Prediction of bankruptcy of companies in the Netherlands : main determinants

Prediction of Bankruptcy of Companies in the Netherlands:

Main Determinants

Master`s Thesis

University of Amsterdam,

Faculty of Economics and Business,

Masters Program in Business Economics: Organization Economics

Oleksandra Isachenko

July 2015

2

TABLE OF CONTENTS

SAMPLE SELECTION AND DATA SOURCES……….. 18

ESTIMATION RESULTS……… 22

CONCLUSIONS……….. 29

3

GLOSSARY

Name Description

1 2

AUC Area under curves

CBS Central bureau of statistics in the Netherlands

CIS Commonwealth of Independent States

CR Current ratio

EU European Union

MDA Multiple discriminant analysis

NAT Net asset turnover

PM Pre-tax profit margin

QR Quick ratio

ROA Return on assets

ROC Receiver operating characteristic

ROE Return on equity

SER Shareholder equity ratio

4

1. INTRODUCTION

Modern economic reality induces managers of enterprises to make decisions under uncertainty. Under conditions of financial and political instability the activity of a company is a subject of various critical situations that can result insolvency or bankruptcy of firm. Bankruptcy is a legal proceeding when a company is unable to repay its debts to creditors. Companies are never protected against failure and are likely to go bankrupt, either in an economic expansion or in a recession.

Methodological approaches concerning prediction of companies going bankrupt include some key indicators that characterize the financial position of a company. These approaches should allow to predict an appearance of crisis situation of a company in advance, before the occurrence of a certain visible signs.

The Netherlands is the sixth-largest economy in euro zone and is known for its stability, low unemployment rate and low inflation (Central bureau of statistics). The current economic situation of the Netherlands is experiencing economic recovery after the crisis of 2008. More and more economic indexes are shifted from recession stage to the recovery stage. The decreasing number of companies that failed in the recent years is also an evidence of economic recovery. Government of the Netherlands creates policies and conditions for the companies that can prevent bankruptcies as much as possible.

First of all, the Netherlands have favorable legal environment for jurisdiction to start a business. An extensive system of legislation, which in comparison with, for example, CIS countries law, has more flexibility. Participants of business are more protected against economic situation and against arbitrary action by public authorities.

Secondly, the Netherlands have a beneficial corporate tax system. According to government regulations the corporate income tax rate for the Netherlands in 2014 was equal to 20% for taxable amounts up to and including EUR 200,000. For taxable amounts which exceed EUR 200,000, the corporate income tax rate was 25%. France has a corporate tax rate of 33.33% on profits for large companies and a reduced tax of 15% on the first 38,120 Euros of profit for small companies. Germany has a corporate income tax rate of 29%, Spain of 30%, Italy of 31%, United States and Japan of 40%. Moreover, double taxation can usually be avoided in the Netherlands.

In addition, the Netherlands have a low interest rate on loans for organizations. According to data of The European Central Bank the lending interest rate for the Netherlands, on average, is 1.6 %. This is low compared with other countries of the EU: 1.9% for France,

5

2.4% for Austria and Germany, 3.4% for Belgium, 3.6% for Italy. And also low if to compare with other well developed countries: 3.07% for Japan, 5.13% for Australia, 6.04% for United States, 8.02% for United Kingdom.

These conditions can cause Dutch companies to have distinct results comparing to other countries and make it particularly interesting to research.

There are a lot of prior researches on investigating determinants that cause companies to go bankrupt in different countries. Among the primary determinants are size and a age of the company. The larger is the firm in terms of total assets the lower is the probability of failure (Beaver, 1966). Moreover, companies which exist for a small period of time have higher probability of going bankrupt (Altman, 1993). The explanation for this is that older firms are tend to be previously successful if they exist for many years and thus this success will help them to survive for a certain time (Levinthal, 1991). Moreover, young firms are tend to be smaller than older firms. One of every four newly created companies in the Netherlands, on average, does not survive for more than two years and round fifty percent does not exist for more than five years (Pellenbarg and Van Steen, 2003). On the basis of this we suggest:

Hypothesis 1: Smaller and younger Dutch companies have higher probability of going

bankrupt comparing to medium-sized and large established companies.

The probability of the companies going bankrupt also depends on their economic performance, the macroeconomic environment and financing conditions (Aleksanyan et al, 2014). Altman (1968) and Beaver (1966) were pioneers in making research on predicting corporate bankruptcy using financial ratios. Their analysis states that financial ratios are useful in predicting failure of medium-sized and large companies. Hossari and Rahman (2005) also have found financial ratios to be powerful in prediction models for assessing the financial distress of a firm. We suggest that for the Netherlands:

Hypothesis 2: Financial ratios are useful in predicting corporate failure of small,

medium-sized and large companies.

According to the findings of Bellovary et al (2007) return on assets ratio has the most predictive power among all financial ratios. Janer (2011) in his paper found asset turnover ratio to be an important predictor for French data, Fijorek et al (2012) argue that two most important financial factors that predict bankruptcy for Polish companies are operating profitability and cash turnover ratio. These financial ratios belong to different classification groups. According to study of Altman (1968) financial ratios can be classified into four categories as follows: liquidity ratios, profitability ratios, solvency ratios, efficiency ratios.

6

Solvency and profitability ratios are important in predicting failure of companies (Yazdanfar et al, 2008). Lennox (1999) explains that a company which is unprofitable, highly leveraged and has low cash flows is more likely to go bankrupt. Furthermore, according to Lizal (2002) liquidity ratios turned out to be the most irrelevant no matter the time-framed used. We use data of Dutch companies to support or disprove findings of current authors and we suggest that:

Hypothesis 3: Profitability and solvency ratios tend to have higher influence than

liquidity and efficiency ratios in predicting bankruptcy of firms.

There are various articles on the prediction of corporate failure, but only few are related to the Netherlands: Bilderbeek (1979)and Slotemaker (2008) where they prove that financial ratios are useful to predict default of Dutch companies.

The goal of this thesis is to create a probabilistic model that can predict bankruptcies of firms. This approach can help us in many scientific and practical ways. From the scientific point of view, finding determinants of firm’s bankruptcy can help us to explain each factor more extensively. For example, why the return on assets has more influence on probability of going bankrupt than return on equity or whether big firms are less influenced by comparable losses than small ones. From the practical point of view, it helps us to make investment choices, to evaluate the required rate on borrowings and decide even whether to lend money or not.

This study conducts an analysis of firm bankruptcy by accounting for firm financial factors. Our analysis concentrates on Dutch firms of different age and in financial and non-financial industries. An econometric analysis based on the data set of 251 companies that are observed for a period of 2008-2013. A long time period allows us to extract more information from the data. An expanded list of variables would lead to defining a predictive power each variable has on bankruptcy’s probability. The data is analyzed using random effect probit regression model. Advantage of using such a random effects model for prediction of corporate failure is to consider not only the company’s individual characteristics but also the uncertainty that cannot be explained by such characteristics. Jarrow (2004) and Hillegeist et al (2004) emphasize that a single-period approach neglects important information when a company is at risk, but remains solvent. Our model takes this into account and addresses the evolutionary nature of bankruptcy by considering the past values of the explanatory variables (lagged values up to three years before the failure).

