Financial Benchmarking of

Telecommunications Companies

Jonas Karlsson
bo Akademi University, Department of Information Systems,
Lemminkisenkatu 14, FIN-20520 Turku, Finland
E-mail: Jonas.Karlsson@abo.fi
Barbro Back
bo Akademi University, Department of Information Systems,
Lemminkisenkatu 14, FIN-20520 Turku, Finland
E-mail: Barbro.Back@abo.fi
Hannu Vanharanta
Pori School of Technology and Economics,
Pohjoisranta 11, FIN-28100 Pori, Finland
E-mail: Hannu.Vanharanta@pori.tut.fi
Ari Visa
Tampere University of Technology, Department of
Information Technology,
Korkeakoulunkatu 10, FIN-33100 Tampere, Finland
E-mail: Ari.Visa@tut.fi


Turku Centre for Computer Science
TUCS Technical Report No 395
February 2001
ISBN 952-12-0788-4
ISSN 1239-1891

Abstract

The aim of this paper is to evaluate the financial performance of
telecommunications companies with the help self-organising maps. A total
of 88 companies is evaluated and benchmarked on the basis of seven key
ratios, calculated for five consecutive years 1995-99. To conduct
benchmarking with several financial key figures in a changing business
environment is difficult and time consuming for todays managers and there
exists a need for an easy and quick tool to accomplish this task.
Furthermore, financial analysts and executives need efficient tools for their
work to make comparisons and to trace company performance over time. In
this study, we have collected financial data from the Internet and used self-
organising maps, a form of neural networks, to analyse telecommunication
companies financial performance. The self-organising map method turned
out to be a good analysing, clustering and visualising tool for depicting
patterns and company movements over the years, within the business line of
telecommunications companies.


Keywords: Financial benchmarking, Self-organising maps, Neural
networks, Financial performance, Telecommunication
















TUCS Research Group
Computational Intelligence in Business


1

1. Introduction

At the same rate as new products are launched onto the market and new competitors
arise, there is an increasing pressure on the companies to be competitive enough to stay
alive. There is a need to develop high-quality products, implement efficient working
methods, and to have a clear hunch about who the major competitors are, what they are
good at, and how they are doing financially. Furthermore, other stakeholders such as
investors and financial analysts have a vast interest in how companies in the same line
of business are doing financially. To be able to understand this, it is important to have
efficient tools. Benchmarking is one way to accomplish necessary improvements in the
companys activities by measuring itself against other companies. [Bendell et al. 1998,
66]

Benchmarking can be divided into qualitative and quantitative benchmarking.
Qualitative benchmarking involves to measure functions and methods of one company
against other companies. Quantitative benchmarking, on the other hand, involves
benchmarking the performance and profitability of one company against others.
Quantitative benchmarking can also be referred to as financial benchmarking. Financial
benchmarking is usually conducted by analysing the financial statements or financial
key ratios of the selected companies. [Karlf 1997, 40-41]

One problem that is encountered immediately is, however, that within a database there
can be very much information, both relevant and irrelevant. How should the collection
be done in order to gather only relevant information? To manually browse through, and
check the database, is cumbersome work that could take a great amount of time to
accomplish. Furthermore, there is a risk of missing relevant information or possible
hidden patterns. [Stein, 1993]

By using ordinary spreadsheet programs one can easily compare two to six companies
at a time according to one ratio at a time. However, if one wants to obtain an overview
of the competitors on the market, or want to take into account several ratios at the same
time, spreadsheet programs are no longer of any use. Neural networks, in the form of
self-organising maps, provide a new tool for clustering and visualisation of large
amounts of information. This technique analyses the different characteristics of the
input, and groups samples with similar characteristics together. In this report, we do a
financial benchmark for telecommunications companies using self-organising maps.
Thus, in the financial comparison conducted in this report the self-organising map will
analyse selected financial key ratios of companies, grouping companies with similar
financial performance together.

The term self-organising map has become a very popular topic in todays information
technology society. Since its invention in 1981 over 4300 research papers have been
written on the subject of self-organising maps [Kohonen, 2000]. Some examples of
more resent research papers include cloud classification [Ambroise et al., 2000], image
object classification [Becanovic, 2000], breast cancer diagnosis [Chen et al., 2000],
classifying and clustering Internet traffic [Raivio et al., 2000], and extracting knowledge
from text documents [Visa et al., 2000]. Generally these reports can be divided into two
2
groups, analyses and surveys. The analyses take a more technological approach towards
the algorithm and function of the self-organising map, while surveys take the practical
application of the self-organising map into consideration [Kaski, 1998]. There are a
very small number of reports concentrating on financial analysis and large databases
available at present.

