You are on page 1of 7

BANKING SECTOR PROFITABILITY ANALYSIS: DECISION TREE

APPROACH
Mirjana Peji Bach
University of Zagreb
Faculty of Economics
Trg J.F.Kennedy 6
Zagreb
e-mail: mpejic@efzg.hr
Ksenija Dumii
University of Zagreb
Faculty of Economics
Trg J.F.Kennedy 6
Zagreb
e-mail: kdumicic@efzg.hr
Nataa arlija
University of Osijek
Faculty of Economics
Gajev trg 7
Osijek
e-mail: natasa@efos.hr

Abstract The paper deals with problem of analyzing the profitability of the banking sector in
Croatia. In our research, profitability is measured by the return on average assets (ROAA). The aim
of the paper is to design a model which would forecast the profitability of banks by their
characteristics and the environment factors in order to maintain the stability of the banking sector.
The decision tree has been developed using C&RT algorithm. The results have shown that ratio of
capital and assets, market share and loan to assets ratio have the positive influence on the profitability
of the banks.
Key words: profitability of the banks, forecasting profitability, decision tree

1. Introduction
Current banking sector profitability analyses have been targeted to forecast bankruptcy of the
bank and authors have used various methods. Barr, Seiford and Thomas (1994) tried to
predict bankrupts using a non-parametric frontier estimation approach.. Lane, Looney and
Wansley (1998) used the Cox model, and other researchers used neural networks (Tam et.al.,
1992; Salchenberger et.al., 1992). These studies are based on the classification approach,
according to which in the past banks have been classified as bankrupted or not. On the other
side, Nuxoll (2003) proposes the benchmarking approach, which is based on the preposition
that best results are achieved if banks follow the financial structure of the best banks, or in
other words by benchmarking best banks.
The goal of this study is to design a model which would forecast the profitability of
banks by their characteristics and the environment factors. All this is used to maintain the
stability of the banking sector. A forecasting model like this would be of great use to the
Croatian National Bank, as to all the Boards of Directors. The study consists of the following
parts. In the second part various ways of banking sector profitability analysis are enlisted.
The third part describes the methodology used in the study (the decision tree). The fourth
part encloses the results and the fifth part comprises final thoughts.

2. Banking sector profitability analysis


Banking sector profitability is measured by the return on average assets (ROAA),
return on average equity (ROAE) and the net interest margin (NIM). Based on these
profitability indicators, recommendations to the boards of directors can be made. In this
study we will try to express profitability with one value, which follows the duPont procedure
of business activity estimation (Pavkovi, 2004).
In this study we will concentrate only on the return on average assets (ROAA), which
is calculated as a ratio of profit and average assets. Hence, it is the banking profit gained for
one Kuna (local currency) of assets.
Factors of banking sector profitability can be divided to two basic groups:
characteristics of a specific bank and environment factors, and the selected profitability
factors were used in a research by Demirguc- Kunt and Levine (2001). Characteristics
specific for a bank are: market share, ratio of capital and assets of the bank, ratio of loans
and assets of the bank, ratio of overheads and assets of the bank and the ratio of non-income
assets and total assets of the bank.
The market share should have a positive effect on banking sector profitability
indicators. Different hypotheses on the functioning of the market in various ways explain
this fact. According to the relative market power hypothesis only monopolistic companies
with high market shares and highly differentiated products can acquire above average profit
margins. The efficient structure hypothesis claims that banks with effective asset structure
achieve highest market shares. Berger (1995) tested these two hypotheses in the financial
market and proved that the size of the bank is connected with profitability, which was also
proved by Frame and Kamerschen (1997). On the other side, Smirlock (1985) shows that
concentration isn't prior connected to superior performance of the leading banks, but the
efficient banks become bigger and gain bigger market shares.
Share of capital in the assets is positively correlated with ROAA. Banks with high
shares of capital in overall assets have lower costs of financing which effects higher
profitability and lower probability of bankruptcy.
Ratio of loans and assets is also positively correlated with profitability indicators. A
bank which approves more loans for a unit of assets with the same interest rate, acquires
higher profit because it earns more on the interest rates. Let us just emphasize that the
growth of profit is not proportional to the growth of approved loans if this is too risky.
The share of non-income assets in the bank assets is negatively correlated with
profitability indicators, although there are exceptions. For example, a bank can transfer the
costs of its non-income assets to its clients, and a bank that pays rent for real estate can have
higher costs than the bank that has its own facility.
The ratio of overheads and bank assets is negatively correlated with profitability
indicators.
The values of these indicators for banks from the sample are shown in Table 1. The
average market share of banks has been decreasing in the past five years. The ratio of capital
and assets of the bank has also been decreasing, but it is still high. That means that banks
have been decreasing the share of capital in the assets, but are still very cautious because of
the suspicion in the stability of the banking sector. The ratio of loans and banks assets is
increasing. The ratio of non-income assets and bank assets and the ratio of overheads and
bank assets do not show a visible trend.

