Determining Accuracy of IN05 Index and its Improvement

The IN05 index is probably the only one model in the world that predicts economic value added (EVA). Determining a positive or negative EVA is conducted based the level of return on equity (ROE). The threshold of the level depends on the changing value of implicit cost of equity (re). The Czech economy has undergone many changes including the economic crisis during the ten years following the creation of the IN05 index. This research uses financial data of 1,224 companies in the manufacturing sector. A very low current prediction power of IN05 was detected in prediction of prosperity. The main benefit of the research is finding a way to increase accuracy of economic value added prediction.


Introduction
Bankruptcy models are becoming more popular especially since 2008 when the whole world was hit by the global financial crisis. These models are designed to help analyze not only one's own company but they also allow business partners to recognize the financial situation of companies of interest.
However, the most important attribute of these models is the accuracy of their ability to predict bankruptcy or financial stability of a company in the future, i.e. ex ante in a few years, as stated by Castagnolo and Ferro (2014), Seppa (2014). A typical company, that is most keen to know the financial situation of its partner companies, is a bank. But for the company it is also very useful and instructive to compare with the companies operating in the same sector.
The models are very popular especially because they can be applied easily and provide unambiguously worded outcomes. The difference between the user attitudes to specific methods of determining the financial health of the company is very clearly handled by Mr. and Mrs. Neumaier (2008). They maintain that prosperity and bankruptcy models are more popular compared e. g. to pyramidal systems of indicators (alternative aggregate indicator). That's because they are considerably simpler and basic knowledge is sufficient for processing e. g. in Excel software. To interpret the results is also easier for users.
first bankruptcy model was made in 1968, described in Altman (1968). This model was designed for manufacturing companies in the USA and predicted bankruptcy for a period of up to three years in advance. One year ahead was the highest accuracy of prediction of bankruptcy amounting to 95%, two years by 72%, three-year forecasts with an accuracy of only 48%. For creation of Z score, Altman used discriminant analysis. Training set included 33 bankruptcy companies and 33 financial healthy also for companies not traded at stock exchange, as stated by Altman (2002). It was in 1983.
In the eighties indexes emerged for companies operating in Germany, Austria and Switzerland. One of these indexes was the Creditworthiness index. The Creditworthiness index classifies companies for creditworthy and bankruptcy. In these countries and also in the Czech Republic, the Creditworthiness index is well known and used. As Choi (2003) notes, even despite that it was validated only on test set of 21 companies. Another worldwide famous model called the Taffler index was created in the UK by Taffler (1982). It was published in 1977 and constructed through the use of linear discriminant analysis. To create the model, 46 bankrupting and 46 prosperous companies were investigated, specifically through financial statements in years 1969 to 1975. The author of the model determines the accuracy of the model as follows: correct identification of the bankruptcy with probability of 97% one year before bankruptcy prediction of a bankruptcy with 70% probability two years ago with probability of 61% three years ago The Neumaiers' created IN indexes. All of them constructed with data set of companies operating manufacturing industry in the Czech Republic. Overall they created four indexes. They called them IN95 index (in order to prevent bankruptcy), IN99 index (for prosperity prediction), IN01 and IN05 index are both used for bankruptcy and also prosperity prediction. The IN05 (2005) index is a modification of the IN01 index (2001). IN05 index is based on the analysis of 1526. Authors declare a success rate at 88% for economic value added prediction and probability of 77% for models were usually created differently, the structure of most of them is similar, as the combination of ratios and assigned weights of importance. However, IN indices for prosperity prediction (IN99 & IN05) were made on a sample of real companies, therefore they are exceptional.
called the P 'model. This model does not use classical statistical methods, but the methods for selecting the data used in data mining. The model was based on analysis of sample including 1560 Slovak companies of which 103 went bankrupt. Authors declare accuracy 78.24%.
based also on linear discriminant analysis with a data set of 1,508 companies operating in the manufacturing industry in the Czech Republic. There were 880 financially stable companies and 628 companies in bankrupt. The authors declare overall accuracy from 87.81% to 91.7% (on the base of weighted accuracy). The ability to determine creditworthiness reaches from 97.89% to 91.47% for the model. However the ability to predict bankruptcy is significantly lower from 69.91% to 61.18%.
The newest knowledge confirms that the time as well as branches are key for some with the sample of 34533 companies from manufacturing and construction branch we found that only 8 of 17 analysed predictors were significant in all of the analysed five year prior bankruptcy, 8 of them in the branch of manufacturing (i.e. EBITDA/TA, S/TA, S/St., EBIT/Int., EBITDA/Int., FA/TA, S/OR and S/CL.), the same set (with exception of S/TA) is also significant in the construction branch. As same predictors are time specific, or even not significant at all, this lead us to conclusion, that only some predictors could be used as early warning indicators or used to predict bankruptcy more years ahead. The presented accuracy is often debatable, mainly due to the fact, that: authors created the model for a small sample and thus the presented accuracy is not statistically based, authors presented the total accuracy based on the average of achieved accuracy in the sphere of bankruptcy prediction as well as prosperity prediction, authors confuse the accuracy and the rate of return, when comparing the accuracy of models the researchers do not regard the fact that the prosperity IN99 and IN05 (or IN01) is derived from EVA. authors take into account or make average of achieved accuracy in the sphere of bankruptcy or prosperity prediction, results of authors are not comparable, as they work with various ratios of bankrupting and financially sound companies. models and states that ratios of ensemble work with researches of specific economic di the bankruptcy moment that is related to the legislation of specific country.
The evaluation of predictive capability of bankruptcy models is usualy performed based on the comparison of current situation with the situation one year before. Thus the prosperity or bankruptcy predicted by the model (in the year X) versus the current situation of the following year (X + 1).
Then the accuracy of bankruptcy model prediction is (in relation to the table 1) defined in various ways: 1) as the conditioned probability j = TP j / (TP j + FP j and Neumaier (2005), Altman (2006), 2) as the probability (rate of return) p j = TP j / (TP j + FN j ) applied by Delina and 3) as the weighted average p j n j = (TP j + TN j ) / (TP j + FP j + TN j + FN j ) applied by 4) as the weighted average with subjectively defined weights for the purpose of 5) as the arithmetic average p j n j = ((TP j / (TP j + FN j )) + (TN j / (TN j + FN j )))/2  (2009) Where: TP j number of companies that went bankrupt and their bankruptcy was predicted by the model FP j number of companies that went bankrupt, but their financial health was predicted TN j number of companies that went not bankrupt and their financial health was also predicted FN j number of financially sound companies for which the bankruptcy was predicted The prosperity model accuracy is not shown in measurement of prosperity, except for one exception. Results of several researches are shown in the table 2. The following validation was focused on measurement of the capability to predict the bankruptcy using the prosperity model (Index of prosperity) and to predict the prosperity using the model (IN99) and the prosperity bankruptcy model (IN01 and IN05) (2015) emphasizes that IN99, IN01 and IN05 are specified by the fact that only those for which there is a presumption of economic value added creation are classified into the category of creditworthy companies, and thus the only presumption of positive and large industrial firms, so for these firms it will have the best expla That is why the current prediction capability of this model will be tested on middlefollowing form: IN05 = 0.13 X 1 + 0.04 X 2 + 3.97 X 3 + 0.21 X 4 + 0.09 X 5 (1) Where: X1 = total assets / liabilities X2 = earnings before interest and taxes / interest paid X3 = earnings before interest and taxes / total assets X4 = revenues / total assets X5 = current assets / current liabilities When IN05 is greater than 1.60 the firm produces a positive EVA value with a probability of 83%. When IN05 is smaller than 0.90, the firm gets in financial distress ("bankruptcy zone") within a year with a probability of 77% (for middle-sized The goal of this research is determining accuracy of IN05 Index in prediction of prosperity and Its Improvement.
1 Data and Methodology