The thesis continues with reviewing literature related to research question. The third section discusses more deeply the methodology used for analysis. In section four we more

7

closely discuss the sample selection and data sources. Section five describes the results of empirical tests. Conclusions and final remarks are presented in section six.

2. LITERATURE REVIEW

The question may arise why firms go bankrupt in general. The first possible reason is inappropriate allocation of resources in the economy. According to neoclassical economists

such as Marshall (1890) and others, firms react to changes in people preferences, e.g. consumers maximize their utilities and firms maximize their profits. Suppose demand for

some specific goods falls due to shifting to the other one. As example, people start drifting from fast food consumption towards health goods (gym and swimming pool). There is a leftward shift in demand for fast food. Eventually marginal firms leave the market until demand equates supply. The second reason for going bankruptcy is an industry structure. Maybe industry possesses ever increasing return to scale feature and according to Spence (1983) all small firms will be beaten out by a large one. Since industry with decreasing marginal costs (same as decreasing average costs) the firm ultimately must make a choice between merger and exit.

The topic of identifying determinants that predict failure of companies has been investigated by many researchers for quite a long period of time. According to Bellovary et al. (2007) since sixties there have been more than 170 articles on this topic around the world. Mostly research papers were written on the US economy, but we can observe a number of papers on other countries: Janer J. (2011) on France, Shirata Y. (2009) on Japan, Dakovic R. et al (2010) for Norway, Aliakbari S. (2009) on the United Kingdom. The most common approaches are multiple discriminant analysis, probit and logit analysis.

According to Altman (1968) multiple discriminant analysis is used to classify an observation into one of several groupings dependent upon the observation`s individual characteristics and to classify or make predictions. The goal of MDA is to obtain a model to predict a single dependent variable as a linear combination of independent variables. It works by creating a new variable that is called the discriminant function score and used to predict to which group an observation belongs. Discriminant function scores are computed using eigenvalues. The computations find the coefficients for the independent variables that maximize the measure of distance between the groups defined by the dependent variable.

8

The first to apply a multiple discriminant analysis to determine financial problems were Walter (1957) and Smith (1965). Walter employed a MDA analysis to classify high and low earnings price ratios of large industrial corporations. Smith used this model to classify firms into standard investment categories. In current practice, the most widely are used failure prediction models of Beaver (1966) and Altman (1968) that also apply multiple discriminant analysis.

The study of Beaver (1966) contains 79 failed firms in period of 1954-1964 years that were matched with non-failed firms in terms of total asset size and industry. He constructed the univariate approach of the multiple discriminant model with accuracy of 87%. Findings indicate that the best univariate discriminator between failed and active companies is cash flow to total debt. This univariate analysis is a benchmark in developing multivariate models.

Alternatively to the univariate analysis used by Beaver, Altman (1968) uses multivariate approach of multiple discriminant model in order to develop a model for bankruptcy prediction. He demonstrates the usefulness of MDA taking combinations of ratios that can be analyzed together. The created Z-score model is a five-factor model which was developed on the basis of bankrupt and non-bankrupt industrial enterprises of United States. Altman applies twenty two most commonly used variables, out of which he employs five most influential variables: working capital over total assets, retained earnings over total assets, earnings before interest and taxes over total assets, market value of equity over book value of total debt and sales over total assets. All variables are adjusted to the size of assets. That eliminates the size influence on these variables and leaves only relative indicators. The variables are found to have the diminishing effect on probability to go bankrupt. Classification accuracy of Altman`s model is 95% with data one year before bankruptcy and 87% with data of two years prior to failure.

Despite the fact that previously mentioned studies provide a high accuracy, MDA models are based on the assumptions that are sometimes violated in predicting failures. Karels and Prakash (1987) argue that the first requirement is that the explanatory variables have a multivariate normal distribution. The second requirement is that the sample of bankrupt and non-bankrupt is assumed to be drawn randomly from the population sample. If these assumptions are satisfied then the multiple discriminant analysis is optimal (Kshirsagar, 1971).

To avoid disadvantages of using MDA approach, logit and probit analysis can be

applied in predicting corporate failure. Ohlson (1980), Zmijewski (1984) and Gentry et al (1987) were the first to apply logit and probit models in predicting default risk.

9

Ohlson (1980) applied nine financial and non-financial ratios of 2058 non-bankrupt and 105 bankrupt companies for the period 1970-1976 years. Ohlson’s logit model is preferred to multiple discriminant analysis, because it allows to use dummy variables and provides us with the probability of going bankrupt as a result. One of the assumptions used in the multiple discriminant analysis model, is that independent variables have a normal distribution, while dummies variables do not have this distribution. Moreover, the multiple discriminant analysis does not deal with probability. Its coefficients have no sense regarding probability, they show the effect on a score variable instead. The results of the study suggest that the ratios representing current liquidity, performance, financial structure and the size of a company are associated with bankruptcy within one year. Classification accuracy of Ohlson`s model is 96.3%.

Zmijewski (1984) developed a new type of research using a probit model to analyze biases caused by sample selection. The data sample consisted of 81 failed and 1,600 active companies in United States. In this research he argues that estimating models on nonrandom samples can result in biased outcomes. He presented two estimation biases: one due to oversampling of distressed companies, and the other from using only complete data. The existence of biased was examined by comparing the unweighted probit results across the estimation samples.

Gentry et al (1987) used a probit model to prove that cash based funds flow factors and financial ratios provide significant information in classifying of failed and non-failed companies. Authors combined twelve funds flow components with nine financial ratios. The probit results show that dividends, investment and receivables are highly related to classification process. Moreover, results indicate that funds flow factors provide better results than financial ratios.

Lennox (1999) in his study uses the sample of 949 listed UK companies for a long period of time to compare logit and probit models with discriminant analysis. Results show that probit and logit models can identify failure of companies more accurately than discriminant analysis. Furthermore, the advantage over discriminant analysis was even greater for well-specified non-linear probit and logit models.

Researches nowadays also involve hazard models and neural networks. According to Shumway (2001) hazard models are preferable to static models that are single-period classification models, because they correct for period at risk and allow for time-varying covariates. When sampling periods are long, it is important to control for the fact that some firms file for bankruptcy after many years of being at risk while other firms fail in their first

10

year. Hazard models make this adjustment, while static models do not adjust for the period at risk. Chava et al (2004) expanded the database to approve superior forecasting performance of Shumway`s hazard model over previously used methods.

The neural network model is probably the most widely used model among the intelligent techniques. This model explores correlations between predictive variables that are then used as additional variables in a non-linear bankruptcy prediction model. Neural networks do not rely on specific assumptions concerning data or predictors. This makes them more reliable than models that would have their assumptions violated. Odom et al (1990) in their research have compared neural networks and multiple discriminant analysis in prediction of bankruptcy. The results show that neural networks perform much better. The same result was obtained by Wilson et al (1994) in accuracy of predicting bankrupt and non-bankrupt firms under various conditions on the basis of small number of financial ratios.