Table 1. Bank activity indicators


Year

Market
share

1999
3,45%
2000
3,45%
2001
3,13%
2002
3,33%
2003
3,13%
Source: The Scope

Ratio of capital Ratio of loans Ratio of non- Ratio of


and assets of the and assets of the income
assets overheads and
bank
bank
and assets of the assets of the bank
bank
53,96%
24,29%
1,86%
2,68%
23,01%
50,97%
2,76%
5,00%
17,37%
50,41%
2,03%
4,15%
15,24%
55,52%
2,29%
3,74%
15,44%
57,65%
2,02%
4,19%

Environmental factors are: GDP real growth rate, inflation rate, average exchange
rate, GDP per capita. Web pages of the Croatian National Bank were used as a source of data
about the macroeconomic indicators (www.hnb.hr). Values of the indicators are shown in
Table 2.
Table 2. Characteristics of the environment as a factor of profitability
Inflation rate
1999 4 %
2000 4,6 %
2001 3,8 %
2002 1,7 %
2003 1,8 %
Source: www.hnb.hr

GDP per capita (EUR)


4102
4560
4998
5451
5747

Growth of GDP
-0,9 %
2,9 %
4,4 %
5,2 %
4,3 %

Exchange rate HRK: EUR


7,5796
7,635
7,469
7,4068
7,5634

The growth of GDP should have a positive effect on the profitability of the bank. The
inflation rate can have a positive and negative effect on the profitability, depending on the
capability of the management of the bank to effectively conduct the resources of the bank
during inflation. Finally, the exchange rate should be negatively correlated with the
profitability of the bank, which is explained by the following. In the case of a strong HRK,
Croatian companies are less competitive on the world market, which decreases the GDP and
this way has a negative impact on profitability.
3. The decision tree
The decision tree can be used for classification and regression problems, and unlike
neuron networks, the decision tree generates a model which can explain the mutual influence
of input and output variables by a set of rules. The generated rules can be expressed like
SQL commands and can simply be built in to the program solution.
For a problem to be appropriate for solving it with a decision tree, it has to have the
following characteristics (Mitchell , 1997): (1) The data has to be described in a form of a
final number of attributes, for example there are attributes for every bank; (2) The number of
attributes is known in advance, for example it is well known how many attributes one bank
can have; and (3) Every part of data should belong to only one category.

The decision tree is a classification algorithm which has a structure of a tree


(McLahlan, 1992). There are two types of nodes connected with branches: leaf node which is
the end of a particular branch, and the decision node which defines a certain condition in a
form of a value described with attributes. It is made by searching for patterns with the
algorithm , and the most famous algorithm are Chaid, exhaustive Chaid, C&RT (Breiman
et.al, 1984) and Quest (Loh et.al., 1997).
The algorithm is made of a selection of attributes for generating the decision nodes,
with all data sorted to one group in the beginning. Data are then divided to branches
according to all possible criteria, and the criterion chosen is the one that divides data to
groups that are more homogenous that the initial group of data. When the data can no longer
be divided into groups that are more homogenous than the initial data, the tree is finished.
Entropy is used as a measure of data group homogenousity.
4. Results
In order to forecast profitability of the banks C&RT and CHAID decision trees have
been developed. All methods are processed with StatSoft Statistica 7.1. Results of C&RT
algorithm is shown here as it gave better model.
Accuracy of the prediction is analyzed. Measures that are usually used are mean
absolute deviation (MAD) and root mean-squared error (RMSE). The lower prognostic
errors mean the higher accuracy of the model. According to both criteria C&RT has been
shown as method which generates the lowest errors (Table 3). This method can be shown in
a form of SQL statements which enables what-if scenario where the aim is to analyze what
could happen if the characteristics of the banks and the environment factors are changed.
Figure 1 shows the structure of the decision tree.
Table 3. Measures of accuracy prediction for the return on average assets (ROAA) for C&RT
and CHAID
Measures of accuracy prediction
MAD
RMSE

ROAA
C&RT CHAID
0,74
0,85
1,03
1,18

Figure 1. The decision tree for the return on average assets (ROAA)
Variables used for splitting nodes are: ratio of non-income assets and total assets,
ratio of capital and assets, market share and loan to total assets ratio. The banks are divided
into the 6 groups shown in table 4.
Table 4. Leaf nodes of the decision tree for the return of average assets (ROAA)
Node number
Average ROAA
Split criteria
3
4,688000
Ratio of non-income assets and total assets
higher than 4,54512
11
2,092586
1 - Ratio of non-income assets and total assets
lower than 4,5412; 2 - ratio of loan to assets
higher than 16,3300
10
1,162308
1- Ratio of non-income assets and total assets
lower than 4,5412; 2 - ratio of loan to assets
lower than od 16,3300
6
0,641538
1- Ratio of non-income assets and total assets
lower than 4,5412; 2 market share lower than
0,7215
8
1,162632
1- Ratio of non-income assets and total assets
lower than 4,5412; 2 market share higher than
0,7215; 3 Ratio of non-income assets and total
assets lower than 1,5485
9
1,641613
1- Ratio of non-income assets and total assets
lower than lower than 4,5412; 2 market share
higher than 0,7215; 3 Ratio of non-income
assets and total assets higher than 1,5485
On the basis of the results of the decision tree, the following results can be made:
Profitability of the banks is positively influenced by ratio of capital and assets of the banks,
market share and loan to assets ratio.