Data and Calculation of IN05 Index
The sample for testing the prosperity prediction using the model IN05 was compound of 1224 companies from the Czech Republic, from the manufacturing industry, with turnover above 300 million that showed no negative symptoms 2012. This sample was compared with the prosperity level of analyzed companies in 2013. It means that the investigated period was one year. Data were drawn from the database of economic entities MagnusWeb of the company Bisnode.
The financial data from the balance sheet and profit and loss statement of chosen companies were applied in the formula (1).
Input values were defined as follows: earnings before interest and taxes revenues are the sum of all revenue categories I. up to XIII.,i.e. no. 1 + no. 4 + no. 19 + no. 26 + no. 28 + no. 31 + no. 33 + no. 37 + no. 39 + no. 42 + no. 44 + no. 46 + no. 53, current liabilities including short-term liabilities and short-term loans, i.e. line Resulting values of intermediate calculations were adapted as follows: within the calculation of interest coverage, i.e. components X2 the problems of zero division were found in about one quarter of cases. Resulting values X2 were limited at -9 in case of the negative EBIT value and +9 in case of positive EBIT values. there were also extremes with resulting values X2 directed towards +/due to low values of cost interests approaching the zero. These extremes were reduced up to the interval <-9; +9>. Such reduction was performed in about one third of cases.