An example of a financial survey is the study conducted by Back et al. (1998). In this
study self-organising maps were used to benchmark and analyse 120 companies in the
pulp and paper market. The main objective of the study was to determine whether the
self-organising map could be used for managing complexity in large databases. The
result of the study indicates that there are benefits from using the self-organising map as
a tool for organising large, standardised databases, and also to identify clusters of
companies with similar characteristics.

Not many financial analyses of the telecommunications market have been conducted
using the self-organising map in the same fashion as the study of pulp and paper
companies. In this way, this report takes a step into a newer and untried field,
furthermore, this study is more concentrated at analysing the companies and their
movement over the years, instead of evaluating whether self-organising maps are a
suitable tool for this kind of comparisons. One should also keep in mind that the
companies in the pulp and paper market are often well-established, 100-year old
companies with a permanent customer base. In the telecommunications sector, most of
the companies are new actors on the market (under 10 years old), often with poor
profitability due to heavy investments needed in order to be competitive. This report is
based on the findings of Karlsson (2000).

The rest of the paper is organised as follows: Section 2 describes the methodology and
the choice of financial ratios and companies. Section 3 presents the construction of the
self-organising maps and Section 4 presents a detailed analysis of the maps. The
conclusions of this paper are presented in Section 5.

3
2. Methodology

In this section we provide a description of the self-organising map, introduce the
companies included in the study as well as describe the choice of the financial key
ratios.

2.1 The Data Mining Tool

The self-organising map technique is a form of a neural network that can be used as a
clustering and visualisation tool. The self-organising map technique creates a two-
dimensional map from the input data. This map resembles a landscape where it is
possible to identify borders that define different clusters. These clusters consist of input
variables with similar characteristics, i.e. in this report they consist of companies with
similar financial performance. [Kohonen 1997, VIII]

With every node i on the map, a parametric reference vector m
i
is associated. The nodes
in the array represent the neurons of the neural network. Data vectors x, representing the
high dimensional input data space, are mapped onto the two-dimensional array. Vector
x can be compared to every m
i
by, for example, using the smallest Euclidian distance

x in order to define the best-matching node. [Kohonen 1997, 86]:

} { ,

x x = [2.1]

where the index c refers to the winning unit. The self-organising map algorithm consists
of two steps that are iterated for every sample:

Step 1: When a sufficient number of input vectors x have been presented, and the
reference vector defined, the calibration of the map begins. This is done by
finding where the best matching neurons on the map lie, and labelling the map
units correspondingly [Kohonen 1997, 90]. For each input vector used in the
training of the self-organising map the best matching neuron c in the map is
located. Usually, the best matching neuron is defined by the distance function
used, in most cases the earlier described Euclidian distance. In other words, the
best matching neuron is the neuron with the weight vector m
i
situated closest to
the input vector. [Kohonen 1997, cited in Kangas 1994, 15]

Step 2: During the learning process those nodes will activate each other to learn
something from the same input x. This will result in a local smoothing effect on
the weight vectors of neurons in this neighbourhood. [Kohonen 1997, 87]

The process can be described as following: first the input vectors, x
i
are
mapped randomly onto the two-dimensional array, the map. The input vector is
compared to all reference vectors m
i
, in order to find the best-matching node, c
(Step 1). The best-matching neuron affects the neurons within the
neighbourhood area h
ci
, i.e. they learn from the best-matching neuron (Step 2).
4

The weight vector values m
i
are tuned so that the match of the modified weight
vectors in the neighbourhood of the best matching neuron and the input vector
x is improved [Kangas 1994, 15]. When using the Euclidian distance function,
the adaptation process has the following general form [Kohonen 1997, 87]:

[ ], ) ( ) ( ) ( ) ( ) 1 ( x

+ = + [2.2]

where t = 0, 1, 2, is an integer, the discrete-time coordinate. The function
h
ci
(t) is the neighbourhood of the winning neuron c, and acts as a smoothing
kernel defined over the lattice points. [Kohonen 1997, 87]

The neighbourhood function h
ci
(t) defines the activity of neighbouring neurons.
Usually the value of the neighbourhood function depends only on the distance
between the neurons and time. The neighbourhood function can be defined as:

,
) ( 2
x ) ( ) (
2
2

[2.3]

where (t) and ) ( are suitable decreasing functions of time, r
i
and r
c
are the
coordinates of units i respective c. [Kohonen 1997, cited in Kangas 1994, 15]

These two steps are iterated and cycled through each input vector until a certain
stopping criterion is reached. The stopping criterion is usually a pre-defined number of
iterations.