Classical economist’s point of view is that the supply curve is a marginal cost for a firm. And while turmoil is turning out with demand, less productive firms are bound to leave the market. Olley and Pakes (1996), for example, employ the productivity approach in their work. Less productive firms are supposed to have higher exposure to bankruptcy. Additionally, there are other risks that threaten firms’ conditions. Even in “capital structure indifferent universe” of Modigliani and Miller (1958), higher debt level pushes up required return on equity. So, high leverage makes firms more vulnerable. There is a wide range of financial ratios that are used in prediction of corporate failure, so it would be beneficial for us to look deeply into previous works.

Bellovary et al. (2007) collected all articles concerning prediction of failure and found the eight most frequently used determinants. These ratios are: working capital to total assets, return on assets, net worth to fixed assets, fixed assets to total assets, current ratio, net worth to total assets, sales to total assets, and cash to total assets. The return on assets variable appeared to be the most related for corporate failure prediction according to the study. However, Lizal (2002) finds return on assets to be insignificant. In Janer (2011) the return on assets is significant only at 95% significance level, but he utilizes the assets turnover ratio, which is significant and sometimes stands as a substitute for return on assets.

Blum (1969) constructed a model that is based on both financial and accounting data. A sample consists of 115 failed companies for the period of 1954-1968 and 115 non-failed companies paired according industry, number of employees and fiscal year. The model distinguishes bankrupt and active companies with an accuracy of 94%, when failure occurred

11

within one year before bankruptcy, 80% for failure two years prior and 70% for failure three, four and five years before.

Shirata (1998) developed a model for predicting bankruptcy of Japanese firms. In this model he did not include profitability and liquidity ratios. He argues that these ratios can predict company failure only till a certain extent. Another point that deserves attention is that his model is universal and not affected by industry and size of assets. The model can predict with an accuracy of more than 86.14 per cent regardless of industry and size.

Yazdanfar (2008) employs 24 financial ratios to identify the main determinants that cause Swedish companies to go bankrupt. The results imply that financial ratios lead to useful predictions of corporate failure. Solvency ratio, quick ratio and return on assets were found to be the most important determinants in one year prior to failure. Also the solvency ratio is more apparent than other bankruptcy determinants for one to three years before bankruptcy. The classification accuracy rate varies from 83.5% for one year prior to failure to 77.8% for three years before the bankruptcy.

Lugovskaya (2009), in a paper that predicts bankruptcy of Russian SMEs, has found that the following six variables were significant for bankruptcy prediction: cash to current liabilities, current assets to current liabilities, (cash + short term debtors) to current liabilities, current liabilities to total assets, return on assets and cash to total assets. Liquidity and profitability ratios showed to be the most important factors. The model has a high classification accuracy. In five years before default 93.1 per cent of the companies are correctly classified.

Not a lot of researches examined the role of cash flows in the prediction of corporate failure. Gilbert et al (1990) concluded that cash flow variables are significant and influence predictive power of the prediction model. Cash flows play a more important role in predicting corporate failure than accrual earnings, because they have a direct connection with the liquidity ability of the company to repay its debt (Bernard and Stober, 1989). On the contrast, Takahashi et al (1984) has found in his study that prediction models based on accrual ratios have higher accuracy rather than based on cash flow ratios and that the later ones do not provide any additional power. Charitou et al (2004) did not find cash flow ratios to have predictive power either.

Concerning data of Dutch companies there was a research done by Slotemaker in 2008. In his research he tried to construct a model for bankruptcy prediction of small sized, private Dutch companies. The data was analyzed using multiple discriminant analysis with five ratios to predict bankruptcy. Results show that a 3-year trend of financial ratios and a 3-year

12

average of financial ratios, are useful to predict corporate failure. Moreover, current ratio, solvability and working capital to total assets ratio have most predictive power in the model while the cash flow ratio and the net income to total assets ratio do not contribute to model at all.

This work, firstly, has long and more up-to-date period of observations, which allows to extract more information from the data. Secondly, it utilizes more explanatory variables that allows to capture different effects. And finally more properly expresses other variables. For example, liquidity is better captured by the current ratio of current assets to current liabilities, than by the ratio of net current assets to total assets.

Searching through the literature we see that multiple discriminant analysis was used before probit and logit models. Nowadays people move toward probit and logit ones for both theoretical and empirical reasons. The probit model requires less restrictive statistical assumptions, and offers a better empirical discrimination. Moreover, the literature overview shows that we will be able to make the first report on the Dutch data for financial and non-financial firms by using a probit model.

3. METHODOLOGY

In this chapter we discuss plausible evaluating methodologies for testing suggested hypothesis. Main concerns are models and variables.

We are interested in finding this marginal firm, which is subject to failure. There are other things that may affect bankruptcy probability such as bad management, declining sales, increase in expenses or bad luck issues. But these effects must be reflected in financial figures, so we will count on them.

We estimate bankruptcy determinants for enterprises. There are virtually two possible cases: a firm goes bankrupt or does not. For this type of the analysis we have three possible methodologies: multiple discriminant analysis, logit and probit models. As discussed above, MDA models are based on assumptions that are sometimes violated in predicting failure of a firm. That is whynowadays authors shift toward logit and probit models. In the thesis we use the probit model on panel data to avoid well-known problems associated with multivariate discriminant analysis.

13

The Probit Model

Probit and logit models are the most widely used representatives of the linear models with binary dependent variables. Both functions are practically identical. Chambers and Cox (1967) found the empirical support that it was only possible to distinguish results of two models when sample sizes were large and certain extreme patterns were observed in the data. Probit regression analysis is used to develop bankruptcy prediction models in one, two, three, four and five years prior to the failure. When using the probit model, we are modelling the probability of company failure, that is:

P 𝑌 = 1 𝑋 = Ф 𝑋`𝛽 (1)

Here Ф(𝑋`𝛽) is the cumulative distribution function of the standard normal distribution. In other words, the probability of a company going bankrupt is a certain function of a linear regression of independent variables.

In general, it is very difficult to interpret estimated coefficients of probit regression. For this reason we need to make an additional step in our computation to get the marginal effects of the independent variables. Marginal effects measure the effect of a unit change of the regressor on the probability of fail given that all other regressors are constant.

When using a panel data we have to take into account the panel nature of our data by using fixed-effects or random-effects models. In the thesis we use a random-effects model because we believe that differences across entities have some influence on the probability of a company going bankrupt. Moreover, using random-effects model allows us to include time invariant variables. The results of the empirical study of Sohn and Kim (2007) indicate that the classification accuracy of the random-effects model is far better than that of the fixed-effects regression model in case of predicting a corporate failure.

Selection of prediction variables

Financial ratios allow us to compare companies across different industries, sizes and to identify strength and weaknesses of a company. After reviewing various literature concerning prediction of companies going bankrupt we define a minimum set of variables. As it was stated before financial ratios can be classified into four categories as follows: profitability ratios, liquidity ratios, solvency and efficiency ratios.

