LONG-TERM PREDICTIVE ABILITY OF BANKRUPTCY MODELS IN THE CZECH REPUBLIC: EVIDENCE FROM 2007-2012

Bankruptcy models are a common tool of financial analysis to predict the financial distress of companies. However, in the recent years, the instability and risk of the overall economic environment have underlined the need for accurate tools to predict bankruptcy and assess the overall performance of companies. In this article, we analyze the ex-ante predictive ability of selected bankruptcy and solvency models commonly used in financial analysis: Kralicek quick test, Taffler model, the IN99 and IN05 indexes, and Altman Z’-score models in the case of Czech companies from 2007 to 2012. We determined the percentage of cases when these models correctly predicted failures of companies up to five years in advance, and found that the IN05 and IN99 credibility indexes provided the best results, as well as the Altman Z’-score model. However, the predictive ability of the Taffler model and Kralicek quicktest has only been limited.


Introduction
The number of bankruptcies in the Czech Republic has kept growing since the introduction of the new insolvency law in 2008, which concerns, in particular, small businesses and individual entrepreneurs. The yields of creditors in insolvency proceedings have reached very low values (see Smr ka and Schönfeld, 2014 for a more detailed discussion). Companies may deal with a crisis by various measures, for instance, reassessment of investments, reduction of personnel costs or strengthening of cash ow (see e.g. Krause, 2013) or reappraisal of outsourcing activities (see e.g. Tyll, 2011). Nevertheless, a timely recognition of a company's crisis could reduce the harmful effects of company failures on individuals as well as on legal persons. In recent years, the instability of the economic environment underlined the need for accurate tools to predict bankruptcy and assess overall performance of companies. In this article, we test the predictive ability of selected bankruptcy and solvency models commonly used in nancial analysis: the Kralicek quick test, Taf er model, the IN05 and IN99 credibility indexes, and Altman Z'-score models in the Czech environment. The article tests whether these models had a real ex ante predictive ability using the data on Czech companies with more than 10 employees in the 2007-2012 period.

Bankruptcy and Solvency Models
A number of models to detect nancial distress have been developed in the past. Despite the variety of models available, business practitioners and researchers often rely on the most popular models, among which we can cite the Altman Z-score (Altman, 1968), Kralicek quick test (Kralicek, 1991), Taf er model (Taf er and Tisshaw, 1977) and in the Czech environment, the indexes of credibility (see, for instance, Neumaierová and Neumairer, 2014). The popularity of these models is due to the fact that they have a simple mechanism of calculation and an intuitive and clear way of interpreting the results. Many authors have tested the predictive ability of bankruptcy models in particular conditions in the past. For instance, Collins (1980) compared the Altman Z-score against more sophisticated ways of bankruptcy prediction and con rmed that the Altman's model performed well enough. Some authors consider the Altman model to be able to predict nancial distress 2-3 years in advance. More recently, Wang and Campbell (2010) examined the ability of Z-score models in China, and Agarwal and Taf er (2007) tested the predictive ability of the Taf er Z-score model on UK-based data and af rmed that the model had a good predictive ability. Concerning the Czech environment, the limits of bankruptcy models have been examined by many authors. ámská (2012)

LONG-TERM PREDICTIVE ABILITY OF BANKRUPTCY MODELS IN THE CZECH REPUBLIC: EVIDENCE FROM 2007-2012
Machek, O.
Bankruptcy models are a common tool of nancial analysis to predict the nancial distress of companies. However, in the recent years, the instability and risk of the overall economic environment have underlined the need for accurate tools to predict bankruptcy and assess the overall performance of companies. In this article, we analyze the ex-ante predictive ability of selected bankruptcy and solvency models commonly used in nancial analysis: Kralicek quick test, Taf er model, the IN99 and IN05 indexes, and Altman Z'-score models in the case of Czech companies from 2007 to 2012. We determined the percentage of cases when these models correctly predicted failures of companies up to ve years in advance, and found that the IN05 and IN99 credibility indexes provided the best results, as well as the Altman Z'-score model. However, the predictive ability of the Taf er model and Kralicek quicktest has only been limited.

JEL classi cation: G30, G33
Keywords: bankruptcy prediction; Altman Z'-score; Taf er model; Kralicek quick test; IN05; IN99 described bankruptcy prediction models which were created during the nineties in Central and Eastern Europe. Perhaps the most comprehensive research on the ef ciency of bankruptcy models in the Czech environment (before the global economic crisis in 2008) was done by Ma asová (2007) who concluded that the IN indexes performed well and the Altman's Z-score models had the best predictive ability in selected industries.