The software we will use when training and creating the self-organising map is called
The Self-Organising Map Program Package Version 3.1 (SOM_PAK), and is based on
the Kohonen self-organising algorithm. The software package has been developed to
demonstrate the implementation of the self-organising algorithm, and was written by the
SOM programming team at the Helsinki University of Technology. The SOM_PAK and
more information regarding the software can be acquired from the homepage of the
Neural Networks Research Centre (http://www.cis.hut.fi/research/). In the following
section, this software will be referred to as SOM_PAK.

To visualise the final self-organising map we will use the unified distance matrix
method (U-matrix). The U-matrix method can be used to discover otherwise invisible
relationships in a high-dimensional data space. It also makes it possible to classify data
sets into clusters of similar values. The simplest U-matrix method is to calculate the
distances between neighbouring neurons, and store them in a matrix, i.e. the output
map, which then can be interpreted. If there are walls between the neurons, the
neighbouring weights are distant, i.e. the values differ significantly. The distance values
are also displayed in colour when the U-matrix is visualised. Hence, dark colours
represent great distances while brighter colours indicate similarities amongst the
neurons. [Ultsch 1993]

The software we have used to visualise the final constructed self-organising maps and
the feature planes in this report is a program called Nenet version 1.1a. This software is
5
developed by the Nenet team at the Helsinki University of Technology. Nenet is a user-
friendly program designed to illustrate the use of self-organising maps, and provides an
easy way to visualise the output maps with not only the U-matrix method but also the
Interpolated 2D U-matrix method, and as parameter level maps. Because the software is
a commercial product the version used in this report is a demo version. More
information about the software and a downloadable demo version can be obtained from
the Nenet homepage (http://koti.mbnet.fi/~phodju/nenet/Nenet/General.html).

2.2 Choice of Companies

We selected the companies from five different regions: Asia, Canada, Continental
Europe, Northern Europe (the Nordic Countries) and the USA. It would have been
interesting to include companies from Africa or South America, but these companies
had very little or no financial information at all on the Internet, so these companies were
excluded. The 88 selected companies are presented in Table 2-1. There is also an
average of every region included as an additional company, bringing the total number
of companies to 93. The averages will make a comparison between the different regions
possible.

2.3 Choice of data and information

The starting point of this report was to use only the Internet as a source of financial
data. Hence, the data searching part of this research was executed by searching for
financial statements on the homepages of the companies, as well as in different
databases on the Internet.

Many of the companies did not have financial information for more than three years on
their homepages on the Internet. Therefore, in most cases, this was complemented with
financial information from databases such as the U.S. Securities and Exchange
Commission (http://www.sec.gov) for American companies, the System for Electronic
Document Analysis and Retrieval (http://www.sedar.com) for Canadian companies and
Japan Financials (http://japanfinancials.com) for Japanese companies. Since no good
database was found for the Continental European companies the financial information
has been complemented with annual reports received via regular mail.

6
N C Y N C Y
47 T 1995-99
1 B 1995-99 48 1995-98
2 1995-99 49 1995-99
3 E 1995-99 50 LS L 1995-99
4 HPY 1995-99 51 L 1995-99
5 N 1995-99 52 MC WC 1995-99
6 N 1995-99 53 Mx 1995-99
7 S 1995-99 54 M 1995-99
8 T 1995-99 55 Nx 1995-99
9 TN 1995-99 56 P 1995-99
10 T 1995-99 57 P 1995-99
11 1 1995-99 58 Q 1995-99
59 SBC 1995-99
60 S 1995-99
12 1995-99 61 T 1995-99
13 1995-99 62 UW 1995-99
14 B T 1995-99 63 V 1995-99
15 C & W 1995-99 64 X 1995-99
16 C T 1995-99 65 86 1995-99
17 T 1995-99
18 F T 1995-99
19 M 1995-99 66 B M 1995-99
20 MTV 1996-99 67 C 1995-99
21 M 1996-99 68 M 1995-99
22 O 1995-99 69 N N 1995-99
23 O 1995-98 70 S 1995-99
24 P 1995-99 71 T 1995-99
25 P T 1995-99 72 & 1995-99
26 R 1995-98
27 S 1996-99
28 S 1995-99 73 1995-99
29 S 1995-99 74 1995-99
30 TW 1995-99 75 1995-99
31 V 1996-99 76 R 1995-99
32 ( 1995-99 77 T 1995-99
78 K 1995-99
79 K 1995-99
33 3C 1995-99 80 M 1995-99
34 C 1995-99 81 M E 1995-99
35 1995-99 82 NEC 1995-99
36 C. 1995-99 83 NTT CM 1995-99
37 T&T 1995-99 84 OK 1995-99
38 x 1995-99 85 S 1995-99
39 B 1995-99 86 S 1995-99
40 BS 1995-99 87 S 1995-99
41 CT 1995-99 88 S 1995-99
42 C 1995-99 89 T 1995-99
43 C 1995-99 90 T 1995-99
44 C 1995-99 91 U 1995-99
45 E 1995-99 92 V N 1996-99
46 TE 1995-99 93 1995-99
N C
C E C
C
C
C C