Profitability ratios measure the ability of company to generate earnings relative to sales, assets and equity. Liquidity ratios determine a company’s ability to pay off its short-terms debt obligations, while solvency ratios measure how a firm is able to pay off its long-term

14

debt and the interest on that debt. Finally, efficiency ratios are used to identify how well a company manages its assets and liabilities.

In order to make our analysis more accurate we need at least one variable per each group. However, including too many variables can cause severe multicollinearity, so under certain criteria we have defined a set of seven financial ratios for the current analysis. Moreover, we include in our model also the following variables: size of the company, age of the company and industry variable.

Three criteria were used to select ten variables from all possible combinations. The first criteria, is the popularity in the literature. The second criteria, is the potential relevance to this study. The third criteria, is availability of the necessary information about a company in the database. The following set of prediction variables should be well enough to explain the probability of bankruptcy events:

1. Return on assets:

𝑅𝑂𝐴 = 𝑁𝑒𝑡 𝐼𝑛𝑐𝑜𝑚𝑒

𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠 (2)

ROA indicates how effectively a company uses its assets in generating revenue. It is the most common financial ratio among various researches on determinants of company`s bankruptcy. This indicator has a few positive features: it provides the information of how well the firm works through net income in the numerator and it is adjusted to the size of the firm through the total assets in denominator. According to Bellovary et al. (2007), this ratio explains more than half of variations in bankruptcy differences. Yazdanfar et al (2008) found return on assets for Swedish companies to be an important determinant of bankruptcy in one year prior to failure. However, other authors, for example Lizar (2002), found this variable to be insignificant for The Czech Republic.

2. Return on equity:

𝑅𝑂𝐸 = 𝑁𝑒𝑡 𝐼𝑛𝑐𝑜𝑚𝑒

𝑆𝑕𝑎𝑟𝑒𝑕𝑜𝑙𝑑𝑒𝑟𝑠 𝐸𝑞𝑢𝑖𝑡𝑦 (3)

Return on equity measures the amount of net income earned as a percentage of shareholder`s equity. Fijorek et al (2012) found that ROE was important in 33% cases of predicting bankruptcy for Polish companies.

15

3. Pre-tax profit margin:

𝑃𝑟𝑒 − 𝑡𝑎𝑥 𝑝𝑟𝑜𝑓𝑖𝑡 𝑚𝑎𝑟𝑔𝑖𝑛 = 𝑃𝑟𝑜𝑓𝑖𝑡 𝑏𝑒𝑓𝑜𝑟𝑒 𝑡𝑎𝑥

𝑂𝑝𝑒𝑟𝑎𝑡𝑖𝑛𝑔 𝑟𝑒𝑣𝑒𝑛𝑢𝑒 (4)

This ratio helps to measure profitability of the firm before deducting taxes. The higher the pre-tax profit margin the more profitable is the company. Aghaie and Saeedi (2009) included this variable in the research, but they dropped it during the picking of the final set of variables. This indicator may well be of both positive and negative signs. On the one hand, high profitability may insure any company from negative operating issues. Price decrease, as a result of a downturn, would hurt the firms with a higher margin less. On the other hand, a high margin might indicate a new rising industry, where companies are more volatile. More volatile companies have a higher probability to go bankrupt.

4. Current ratio:

𝐶𝑢𝑟𝑟𝑒𝑛𝑡 𝑟𝑎𝑡𝑖𝑜 = 𝐶𝑢𝑟𝑟𝑒𝑛𝑡 𝐴𝑠𝑠𝑒𝑡𝑠

𝐶𝑢𝑟𝑟𝑒𝑛𝑡 𝐿𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑖𝑒𝑠 (5)

The current ratio indicates how a company is able to meet its short-term debt obligations. It is the second most common ratio in the literature concerning prediction of bankruptcy. In theory, the higher the current ratio, the better. Mervin (1942) found the current ratio to be a significant indicator of failure of a company. However, FitzPatrick (1932), Smith and Winacor (1935) found this ratio to have a less predictive power than the return on assets ratio.

5. Quick ratio:

𝑄𝑢𝑖𝑐𝑘 𝑟𝑎𝑡𝑖𝑜 =𝐶𝑢𝑟𝑟𝑒𝑛𝑡 𝐴𝑠𝑠𝑒𝑡𝑠 − 𝑆𝑡𝑜𝑐𝑘𝑠

𝐶𝑢𝑟𝑟𝑒𝑛𝑡 𝐿𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑖𝑒𝑠 (6)

The quick ratio is an important ratio because it measures the liquidity of firm more precisely. It concentrates on more liquid current assets (cash, marketable securities, receivables) in relation to current obligations (Gon Kim et al, 2005). When making valuation of the liquidity of the firm it is a common practice to look on both ratios: current ratio and quick ratio. Yazdanfar et al (2008) found the quick ratio for Swedish companies to be an important determinant of bankruptcy in one year prior to failure.

6. Shareholder equity ratio:

𝑆𝑕𝑎𝑟𝑒𝑕𝑜𝑙𝑑𝑒𝑟 𝑒𝑞𝑢𝑖𝑡𝑦 𝑟𝑎𝑡𝑖𝑜 =𝑆𝑕𝑎𝑟𝑒𝑕𝑜𝑙𝑑𝑒𝑟𝑠 𝐸𝑞𝑢𝑖𝑡𝑦

16

This variable determines the amount of assets that are financed by investors. The higher shareholder equity ratio the more favorable it is for company. Firms with a higher ratio should have less debt and financing costs comparing to companies with a low equity ratio. Pompe et al (2005) found this ratio to be more significant for Belgium than in the study of Beaver (1966) for USA companies.

7. Net asset turnover:

𝑁𝐴𝑇 = 𝑂𝑝𝑒𝑟𝑎𝑡𝑖𝑛𝑔 𝑅𝑒𝑣𝑒𝑛𝑢𝑒

𝑆𝑕𝑎𝑟𝑒𝑕𝑜𝑙𝑑𝑒𝑟𝑠 𝑒𝑞𝑢𝑖𝑡𝑦 + 𝑁𝑜𝑛 − 𝑐𝑢𝑟𝑟𝑒𝑛𝑡 𝐿𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑖𝑒𝑠 (8)

The net asset turnover ratio indicates the amount of revenues or sales generated by assets. The higher the ratio, the better it is since it means that higher revenue is generated by the assets. This is one of the five variables used by Altman (1968).

8. Size of a company

The size variable is measured as a natural logarithm of total assets. It is assumed that firms grow up due to different economic benefits: economy of scale, market power. If it is so, then bigger firms should be less vulnerable than smaller ones. Big firms with the same ratios as small ones should be less probable to go bankrupt. We expect this coefficient to have a negative sign in our research.

9. Industry variable

The industry variable in this analysis is presented as a dummy variable that takes the value only of 0 or 1. We use industry dummies to control for industry specific shocks and industry export orientation. Industry classification was made according to Statistical Classification of Economic Activities in the European Community Rev.2 (NACE Rev.2) as shown in the Table 1.

