000||0.474||0.114||0.072|
|General liquidity 2015||3.018||5.470||0.348||72.558||4.047||2.693|
|General liquidity 2013||2.359||3.843||0.106||39.628||3.181||2.100|
|Standard deviation of general liquidity||0.946||3.581||0.000||46.861||2.134||0.5707|
|Table 3 The Statistics of The Sample Data in the Failure Group (236 CASES)|
|Variables||Mean||Standard Deviation||Min||Max||The mean of Failed firms (46 cases)||The mean of financially healthy firms (190 cases)|
|Ln assets 2015||6.400||1.764||0.798||10.585||4.914||6.760|
|Ln assets 2013||6.354||1.729||-0.564||10.777||5.212||6.631|
|Standard deviation of ln assets||0.179||0.250||0.000||1.864||0.379||0.131|
|Standard deviation of ROA||0.192||1.141||0.000||12.904||0.791||0.047|
|Standard deviation of indebtedness||0.328||1.330||0.000||12.883||1.322||0.087|
|Standard deviation of tangibles||0.037||0.050||0.000||0.447||0.039||0.037|
|Dummy intangibles 2015||0.318||0.467||0.000||1.000||0.130||0.363|
|Dummy intangibles 2013||0.347||0.477||0.000||1.000||0.239||0.374|
|Current liabilities to total liabilities 2015||0.696||0.285||0.016||1.000||0.678||0.701|
|Current liabilities to total liabilities 2013||0.701||0.285||0.007||1.000||0.670||0.709|
|Standard deviation of current liabilities to total liabilities||0.076||0.097||0.000||0.522||0.092||0.072|
|General liquidity 2015||3.067||4.875||0.049||52.772||4.611||2.693|
|General liquidity 2013||2.390||3.021||0.039||25.916||3.591||2.010|
|Standard deviation of general liquidity||0.898||2.420||0.000||27.734||2.252||0.571|
The failure group (including the financially healthy firms and the failed firms) and the financial distress group (including the financially healthy firms and the financially distressed firms) are separately put into the binary logit models. In concrete, each group has four regressions with different independent variables (the 2015 original variables, the 2013 original variables, the 2015 original variables together with the standard deviations of the 2013, 2014 and 2015 data, and the 2013 original variables together with the standard deviations of the 2013, 2014 and 2015 data). The detailed results are shown in the following Tables 4, 5, 6 and 7. Here we believe that one variable is statistically significant if its P-value is lower than 0.1.
|Table 4 The Results of Logistic Regressions of the Failure Group (Only Original Variables)|
|Independent variables||2015 data||2013 data|
|Number of observations: 236||Number of observations: 236|
|LR chi2(7): 146.96||LR chi2(7): 48.72|
|Prob > chi2: 0.0000||Prob > chi2: 0.0000|
|Log likelihood: -42.9299||Log likelihood: -92.0532|
|Pseudo R2: 0.6312||Pseudo R2: 0.2092|