Intelligent systems in accounting, finance and managementassessing predictive performance of ann-based classifiers, страница 14

Figure 2. The structure of the operating margin ratio

Table III. Descriptive statistics for the financial ratios

^a Kolmogorov–Smirnov Z.

outliers values (1 far outlier + 10 outliers) with u + 1.5d (= 43.62 for the OM ratio). We proceed likewise with all ratios.

In Table III we present the descriptive statistics, including the skewness, kurtosis and Kolmogorov–Smirnov normality test for the financial ratios of telecom companies. When the data are normally distributed, the values for both skewness and kurtosis are zero. A positive value for skewness indicates that the distribution has more values less than the mean and a long right tail; a negative value for skewness indicates that the distribution has more values greater than the mean and a long left tail. A negative value for kurtosis indicates flatness (flat centre, thin tails), and a positive kurtosis indicates peakedness (spiky centre, fat tails) (SPSS for Windows, Release 11.5.1). A skewness or kurtosis value greater than ±2.0 indicates that the distribution differs significantly from a normal distribution (SPSS for Windows, Release 11.5.1).

The skewness and kurtosis values fall into the range (−2, 2) of approximately normal distributions for all the variables. However, we encountered no zero values. Five financial ratios have positive skewness, which indicates a slight asymmetric distribution with tails extending more towards positive values (there are more companies performing below the sector average). Four financial ratios are somewhat ‘peaked’ (positive values for kurtosis), with operating margin being the most ‘peaked’ and the other three financial ratios have flatter centres and thinner tails when compared with the normal distribution. Results of the Kolmogorov–Smirnov test show that the normality assumption is rejected for all financial ratios at a significance level of α = 0.05. These results support the use of ANNs for financial analysis (e.g. financial classification models) over the traditional statistical methods, since neural networks are free of any distributional assumptions.

We used the real dataset with ‘levelled’ outliers and far outliers to generate the fictive datasets.

5.2.  Generating the Fictive Datasets