T [2-1. T .
7
2.4 Choice of Ratios

To conduct the financial benchmarking of telecommunications companies, the
companies financial statements have been used as information source. From these
financial statements seven key ratios have been calculated for each company and used
as input data when training the self-organising map. These key ratios will be briefly
presented in the following section.

The selection of relevant key ratios was based on an empirical study by Lehtinen (1996)
where the international accounting differences were analysed in more detail, especially
as to the reliability and validity of the ratios. Seven financial key ratios, which fulfilled
the criteria of good validity and reliability, were selected and calculated for each of the
companies. The key ratios can be divided into four different classes: profitability ratios,
liquidity ratios, solidity ratios and efficiency ratios. In financial benchmarking it is
common to choose ratios that measure different aspects on financial behaviour. In this
financial benchmark more emphasis was put on the profitability since it is regarded to
be a driving force behind the companies operating in the telecommunications market.
Three profitability ratios were selected, Operating Margin, Return on Total Assets and
Return on Equity. In the class liquidity only one ratio was selected, the Current Ratio.
The solidity of a company was regarded to be nearly as important as the profitability
and thereby two ratios were selected, Equity to Capital and Interest Coverage. In the
final class, efficiency, only one ratio was selected, Receivables Turnover ratio. The
formulas for the calculation of the selected key ratios are presented below:

1. O M
2. ROT
3. ROE
4. C R
5. E C

6. I C

7. R T
. 100
S N
P O
. 100
) (

+
T
Ex I I T
. 100
) (

+ E R C S
I N
.
L C
C
. 100
) (

+
T
E R C S
.
Ex I
I N Tx I Ex I + +
.
) ( R
S N
P
R
L
R
S
R
E
R


8
3. Training the Maps

During the training process, several tests were carried out in order to determine suitable
parameters. The hexagonal lattice type was preferred for the visualisation of the output
map. Furthermore, the map ought to be of a rectangular form, rather than square, in
order to achieve a stable orientation in the data space [Kohonen 1996, 3-4]. Commonly
the x-axis should be about 30 per cent greater than the y-axis, thus forming a
rectangular output map. Another recommendation is that the training length of the
second part should be at least 500 times the number of network units [Kohonen 1997,
88].

To ease the neural networks learning process and improve the quality of the map the
input was standardised. For example, one of the selected key ratios would have a range
of 0 to 1, while another key ratio has a range of 100 to 100, then the contribution of the
second input will likely be given more weight than the first one. Because of this it is
essential to standardise the input data so that their value reflects their importance, or at
least that the value is similar in relation to the other input data [Bishop 1995]. In this
research the standardisation has been done by scaling the input variables by the variance
according to the following formula:

.
) (

x x
x

= [3.1]

During the training process the self-organising map was still placing too much weight
on the extreme values, even after the standardisation. To cope with this in order to
receive an interpretable map Johnson and Wichern (1997) suggests that one should
modify the input data by putting a limit to how great values the extreme observations
were allowed to take. In this research the extreme values have been limited to 50
respectively 50.

The constructed map was trained using input data for the years 1995-99, so only one
map was created and analysed. The reason for why all years were used to train one map,
and not one map for each year, is that now the same clusters appear for all years, and in
the same places. If one map would be trained for each year, different clusters would
probably appear, and would have to be analysed and interpreted separately.