Table 1: Industry classification Industry

1 Manufacturing

2 Electricity, gas, steam and air conditioning supply 3 Construction

4 Wholesale and retail trade 5 Transportation and storage 6 Information and communication 7 Financial and insurance activities

8 Professional, scientific and technical activities 9 Administrative and support service activities 10 Human health and social work activities

17

10. Age of a company

The age of a company is usually significant in evaluating risk. An analysis between age of the business and failure of the company when age of companies is combined in groups usually yields more statistically significant results than each year evaluated separately (Garavaglia and Sharma, 1994). Pompe et al (2005) divide their sample of companies according to age of a firm into two groups: young firms whose life span is 8 years or less and old firms that had a life of more than 8 years. Aleksanyan et al (2014) in their work define three groups of firms according to their age: young firms (less than 5 years), middle-aged firms (5-10 years) and old firms (more than 10 years). In this research we consider two groups of companies according to their age: young (less than 10 years) and old (10 years and more). The age of the company is not given in the database. Thus, we calculated it as the time between year of incorporation and year of the last available financial report.

Based on the results of previous works, “common sense” and theory we make expectations concerning the signs of coefficients of the explanatory variables. Expected signs are introduced in Table 2.

Table 2: Expected signs for independent variables

Name of variable Expected sign

1 2

Return on Assets Negative

Return on Equity Negative

Pre-tax profit margin Negative

Current ratio Negative

Quick ratio Negative

Shareholder equity ratio Negative

Net Asset Turnover Negative

Size Negative

Age dummy Negative

18

4. SAMPLE SELECTION AND DATA SOURCES

Almost in all countries around the world enterprises are required to submit their annual financial reports at least once a year. The Netherlands is not an exception. Necessary data for our research contains in the statement of financial position (balance sheet) and income statement. These data can be obtained from The Dutch Chamber of Commerce (KVK – Kamer van Koophandel). Every company in Holland should make registration in the trade register. The Dutch Chamber of Commerce keeps on track this trade register. Moreover, there is a big Dutch bankruptcy database that makes updates on a daily basis: faillissement.com.

In the current research we use the data of the global database Orbis from Bureau van Dijk (BvD). This database contains comprehensive information on more than 150 million companies worldwide with possibility of selection bankrupt and non-bankrupt firms.

The Orbis database suggests various options for selecting company`s status. The legal statuses are combined into three categories: active, inactive and unknown. Several levels are possible for each category. To select the appropriate status of a company we need first to define what is meant by a bankrupt and non-bankrupt firm.

Bankruptcy usually refers to declared inability of a company to pay its creditors. In this study bankrupt company is defined as a firm that is in the process of bankruptcy and all assets are being sold to repay the creditors. A non-bankrupt company is defined as a company that still operates and has trading activity. Thus, non-bankrupt companies with “active” status in Orbis database and bankrupt companies with “inactive (bankruptcy)” status were selected for this research.

Other criteria for selecting companies were location and availability of financial information. The goal of this research is to identify determinants that cause companies of the Netherlands to go bankrupt. Information needed for defining financial ratios contains in the financial and income statements of enterprises. Under Dutch law small and medium-sized companies are allowed to fill reduced financial statements in a simplified form. This means that certain financial information can be missed. Moreover, a lot of companies have gap years in reporting, whereas others have reports just for one period. That is why we have eliminated companies with insufficient financial information.

The data for this analysis is not limited by the number of employees. The reason is that this information is not available for every company in the sample. Moreover, we don`t use a matching method of failed and non-failed firms in the thesis. Altman (1968) pointed that there is no completely clear relationship between the size of company, the number of

19

employees and the financial ratios. Beaver (1966) also suggests that ratios are uncorrelated with industry and assets size. Main reason is that with the matching method it will not be possible to investigate the effect of company`s size, industry and age on the probability of going bankrupt.

For our analysis we use panel data that is unbalanced. Notebly, Orbis database includes only a small number of companies that went bankrupt and have available financial reports for at most three years prior to failure. The reasons is that a company subject to a court-administered proceeding (reorganization or liquidation) does not always provide its balance sheet for the year of failure. For this reason, our sample of bankrupt firms includes 69 companies with balance sheets provided with at most three years before firm failure. Sample of non-bankrupt firms consists of 182 company that were randomly selected out of 5,006 active companies. The number of observed firms differs from period to period (Table 3). Summary statistics of independent variables is provided in Table 4.

Table 3: Number of observations

Note. Source is Orbis database from Bureau van Dijk (BvD).

Table 4: Descriptive statistics of observed firms

Year `08 `09 `10 `11 `12 `13

Average total assets, mil 10.8 10.94 12.01 12.06 12.45 13.29

Average ROA, % 6.6 3.6 4.42 3.09 2.18 2.14

Average CR 1.47 1.48 1.48 1.52 1.45 1.34

Average NAT 13.25 10.17 10.06 10.1 8.74 9.29

From Table 4 we can observe the growth of total assets in all periods. This observation supports the fact that we follow the same companies, and they grow during the history. Decreasing ROA indicates that companies become less profitable. The factors that can cause return on assets to decline are decreasing net income and increasing total assets, such as high purchases of fixed assets or poor collection of accounts receivable. The average current ratio

Year `08 `09 `10 `11 `12 `13 Total

Total

observations 59 135 251 251 192 116 1004

% in total

20

does not fluctuate significantly during the observed period. The higher the current ratio the more company is able to pay it`s short-term debts. A current ratio of approximately 1.0 indicates that the company can have difficulties to cover her current liabilities and a measure of 2.0 is a good indicator to identify a company with a favorable liquidity position (Cagle et al, 2013). Our average current ratio is slightly below 1.5. That is an acceptable value for companies according to different literature. Moreover, we calculate the net asset turnover for all firms, with interest how this variable affects bankruptcy vulnerability. Logically thinking, companies with fast assets turnover are less sensitive to market downturns, because they can faster turn their assets into money. As previously mentioned, the net asset turnover ratio measures how efficiently a firm`s assets generate revenue. In Table 4, for example, in 2008 year the average net asset turnover for companies was 13.25, that means every 1 EUR worth of assets generated 13.25 EUR of revenue. The higher the ratio – the better. We can see that the average net asset turnover declines over the observed period. This can be a sign that companies are overinvesting in assets. Other reasons for decrease in the average net asset turnover can be the asset`s age or the industry in which the company operates. Older assets and capital-intensive industries tend to have lower net asset turnover.

An important problem that can arise during the analysis is the multicollinearity of independent variables. Since some financial ratios use the same variables in the calculation (net income, current assets, total assets) there is a high risk of multicollinearity. The problem of multicollinearity can cause increase of standard errors of coefficients and as a result some coefficients for independent variables may be found statistically insignificant. To check for multicollinearity we use two tests: the variance inflation factor (VIF) and Pearson correlation matrix. The VIF can be calculated for each predictor by doing linear regression of that predictor on all other predictors, thus ignoring the dependent variable. For this reason we can run a simple regression with the same list of independent and dependent variables for identifying VIF. Results of both tests are provided in Table 5 and Table 6.

