Research Process

This chapter examines the second research question of whether Altman’s models can adequately predict bank financial unsoundness.

Figure 4.1: Research Process

Source: Author using draw.io

In the first step of the research process Altman’s Z (1993) and EM Score (1995) models were selected. These two models are presented in Table 4.3.

Te

s

ting proce

s

s

Re

-e

s

tima

tion proc

e

s

s

Table 4.3 Chosen Altman Models

Function Variables Application Area Cut-of Scores

Z(1993) = 6.56Х1 +

3.26Х2 + 6.72Х3 +

1.05Х4

Х1 – Working Capital / Total Assets;

Х2 – Retained Earnings / Total Assets;

Х3 – EBIT / Total Assets;

Х4 – Book Value Equity / Book Value

of Total Liabilities Non- manufacturing companies Bankrupt<1.1< Grey>2.6> Safe EM Score (1995) = 6.56Х1 + 3.26Х2 + 6.72Х3 + 1.05Х4 + 3.25

Х1 – Working Capital / Total Assets;

Х2 – Retained Earnings / Total Assets;

Х3 – EBIT / Total Assets;

Х4 – Book Value Equity / Book Value

of Total Liabilities Non- manufacturing companies in Emerging Markets Bankrupt<1.1< Grey>2.6> Safe Source: Altman (2000)

All four Altman variables were selected and analysed in Section 3.4.

Secondly, these models were tested on annual data of 12 Kazakhstan banks during the period from 1st _{January, 2008 to 1}st _{January, 2014 (Appendix 4A).}

Thirdly, in the previous chapter two groups of sound and unsound banks were obtained using a cluster-based methodology of financial soundness assessment. The financially unsound banks are BTA Bank, Kazkommertsbank, ATF Bank, Alliance Bank, Temirbank and Nurbank. The government acquired 75% of BTA bank’s equity and 20% of Kazkommertsbank’s equity. Equity of the Alliance bank was sold by the owners for $1 to the government. The ATF Bank, Temirbank and Nurbank were unable to meet their scheduled payments. In 2015 the BTA Bank merged with Kazkommertsbank and the Alliance Bank and Temirbank merged with Forte bank. Thus, the unsoundness of 6 banks obtained by cluster analysis on 1st _{January 2014 was} established.

The 6 sound banks were selected from a group of financially sound banks based on asset size, specialization and branch network. Thus the sample is composed of 12 banks with a share of assets in the total assets of the banking sector at 81.3% (Table 4.4). These 12 banks represent almost the entire banking sector of Kazakhstan. 84 observations from annual financial reports are used in the analysis.

Table 4.4: Selected Sample of Banks and Asset Share, 1st _{January, 2014}

№ Unsound Bank Share in

Assets of Banking Sector, %

Ranking Sound Bank Share in

Assets of Banking Sector, %

1 Kazkommertsbank 16.2 1 Halyk Bank of

Kazakhstan

15.8

2 BTA Bank 9.8 2 Bank Centercredit 6.9

3 ATF Bank 5.8 3 SB Sberbank 6.7

4 Alliance Bank 3.6 4 Tsesnabank 6.0

5 Temirbank 2.0 5 Kaspi Bank 5.5

6 Nurbank 1.6 6 Bank RBK 1.4

Total 39 Total 42.3

Total of two groups 81.3

Source: Author

The small sample is typical for studies on data on an individual country. In the previous section journal articles with small samples were studied. For example, Othman (2013) investigates 13 Malaysian Islamic and 10 conventional banks; Rankov and Kotlica (2013) examined 10 Serbian banks; Chieng (2013) analyses 4 distressed and 4 control banks from the Eurozone; Pradhan (2014) examines 3 Indian banks. When a sample is small it is impossible to divide it into ‘training’ and ‘holdout’ types. Altman (1995a) noted that, in the case of a lack of observations, it is not possible to test the model on a new meaningful 'holdout' group. Bellovary et al. (2007) reviews bankruptcy prediction studies from 1930 to 2007 and notes that roughly less than half of the studies use hold-out sample.

In this study there is not enough data to allow for testing. That is why a leave-one-out classification is used as a form of cross-validation of the classification table. Under this approach, a discriminant function based on all cases except the selected example is used to classify this case (Nasledov, 2013).

Fourth, as seen from Table 4.3, these two models differ by a constant at 3.25, with the same variables and cut off points. To assess the probability of bankruptcy for both Altman’s Z” and EM Score models, cut off points are proposed where a value of less than 1.1 gives a high probability of bankruptcy; 1.10 to 2.6 is ‘grey zone’ and gives a distress situation; and a value equal to or more than 2.6 gives a low probability of bankruptcy. This study, as was mentioned above, focused on bank financial unsoundness, which is the earlier step of distress and not bankruptcy (Section 4.2.1). Therefore, a value less than 2.6 classifies a bank as unsound and a value higher than 2.6 will rate a bank as sound. A ‘grey zone’ will be clearly interpreted as unsound.

In order to improve the predictability of Altman’s models, a technique from Wu et al. (2010) is adapted. The obtained Z (1993) and EM Scores (1995) are ranked from lowest to highest. It is assumed that the optimal cutoff point is between 25 and 95 percentiles. The predictability and the Type I and II errors are calculated with the step at 5 percentile within this range. Close to the segment with highest values calculations are made with the step at 1 percentile. A new cutoff point is set for the percentile at which the sum of Types I and II classification errors is minimized.

Finally, Altman’s Z model was re-estimated as both the Z (1993) and EM Scores (1995) models consist of four similar variables and differ from each other only by a constant 3.25. The process of re-estimating Altman model is designed according to that of Moyer (1977). The Direct approach includes in the discriminant function each of the four variables specified by Altman. The Wilks’ approach enters variables into the function in a stepwise manner up to the point where the Wilks' lambda is minimized. Both approaches will be used to compare their abilities to assess financial unsoundness of banks.

The significance of the re-estimated models is determined by the Wilks’ Lambda, the Chi-square and by the statistical significance. The closer is the Wilks’ Lambda value to 1, the superior is the model’s quality. The Chi-square measure defines the power at which the discriminant function distinguishes between groups. The higher is the value, the greater can the discriminant function distinguishes between groups and the more effectively it fulfills its intended use. Its consistency can be judged by the statistical significance which must be less than 0.05.

In the process of re-estimation the two models of ZD and ZW were obtained. For each re- estimated model new cut off points and predictive accuracy were calculated. Othman (2013) noted that the optimum cut-off score is approximately equal to zero and is the weighted average of the discriminant score of the sound and unsound bank groups. If the discriminant score is less than the cut-off score, the bank is classified as unsound and, if the discriminant score is more than the cut-off point, the bank is classified as sound.