The trained map is presented in Figure 3-1, and is of the size 9 x 6 neurons. On this size
of map, it is still rather easy to define the different clusters by looking at the colour
shades of the borders between the hexagons. The brighter colours of the hexagons imply
similar characteristics, while darker colours represent greater distances. The coloured
borders between the hexagons are of great value when trying to determine and interpret
clusters. Furthermore, it is also possible to visualise company movements in an
interpretable fashion on this somewhat larger map.

9

F [3-1. T : 9 x 6 .

Using the feature planes it is possible to visualise the values of a single vector column,
i.e. in this research the maps for one financial key ratio. These feature planes can be
analysed in order to discover how well the companies have been doing regarding one
single financial ratio [Kohonen 1996]. Thus, with these feature planes it is rather easy to
see where the companies with good profitability are located on the map, and in the same
fashion where the companies with poor profitability are located. The feature planes
provided excellent assistance in determining and analysing the company clusters on the
self-organising map.

I C E T C
R T
C R
O M ROT ROE
Ex
V

B
V B
T

Figure [3-2]: The Feature planes.

10
The feature planes show a map for each of the financial key ratios, where warmer
colours, i.e. red, represent high values, which in our case implies good values, and
darker colours show low values, which in our case implies poor values. A high value
does not necessarily mean a good value, but it does in the case of our selected key
ratios.

3.1 Defining the Clusters

By analysing the output map more carefully, six major clusters of companies were
identified. To identify the clusters we used both the U-matrix map [Ultsch, 1993] and
the feature planes [Kohonen, 1996]. By analysing the colours of the borders between
the hexagons, as well as the colour of the hexagon itself it is possible to find similarities
as well as differences. Furthermore, the values of the neurons have been evaluated in
order to determine that the clusters are right. The identified clusters are presented in
Figure 3-1 in the form of an U-matrix map:


F [3-3. T - .

In Figure 3-1 six different clusters of companies were identified. These clusters have
been labelled: A
1
, A
2
, B, C
1
, C
2
and D. The conclusions of the interpretations is as
follows:

- Group A
1
and Group A
2
represent the best in class companies. For the
companies situated in subgroup A
1,
profitability is very good, with very high
values in the financial ratios Operating Margin, ROTA, and ROE. Solidity is
decent, i.e. the values of the Equity to Capital ratio and the Interest Coverage
ratio vary from good to average. Generally, liquidity is slightly worse, i.e.
lower values in Current Ratio. For instance, companies like British Telecom
(97-99), Nokia (97-99), Samsung (95, 99), Sonera (96-97, 99), and Vodafone
(97, 99) are situated in this group.

- Group A
2
is the second subgroup of the best in class group. The companies
situated in this group are characterised by slightly lower profitability than
Group A
1
, but instead liquidity and solidity are much better. These
companies generally have the best values in Current Ratio on the map.
11
Furthermore, these companies possess very high values in the Equity to
Capital ratio, and are almost the only companies to show great values in the
Interest Coverage ratio. Efficiency is average. This group includes
companies like Benefon (95-97), Motorola (95), Sonera (98), and Vodafone
(95-96).

- Group B is where the companies with slightly poorer performance than
those in Group A
1
and A
2
are situated. These companies are distinguished by
good profitability, and especially the ROE ratio is excellent. These
companies also have somewhat worse liquidity and solidity than the
companies in Group A. Instead, the Receivables Turnover ratio is generally
better. Examples of companies are Alcatel (97-98), AT&T (95-99), Ericsson
(95-99), Lucent (96-99), Nokia (95-96), and Vodafone (98).

- Group C
1
is the better of two subgroups in Group C. Here the companies
possess decent profitability, good liquidity, and also good values in the
Equity to Capital ratio. But the values in Interest Coverage and Receivables
Turnover are worse. In this group, companies such as Alcatel (99), NTT
Docomo (95-99), Sonera (95), Sony (96-99), and Sprint (96-98) can be
found.

- Group C
2
is the slightly poorer of the two middle groups. These companies
have decent profitability, but poor liquidity. Interest Coverage and
Receivables Turnover are also poor, but Equity to Capital, on the other hand,
is very good. Companies situated in this group include for example British
Telecom (95-96), Matsushita (95-99) Motorola (96-99), and Telia (95-99).

- Group D is the poorest group. The companies with poor financial
performance can be found in this group. Distinguishing features are
commonly poor profitability and solidity. Liquidity is average and
Receivables Turnover varies from very good to poor. Generally this group
contains service providers from Europe and the USA, but also some
Japanese companies, mostly for the years 1998-99. Examples of companies
in this group include Alcatel (95), Benefon (98-99), Orange (95-98), MCI
WorldCom (96-99), Sony (95), and Sprint (99).