Altman Z'-Score model
Perhaps the most famous model is the Altman Z-Score which was originally published in 1968 (Altman, 1968) and further modi ed to better re ect particular operating conditions. The model is based on discriminate analysis. The Altman Z'-score for private rms can be speci ed as: where T 1 is the ratio of net working capital (current assets less current liabilities) over total assets, T 2 is the ratio of retained earnings over total assets, T 3 is the ratio of earnings before interest and taxes (EBIT) over total assets, T 4 is the ratio of equity over total liabilities and T 5 is the asset turnover (sales over total assets). According to the resulting value of the Z'-score, companies can be classi ed into the following groups:

Kralicek Quick Test
The quick test developed by Kralicek (1991) which was further modi ed in 1999 is an example of "solvency models" and evaluates the company's nancial and revenue position.
It takes into account multiple nancial ratios and assigns the following scores according to the resulting values (table 2).

Taffl er Model
The Taf er model developed by Taf er and Tisshaw (1997) is based on calculating the following score: where T 1 denotes earnings before taxes (EBT) over short-term liabilities, T 2 denotes current assets over total liabilities, T 3 denotes short-term liabilities over total assets and T 4 denotes the asset turnover (sales over assets). According to the resulting value of the nal score, companies can be classi ed into the following groups:  where T 1 denotes assets over liabilities, T 2 denotes EBIT over assets, T 3 denotes revenue over assets and T 4 is the ratio of current assets over the sum of short-term liabilities and short-term bank loans.
IN05 can be calculated as: where T 1 denotes assets over liabilities, T 2 denotes EBIT over interests, T 3 denotes EBIT over assets, T 4 denotes revenue over assets and T 5 is the ratio of current assets over short-term liabilities.
According to the value of the nal score, companies can be classi ed into the following groups:

Data and Methodology
We used the Albertina database maintained by the Bisnode company. This database contains nancial data about all subjects with a registered ID in the Czech Republic. We focused on companies with more than 10 employees in the 2007-2012 period (8 924 companies with a unique ID). The research was based on an observation whether the models did or didn't recognize companies which would go bankrupt in the following years. Since we had observations from the period 2007-2012, we were able to analyze the predictive ability of the respective models ve years in advance (for 2012), four years in advance (for 2011) and three years in advance (for 2010).

Results and Discussion
In the following table, we summarize the critical values of the above-mentioned prediction models. A model is supposed to detect forthcoming bankruptcy if the resulting value lies within the below-de ned intervals. The results are summarized in tables 6, 7 and 8. The ve-year predictive ability for 2012 of individual models varies considerably. It is clear that IN05 index of credibility provided the best results in all previous ve years, followed by the Altman Z'score and IN99 index. The Taf er and Quicktest models had the worst predictive ability. An interesting observation is that there is no signi cant decrease of predictive ability of most models over the course of time. The four-year predictive ability for 2011 of individual models also varies substantially. In this case, IN05 provided the best results again, followed by Altman Z'-score and IN99 index. Again, the Taf er and Quicktest models had the worst predictive ability. The three-year predictive ability for 2010 provides similar results. Both credibility indexes together with the Altman Z'-score provided the best results. The results show that the IN05 model had the best overall accuracy for companies which went bankrupt in 2012 and 2011 and provided acceptable outcomes even ve years prior to going bankrupt. Among other models which had the highest accuracy, we can cite the IN99 model which is another credibility index and Altman Z'-score. These models seem to perform well in the Czech conditions. The Taf er model seems to provide the worst outcomes, which is consistent with the ndings of Ma asová (2007). The results may be explained by the fact that credibility indexes have been designed especially for the Czech environment and should therefore best re ect the particularities of it.

Managerial Implications
Managers and analysts often rely on popular bankruptcy models because of the fact that they are easily measurable and their interpretation is straightforward. However, the predictive ability of these models, in particular Taf er and Quicktest, seems to be limited. We can con rm the usefulness of the credibility indexes (IN99 and IN05) in predicting a company's distress in the Czech environment, which is consistent with the ndings of most authors, as well as the Altman Z'-score model. Moreover, it seems that the models do not signi cantly lose their predictive ability over the course of time; even ve years prior to going bankrupt, a model may provide a useful warning on the possible nancial distress of a company. However, all the analyzed modes are based on empirical analysis of historical data, without taking into account the present circumstances. Also, such models share the limitations of the accounting model, including the accounting concepts on which they are based.