21

Table 5: Variance inflation factor

VIF 1/VIF 1. ROE 1.22 0.8186 2. PM 1.4 0.7133 3. NAT 1.12 0.8918 4. QR 1.4 0.7163 5. CR 2.06 0.4852 6. SER 1.65 0.6055 7. Size 1.04 0.9616 Mean 1.41

Note. 1/VIF is the tolerance.

Table 6: Pearson correlation matrix

ROA ROE PM NAT QR CR SER Size

1. ROA 1.0000 2. ROE 0.6442* 1.0000 3. PM 0.6459* 0.4042* 1.0000 4. NAT -0.0188 0.0252 -0.0901* 1.0000 5. QR 0.1220* 0.0420 0.1531* -0.0297 1.0000 6. CR 0.2883* 0.1033* 0.3688* -0.1002* 0.5283* 1.0000 7. SER 0.3582* 0.1700* 0.3251* -0.2811* 0.2539* 0.5666* 1.0000 8. Size -0.0041 0.0726* 0.0610 -0.1163* -0.0546 -0.0908* -0.0465 1.0000

Note. Correlation coefficients are significant at *p<0.05.

For the VIF test we run regression with the ROA as dependent variable and all other variables as independent. This auxiliary regression model with a VIF less than 10 indicates that our analysis is free from multicollinearity. The second test, however, indicates that some of the measurements have a high degree of correlation. For example, return on assets and return on equity have strong and positive correlation. That can be explained by common numerator – net income. As high degree of correlation can lead to misleading results, further in our analysis we make separate regressions for each variable to find out the effect and the predictive power of each variable.

22

5. ESTIMATION RESULTS

An important initial step in the analysis is the identification of possible differences between failed and non-failed companies. For this reason we look at main descriptive statistics: mean and standard deviation for bankrupt and non-bankrupt firms. This overview could be treated as the eyeball test. Obviously, firms that are prone to bankruptcy should emanate deteriorating trends in their operating ratios. Further, we start building our random-effects probit model. In the previous chapters we already mentioned which ratios will be used, but we are not going to put all financial ratios at once. The objective of this probit analysis is to evaluate the predictive ability of the individual variables. Thus, we are going to run separate regressions on each independent variable to figure out which one is more predictive. The latest financial data of the companies were used for this purpose. Next, we are going to run separate regressions on each lagged independent variable to figure out which is a better predictor in one, two and three years prior to failure.

Bankruptcy of a company is not an accidental event. It is rather an end of a long process experienced by firms that is different from the process of the healthy firm. Taking this into account we expect observable differences for bankrupt and non-bankrupt firms (Table 7).

Table 7: Summary statistics for failed companies one, two and three years prior to bankruptcy and non-failed companies.

Three years prior to bankruptcy

Two years prior to bankruptcy

One year prior to

bankruptcy Non-bankrupt firms

N = 69 N = 69 N = 69 N = 935

Variable Mean Stdv Mean Stdv Mean Stdv Mean Stdv

ROA, % 3.44 0.11 3.14 0.09 0.53 0.09 3.8 0.09 ROE, % 5.01 0.51 -6.8 1.17 -6.31 0.355 8.5 0.46 PM 1.04 14.24 3.69 12.96 0.86 10.76 3.33 9.17 NAT 9.88 13.15 8.58 11.19 10.68 15.79 9.76 26.64 CR 1.41 0.9 1.5 1.15 1.57 1.68 1.46 0.98 QR 0.98 0.62 1.05 0.84 1.04 1.2 1.09 0.85 SER, % 28.2 0.16 31.68 0.18 30.5 0.18 32.25 0.18 Size 9.9 1.85 9.88 1.84 9.87 1.88 12.02 1.95

Note. N is a number of observations. Stdv is a standard deviation.

As descriptive statistics indicate, before the failure the financial situation of the companies demonstrates signs of difficulties and different behavior compared to active companies. Failed companies exhibit a positive ROA, but their profitability deteriorates as

23

the interval time before the failure event elapses. This decline indicates that firms moving towards failure have decreasing net income and have difficulties with finding and capturing attractive opportunities. Return on equity drops significantly two years prior to bankruptcy. Low return on equity indicates that the management of companies do not use investor`s funds effectively and companies experience substantial losses. From financial perspective it means that the company will become bankrupt. The analysis also indicates a low level of another profitability ratio, the pre-tax profit margin. Decline in pre-tax profit margin implies that fewer funds are available for operating expenses and taxes. This can happen because of lower sales performance, increase in cost of goods sold or an intense competition. For bankrupt firms current and quick ratios, on average, are almost the same as for non-bankrupt ones. However, in some years, ratios are higher for bankrupt firms. Further research is needed to explain why such differences may occur. Maybe in booming economy they attract a lot of long-term capital and overinvest in current capital. High net asset turnover is observed among both failed and non-failed companies. Usually high net asset turnover ratio implies that firm`s funds are used efficiently. However, this ratio may be artificially overestimated when passing to use of leased assets.

The first step in building up our model is to run regression only on intercept. This gives us an opportunity afterwards to find goodness of fit of estimated models. Second step is to run a separate regression on each variable. This allows us to look at each variable in isolation and find out which variable predicts better. Next, we run regressions including all our independent variables. Results for the first bunch of regressions are presented in Table 8. As noted previously, estimated coefficients of probit regression are difficult to interpret. Therefore, coefficients that are presented in Table 8 are marginal effects for the average firm.

To evaluate the goodness-of-fit of our probit model, several pseudo 𝑅2 _{can be applied.}

We employ the most widely used pseudo 𝑅2_{of McFadden that is calculated as:}

𝑃𝑠𝑒𝑢𝑑𝑜 𝑅2 _{= 1 −} 𝑙𝑛 𝐿 (𝑀𝑜𝑑𝑒𝑙)

𝑙𝑛 𝐿 (𝐼𝑛𝑡𝑒𝑟𝑐𝑒𝑝𝑡) (9) Where:

𝐿 (𝑀𝑜𝑑𝑒𝑙) – estimated likelihood of model with predictors;

24

Table 8: Estimation results of probit models.

Explanatory variables Step 1 Step 2 Step 3 Step 4

1 3 4 5 6 ROA -.370*** -.232*** -.215*** -.140*** (.08) (.09) (.08) (.08) ROE -.05*** -.012 -.013 -.007 (.01) (.009) (.009) (.009) PM -.002*** .001 .001 .0002 (.001) (.007) (.0001) (.0001) CR .005 .005 .006 .005 (.006) (.007) (.007) (.007) QR .004 .003 .003 .002 (.003) (.002) (.002) (.003) NAT .0002 -.0002 -.0002 -.0002 (.0003) (.0003) (.0003) (.0003) SER -.062 -.069* -.075* -.037* (.043) (.041) (.04) (.03) Size -.022*** -.022*** -.015*** (.003) (.003) (.003) Age -.023** -.018** (.01) (.02) Number of observations 1004 1004 1004 1004

Industry dummies No No No Yes

Pseudo 𝑅2 _{No 53.04% 53.93% 54.51%}

Note. Significance levels (p-values): *p<0.1; **p<0.05; ***p<0.01. Values appearing in parentheses are the relevant standard errors.