Procedure of Calculation of Economic Value Added EVA
If the model IN05 in the year X (in this research the year 2012) achieves for a specific company a value higher than 1.6, the company should reach the positive EVA value in the following year X+1 (in this research the year 2013).
The abbreviation EVA refers to the so-called economic value added. Officially, the historical beginning of this indicator is considered the year of 1982. At that time Joel Stern founded the Stern Stewart & Co. and began to use it for the performance measurement of companies. As the unofficial beginning of this indicator can be considered earlier in period of 90s of the 19th century when Alfred Marshall mentioned economic and accounting profit as the first as stated by Kyriazis and Anastassis (2007).
We rank the EVA indicator among the so called value indicators. Another well-known example is the indicator of marked value added (MVA). The task of the EVA is to express the value of economic profit. Economic profit includes cost of equity, it means we must take opportunity costs into account.
Economic value added represents part of company profit generated over the expectations of investors. This money can be employed to invest in company development as stressed by Tortella and Brusco (2003), Kangarloei et al. (2012).
There are several options how to calculate the value of EVA. Let's introduce the main ones. E.g. Vochozka and Stehel (2013) express it in this way: Another way of EVA calculation is using the indicator NOPAT (Net Operating Profit After Taxes). Then the formula looks as follows according to Tortella and Brusco (2003): In the opinion of the author the most objective accuracy measurement, for the sphere of prosperity, is the calculation based on the approach stated in column h.) tab. 4. This one takes into account the whole correct predictions and compares it with the whole tested ensemble of companies, including those that are within the grey zone. Based on this index the capability of model IN05 to predict the positive EVA achieves 49,26%. If we do not take into account the uncertainty area (grey zone), the prediction capability grows up to 75.38%. In order to increase the accuracy, the tests were performed when original evaluation scale limits <0.90;1.60> were shifted. According to the methodology g.) tab. 5 the accuracy was growing with lowered upper limit. The highest accuracy of 75.50% was achieved when the upper limit was lowered at the value 1.4. At the same time the model was not able to include 26.63% of companies into the category +EVA or EVA and incorrectly included 44.61% of companies from the tested sample of 1224 companies. Table 5 Results of IN05 Application with Moved Limits, without Grey Zone Upper limit 1.6 1.5 1.4 1.3 1.2 1.1 1.0 0.9 0.8 0.7 0.6 0.5 Lower limit 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.8 0.7 0.6 0.5 Source: Author Relation between threshold (limit) and model accuracy is seen also in graphical form below. There is evident that to increase the accuracy we need to raise the grey zone. E.g. accuracy 89.37% (type f. with limits 0.9/1.6)) is not the maximum achievable value. This accuracy can be even more increased. But with limits 0.9/1.6 there is connected also higher grey zone as seen on the graph. Author does not recognize the maximization of accuracy through the existence of the grey zone.
Grey zone

Source: Author
The methodology can not be used to find the maximum accuracy. This is because it ignores good prediction of negative EVA (failed prediction of negative EVA). The methodology g seems to be appropriate because it reflects the ability to predict both positive and negative economic value added. However, the result can be misleading due to the fact that a large portion of the analyzed sample is not classified (companies included in grey zone). As the most suitable methodology seems methodology h. The methodology works with the whole sample and the total accuracy includes the whole sample, both classified and unclassified companies.
is the most suitable to move limits at the value 0.7 and to eliminate the grey zone at the same time. The model accuracy, in predicting the prosperity according to the methodology h.), achieves the maximum accuracy amounting to 74.35% with zero grey zone and 25.65% of error rate from the complete tested sample. i. j.

Conclusions
It is not easy to assess the financial stability of business partners or debtors and/or to conduct it on the basis of publicly available data in a speedy, simple and precise manner. It is obvious that these diagnostic/predictive models always operate with some degree of inaccuracy. The question is, what can the maximum achievable value of accuracy be?
Finite precision is affected by many factors. Especially in the phase of model creation these factors are size and homogeneity of the training set, data quality, methodology for data processing and technique of model creation. Following this phase of accuracy testing the key factors are size and homogeneity of the validation set, data quality (accounting provides a fair view), process for treating of data, as well as the methodology chosen for the measurement of total model accuracy.
The aim of this research was to define the accuracy of the IN05 model after the period of 9 years from its creation. This prosperity-bankruptcy model was, as one of the few, created based on real data using the multiple discrimination analysis. The accuracy was derived from the capability of the model to predict the economic value added in the following year. This enables to verify its up-to-date condition. The results of calculations of the tested sample showed, with statistical significance, that the accuracy of prosperity prediction has increased from 83 to 89.37%, when the accuracy is calculated based on TP j / (TP j + FP j ). At the same time, 50.74% of companies (from the whole sample 1224) were classified incorrectly.
The testing of the threshold of <0.90;1.6> shift up to the unified critical limit at 0.7 resulted in the best result of the prediction capability of this model, based on (TP j + TN j ) / (TP j + FP j + TN j + FN j + GZ). The achieved accuracy of 74.35% is the ratio of all correctly classified companies related to the size of the whole sample (1224). The conditioned probability of the correct classification +EVA (based on TP j / (TP j + FP j ) is 73.63%. In order to reach a quick classification of the company, using the IN05 idex, aiming to define its prosperity, the author recommends using the newly defined limit of 0.7.
In addition, the research conducted confirmed that the declared accuracy of models can be considerably different depending on the methodology of accuracy quantification (tab. 4 columns from e.) to h.)). In conclusion, it shall be pointed out that the research conducted confirmed that the accuracy of models can change with time, especially if their accuracy evaluation is based on spread value connected with variable r e value. For example, the mean of r e and ROE changed as follows in these years: r e 10.58% in 2007, 11.14% in 2008, 15.36% in 2009, 13.52% in 2010, 13.66% in 2011, 8.40% in 2012, 7.42% in 2013, ROE 15.11% in 2007, 10.85% in 2008, 6.45% in 2009, 11.61% in 2010, 11.52% in 2011, 21.88% in 2012.33% in 2013.