Detailed analysis has shown that banks with the ratio of non-income assets and total
assets lower than 4,54 and ratio of capital and assets higher than 17,54 will have higher
profitability than the banks with the similar ratio of non-income assets and total assets and
lower ratio of capital and assets.
Banks with the ratio of non-income assets and total assets higher than 4,54 have
higher profitability compared to the banks with the ratio lower than 4,54.
Banks with the market share higher than 0,72 with the ratio of non-income assets and
total assets lower than 4,54 and with the ratio of capital and assets lower than 17,54 will be
more profitable compared to the banks with lower market share and similar values of all
other mentioned ratios. This confirms the previous researches which state that banks
profitability is highly influenced by market share (Berger 2005, Frame and Kamerschen
2007).
Banks that belong to the same group according to the ratio of non-income assets and
total assets and ratio of capital and assets will have different profitability due to the loan to
assets ratio in a way that higher profitability will be accomplished by the banks with the
higher value of loan to assets ratio (Bourke, 1989).
5. Conclusion
The aim of the paper is to design a model which would forecast the profitability of
banks by their characteristics and the environment factors in order to maintain the stability of
the banking sector. In our research profitability is measured by the return on average assets
(ROAA) as the ratio of net income and average total assets. Data for this research consisted
of data about the banks in Croatia over the period from 1999 to 2003. Also, methodology of
decision tree is given with the results of the decision tree model (C&RT) for the banks in
Croatia. Results have shown that profitability of the banks is positively influenced by ratio of
capital and assets, market share and loan to assets ratio. Particularly, of the banks with the
similar ratio of non-income assets and total assets higher profitability is accomplished by the
banks with higher ratio of capital and assets. Further, if the banks belong to the group of
those with similar values of ratio of non-income assets to total assets and ratio of capital and
assets, profitability will be increased by higher value of market share as well as higher loan
to assets ratio. Although it was expected that lower value of non-income assets to total assets
ratio would increase the profitability, the case of Croatian banks has shown opposite
influence. An explanation could be found in the fact that banks in Croatia realized their
profitability on income from services and less on income stated in assets. In order to
investigate this phenomenon it would be interesting to analyze income structure of the banks
as well as non-income assets which we suggest as guidelines for further research.
6. References
1. Barr, R., L. M. Seiford and F.Thomas., 1994. Forecasting Bank Failure: a nonparametric frontier estimation approach. Recherches Economiques de Louvain
60(4), 417-429.
2. Berger, A., 1995. The relationship between capital and earnings in banking.
Journal of Money, Credit and Banking, 27, 404-431.
3. Bourke, P., 1989., Concentration and other determinants of bank
profitability in Europe, North America and Australia. Journal of
Banking and Finance 13, 65-79.

4. Breiman, L., Friedman, J. H., Olshen, R. A. and Stone, C. J., 1984. Classification
and Regression Trees. Belmont: Wadsworth.
5. Demirguc-Kunt, A. and Levine, R., 2001. Financial Structure and Bank
Profitability in Financial Structure and Economic Growth: A Cross-Country
Comparison of Banks, Markets, and Development, Eds. Cambridge, MA: MIT Press.
6. Frame, W. S., and D. R. Kamerschen. 1997. The Profit-Structure Relationship in
Legally Protected Banking Markets Using Efficiency Measures. Review of
Industrial Organization, 12, 9-22.
7. Lane, W. R., S. W. Looney and J. W. Wansley., 1986. An Application of The Cox
Proportional Hazards Model to Bank Failure. Journal of Banking and Finance. 10,
511-531.
8. Loh W. Y. and Shih Y. S., 1997. Split Selection Methods for Classification trees.
Statistica Sinica 7, 815-840.
9. Han, J., and Kamber, K., 2000. Data Mining: Concepts and Techniques. San
Francisco: Morgan Kaufman.
10.
McLachlan, G. J., 1992. Discriminant Analysis and Statistical Pattern
Recognition. New York: Wiley Interscience.
11.
Mitchell, T., 1997. Decision Trees, in T. Mitchell. Machine Learning,
London: McGraw-Hill.
12.
Nuxoll, D.A., O'Keefe, J., and Samolyk, K., 2003. Do Local Economic data
improve off-site bank-monitoring model?. FDIC Banking Review, 15(2), 39-53.
13.
Pavkovi, A., 2004. Instrumenti vrednovanja uspjenosti poslovnih banaka.
Zbornik radova Ekonomskog fakulteta u Zagrebu, 2(1), 179-191.
14.
Salchenberger, L. M.; Cinar, E. M.; and Lash, N. A., 1992. Neural networks:
A new tool for predicting thrift failures. Decision Sciences, 23(4), 899-916.
15.
Smirlock, M., 1985. Evidence on the (Non) Relationship between
Concentration and profitability in banking. Journal of Money, Credit and Banking,
17(1), 69-83.
16.
Tam, K.Y. and Kiang, M.Y., 1992. Managerial Applications of Neural
Networks: The Case of Bank Failure Predictions, Management Science, 38, 926947.