Step 1. Marginal effects for separate regressions for each independent variable.

Step 2. Marginal effects of regression that includes all financial ratios and size variable. Step 3. Marginal effects of regression that includes all financial ratios, size and age variable. Step 4.Marginal effects of regression that includes all explanatory variables.

Higher values of Pseudo 𝑅2 _{indicate better fit of the model. In general, adding variables}

to the model increase the indicator of fit and a full model is far better fit than the intercept model. From our regressions in Table 8 we see that the explanatory power of the model increases as we add variables. So, some of the added variables have a good explanatory power. The obtained signs for ROA and ROE denote consistency with our expectations. A higher value of ROA and ROE is associated with lower probability of failure. This is in

25

accordance with findings of Altman (1968), Janer (2011) and Lizal (2007). Likewise, the size variable indicates that larger companies are less likely to go bankrupt. The PM, CR, QR, NAT ratios mostly appear in the probit functions with positive coefficients, what contradicts with our expectations. However, estimated coefficients for these ratios are statistically insignificant, so we cannot rely on them. This can be explained by high correlation among variables and by small number of observations.

Adding age and industry dummies increase the explanatory power of our model. We find a negative and significant effect of the age variable. As expected and according to the theoretical literature, young firms tend to be more vulnerable to bankruptcy risk than elder one. These findings support the first hypothesis that smaller and younger companies are more likely to fail.

We observe the significance of some industry variables in the extended model. Extractive, wholesale and retail trade industry is significantly more risky than average firms. So providing credit lines or investing equity in it, the investor should demand higher return. On the contrary, the financial sector, human health and social work activities are less risky than others, so risk-averse investors and pension funds should focus on them.

According to the literature financial ratios start to predict bankruptcy prior to the actual date (year) of failure. For this reason in the next part of our analysis we use variables that are lagged three periods (years). Table 9 presents the probit regressions results ran for each explanatory variable separately for one, two and three years before failure. Estimated coefficients are marginal effects for an average firm.

Not unexpectedly, the results of looking on every lagged variable in isolation yield less significant results over longer periods prior to bankruptcy. ROA and size tend to be the most important determinants of corporate failure. The coefficient of return on assets is always negative and significant. This implies that, other things equal, profitable companies have a lower probability to fail, as expected. However, this effect diminishes when increasing years before bankruptcy. In theoretical literature on prediction of bankruptcy, the return on assets ratio appears almost in all studies as statistically significant in prediction of corporate failure three years prior to default (Pervan et al., 2011).

26

Table 9: Estimation results of separate probit regressions

Explanatory variables Year prior to failure

1 2 3 ROA -.234*** -.124*** -.089* (.085) (.087) (.084) ROE -.022* -.018* -.001 (.012) (.012) (.015) PM -.002** .0004 -.001* (.001) (.001) (.001) NAT .0001 -.0002 -8.52e-06 (.0003) (.0004) (.0003) CR .005 .002 -.004 (.007) (.007) (.009) QR -.004 -.003 -.013 (.009) (.009) (.012) SER -.032 -.007 -.082* (.044) (.044) (.045) Size -.025*** -.025*** -.025*** (.003) (.003) (.003) Number of observations 1003 1002 1001

Note. Significance levels (p-values): *p<0.1; **p<0.05; ***p<0.01. Values appearing in parentheses are the relevant standard errors.

Presented coefficients are the marginal effects of regressions done separately for each lagged variable.

The results for size variable have negative and strongly statistical coefficient for all time periods to failure. This confirms our hypothesis that small companies are more likely to go bankrupt. Moreover, ROE and PM ratios have also some significant impact on predicting bankruptcy of a company in one, two and three years prior to failure. But, these coefficients lose significance as a bankruptcy prediction measure when the time horizon exceeds one year. Finally, liquidity, solvency and efficiency ratios as can be seen are irrelevant to bankruptcy prediction no matter of time period prior to failure. This implies that profitability ratios are more significant than other financial ratios in prediction of default in one, two and three years before the failure.

Results of probit models that include all variables for the different time horizons prior to failure are presented in Table 10. Estimated coefficients are marginal effects for an average firm.

27

Table 10: Estimation results of probit regressions

Explanatory variables Year prior to failure

1 2 3 ROA -.196** -.031 .065 (.099) (.087) (.091) ROE .019 -.009 .014 .015 (.009) (.015) PM -.001 .001 -.002** (.001) (.001) (.001) NAT -.0002 -.001 -.001 (.0002) (.0004) (.0004) CR .023*** .005 .02** (.008) (.01) (.009) QR -.019* -.019* -.022* (.01) (.012) (.011) SER -.057 -.03 -.107** (.041) (.045) (.044) Size -.023*** -.024*** -.022*** (.003) (.003) (.003) Pseudo 𝑅2 _{49.8% 41.99% 43.61%} Number of observations 1003 1002 1001

Note. Significance levels (p-values): *p<0.1; **p<0.05; ***p<0.01. Values appearing in parentheses are the relevant standard errors. Presented coefficients are the marginal effects of regressions.

The explanatory power of the variables grows when the time to failure shortens. Pseudo 𝑅2 _{increases from 43.61% (three years prior to failure) to 49.8% (one year prior to failure).}

However, it is surprisingly that the explanatory power of the model three years prior to bankruptcy is higher than for the model of two-years prior to failure. Again consistent with expectations is the finding that the size variable has the strongest significance that doesn’t vary with time period before the bankruptcy. The return on assets ratio, however, does not show any systematic pattern. The coefficients of liquidity and solvency do not show any systematic pattern either. Nevertheless, current ratio and quick ratio appear to be important only in the probit functions 1-3 years before default. Bellovary et al (2007) consider current ratio to be a good indicator in failure prediction. This finding is supported also by Smith and Winakor (1935), Mervin (1942) and Janer (2011).

Predictive performance

Previously we checked our models on a goodness-of-fit with McFadden`s pseudo 𝑅2, but there is also another way to assess the fit of our probit regression. For model assessment we use the receiver operating characteristic (ROC) curves. Dakovic et al (2010) used this

28

method in evaluation the explanatory power of their models for prediction bankruptcy of companies in Norway. Agarwal et al (2007) employed in their paper ROC curves for evaluating performance accuracy of market-based and accounting-based bankruptcy prediction models.

The analysis uses the ROC curve, the graph of the sensitivity versus specificity. The sensitivity is the proportion of positive cases that are correctly classified as being positive. The specificity is the proportion of negative cases that are correctly classified as being negative. More precisely, we use the area under the ROC curve. It is usually measured relatively to the area of the unit square. The rule of thumb indicates a value of 0.5 as a model with no predictive ability and a value of 1.0 as a perfect classification. To estimate the ROC measure we use the 2008-2012 period to compute coefficients of models and then we use these coefficients to forecast bankruptcy of companies for 2013 year. Table 11 reports areas below the ROC curves. The graphs of ROC curves can be found in Fig.1.

29

Table 11: Results for area below the ROC curves

Model Number of observations ROC Area Std. error

1. Model 1 (p1) 885 0.8466 0.0199

2. Model 2 (p2) 885 0.8374 0.0198

3. Model 3 (p3) 885 0.8322 0.0185

4. Model 4 (p4) 885 0.8385 0.0183

Note. Values are significant at p<0.05.

Model 1. Probit regression which includes financial ratios, size and age variables.

Model 2. Probit regression which includes financial ratios and size variable that are lagged one period, two periods (Model 3) and three periods (Model 4).

The greater the area under the curve (AUC), the better is the performance of the test. We observe a high AUC for all models. That is considered to be a good indicator of prediction accuracy. The first model has a higher value to compare with other models.

Dakovic et al (2010) found the area under the ROC curves for their model to be 0.83, however, the model with Altman`s variable has a lower accuracy. Lanine and Vennet (2006) in the study on examining the efficiency and the accuracy of the logit model in assessing the risk of failure of Russian commercial banks acquired an AUC value of 0.75.

6. CONCLUSIONS

The goal of this thesis was to create a probabilistic model for predicting corporate failure and on its basis to test hypothesis concerning the main determinants that can cause companies to go bankrupt in the Netherlands.

After having a look on existing models of bankruptcy prediction we can imply that all models differ in assumptions, results, limitations and predictive power. Many researches were developed in different countries. However, in the Netherlands only two studies were conducted in 1979 and 2008, using multiple discriminant analysis. Yet they ignored size, age and industry variables effect on the prediction model of bankruptcy and used data for pre-crisis period. Therefore, we were motivated to develop the model on failure prediction using recent information of Dutch companies.

The results we have obtained partially support our suggested hypothesis and are consistent with other studies on corporate prediction. The size variable has negative relationship with corporate failure. Meaning that the smaller is the company in terms of total assets the more likely it will go bankrupt. Most common reasons for small businesses to fail

30

are insufficient capital, poor management and lack of planning (Bruno et al, 1987). Also size variable has strongly statistical coefficient for all time periods to failure. The same situation is with the age of a company variable. The fewer years it operates on the market the higher is the probability of default. Based on these results the first hypothesis “Smaller and younger Dutch companies have a higher probability of going bankrupt comparing to medium-sized and large established companies” is approved.

Performance tests of our study indicate that models using financial ratios have high predictive power. These findings support our second hypothesis “Financial ratios are useful in predicting corporate failure of small, medium-sized and large companies in the Netherlands”.

The third and final hypothesis “Profitability and solvency ratios tend to have higher influence than liquidity and efficiency ratios in predicting bankruptcy of firms” is accepted only partially. The return on assets ratio is negative and significant for the one, two and three years prior to failure. The return on equity and the pre-tax profit margin ratios have less predictive power and are less significant no matter of time period prior to failure. The solvency ratio appeared to has less influence on failure prediction comparing to liquidity and efficiency ratios. Moreover, some results show that the shareholder equity ratio is insignificant and has no predictive power on default of a company.

Clearly, there is a number of limitations in this research. One of the limitations is a relatively small sample of bankrupt companies. A larger sample can increase the chance to identify outliers among the population, increase the chance of obtaining significant results and what is more important; increase accuracy of obtained estimations. Another limitation is the period of study. However, obtaining data for a longer period is very difficult and costly.

Financial determinants play a key role in bankruptcy of companies. However, there are also other factors that influence failure of a company. Among them: bad management, unfavorable market conditions, crisis of resources. Thus, when applying the prediction models we should also take into consideration these factors.

To summarize, this study makes a contribution to prior studies in the following respects. First, the financial ratios of all classification groups were included in the model. Second, using a more recent company sample extends previous studies developed in the Netherlands. Third, it employs probit regression in analysis and it uses ROC curves to validate the results. According to receiver operating characteristic curves our model has a good predictive accuracy.

31

Although, the proposed model on prediction of corporate failure will be additionally helpful for investors on making investment decisions, credit issuing institutes on lending decisions and also for owners or managers to predict problems beforehand and make necessary actions to prevent bankruptcy.

The main part in a strategy of prevention of bankruptcy, solving problems of liquidity and solvency consists in professional managing of current assets. From one point of view, this assumes an optimization of sources of current assets on the basis of acquired strategy. From another point of view, this assumes allocation of current assets between fixed assets in manufacturing and distribution fields.

32

REFERENCES

Agarval, V., Taffler., R., (2007). “Comparing the performance of market-based and accounting-based bankruptcy prediction models”. Journal of Banking and Finance, 32, pp. 1541-1551.

Aleksanyan, L., Huiban J.P., (2014). “Economic and financial determinants of firm bankuptcy: Evidence from the French food industry”, University of Paris.

Aliakbari, Saeideh, (2009). “Prediction of corporate bankruptcy for the UK firms in manufacturing industry”. A Master’s Thesis, Brunel University.

Altman, E., (1968). “Financial ratios, discriminant analysis and the prediction of corporate bankruptcy”, The Journal of Finance 23(4), pp. 589-609.

Altman, E., (1993). “Corporate financial distress and bankruptcy: a complete guide to predicting and avoiding distress and profiting from bankruptcy”. John Wiley & Sons, New York.

Beaver, W., (1966). “Financial ratios as predictors of failure,” Journal of Accounting Research, 4, pp. 71–111.

Bellovary, J., Giacomino, D., Michael, A. (2007). “A review of bankruptcy prediction studies: 1930 to present”, Journal of Financial Education, 33, pp. 1-42.

Bernard, V. I., Stober, T. I. (1989). “The nature and amount of information in cash flows and accruals”, Accounting Review, 64(4), pp. 624–652.

Bilderbeek, J. (1979). “An empirical study of the predictive ability of financial ratios in the Netherlands. Zeitschrift fur Betriebswirtschaft, 5.

Brouwer, A., van Wissen L. (2005), “De Jaren Tellen al Vroeg Mee: Demografische Aspecten van Opheffingen.” In: P. Pellenbarg, P. van Steen & L. van Wissen, eds., Ruimtelijke Aspecten van de Bedrijvendynamiek in Nederland, pp. 127–140.

Bruno, A.V., Leidecker, J.K., Harder, J.W., (1987). “Why firms fail”. Business Horizons, pp. 50-58.

Cagle, Corey., S., Campbell, Sharon, N., Jones, Keith, T., (2013). “Analyzing liquidity: using the cash conversion cycle: method incorporating time complements static measures such as the more common current ratio”, Journal of Accountancy.

Casey, C., Bartczak, N., (1985). “Using operating cash flow data to predict financial distress: some extensions”. Journal of Accounting Research, 23(1), pp. 384–401.

Central Bureau of Statistics (CBS, 2015), Statline. Available at <www.cbs.nl>, Accessed on April 17, 2015.

Chambers, E. A., Cox D.R., (1967). “Discrimination between alternative binary response models”. Biometrika, 54, pp. 573–578.