The characteristics of the identified clusters are summarised in Table 3-1. Each of the
selected financial key ratios have been valued from excellent to very poor within each
of the analysed clusters of companies. This valuation is based on the interpretation of
the feature planes. As mentioned earlier the red colours on the feature planes indicate
high values in the respective key ratio, while dark colours represent lower values.
12

2 527 52( & ( , 5
0 5 & & 7

Ex V V P P V P

V Ex V P
% V Ex V P P
&

V P V P
&

P V V P V P
P P V P P P V P P

T [3-1. T .

Based on Table 3-1, it is clear that both subgroups of Group A are good groups, but in
slightly different manner. Both groups show good profitability, even though subgroup
A
1
is performing slightly better. On the other hand, subgroup A
2
is characterised by
better liquidity and solidity. From this table it also becomes obvious that the key ratio
Receivables Turnover is not a very suitable measure when performing this type of
financial study of the telecommunications market. With values ranging from average to
very poor, it is likely that not much weight has been placed on this ratio when training
the map. In fact, most of the companies experienced good Receivables Turnover.
However, the values of this ratio are very low compared to, for example the ROE ratio.
This is probably the reason for why the self-organising map has placed so little weight
on this ratio. The conclusion is that although the Receivables Turnover ratio was
considered to have great reliability and very good validity, it was not a useful financial
key ratio in this report. The same problem can also be found with the Interest Coverage
ratio, relatively low values give low weight in the training process.

4. Benchmark Analysis of the Companies Over Time

In section 3, the clusters of companies on the self-organising output map were identified
and analysed, in this section a more detailed analysis is conducted concerning company
movement during the years 1995-99. Furthermore, the competing companies in the
specific markets will be benchmarked against each other. The companies on the
analysed self-organising maps will be labelled numerically according to Table 2-1 with
the company numbers, except for the map with country averages. Section 4.1 presents
country averages, section 4.2 presents the largest manufacturers and section 4.3 presents
the largest service providers. In section 4.4 the Northern European companies will be
benchmarked against each other, in section 4.5 the Continental European companies, in
section 4.6 the American companies, in section 4.7 the Canadian companies and finally
the Asian companies in section 4.8.

4.1 Country Averages

In this section, the company averages will be benchmarked against each other, the
accounting diversities between the five different regions will be analysed to see whether
the locations on the map can be explained. The figures of the five regions are calculated
as an average of the companies operating within the regions, i.e. the Nordic average
consists of the Northern European companies.
13


F [4-1. T .

The map shows that the company averages are actually positioned in the average
groups, as could be estimated in advance. No company average is situated in the best or
poorest groups. The only average to enter the somewhat poorer middle group, Group C
2
,
is the Canadian Average for the years 1995-98. What is interesting with this average is
that it is characterised by very high values in Equity to Capital. All of the Canadian
companies have very high share capital in proportion to total assets. Hence, the
Canadian companies seem to finance their operations rather heavily using their equity.
The presumption was that the American and Canadian companies would have rather
similar accounting methods and capital structures. Upon closer examination, it was
discovered that many of the American companies also have very high Equity to Capital
ratios. However, there exist large variations amongst these companies, many companies
having very low Equity to Capital ratios, thus decreasing the American average.
Another difference between the Canadian and American averages is that only service
providers were selected from Canada, hence Canadian profitability is usually very low.

The American average is situated a bit further away from the other company averages in
the middle group (pink circles). One reason for this may be the unrestrictive accounting
principles used in the United States of America. The Asian average seems to have
similar characteristics as the Nordic and European average, situated in the middle group
C
1
, even though a slight backtracking occurred in 1999, probably due to the Asian crisis
(red arrow).

As for the Asian crisis, the major crisis took place in 1997-98, but it is likely that the
crisis only affects the financial figures for 1998-99. The impact of the crisis on the
Asian telecommunications companies will be analysed in more detail in Chapter 6.8. In
advance, it was anticipated that the Asian average would be situated in a middle to poor
group, due to very strict accounting standards, but no major differences can be found in
comparison with, for example, the Nordic average.

14
As expected, the Nordic and European averages display rather similar characteristics,
and are consequently situated in the same group, Group C
1
. The only major movement
to speak of is that the European average was situated close to Group D in 1995-96.
However, in 1997-99, the financial figures have obviously improved, since the
European average is now situated close to Group A
1
(blue arrows).

4.2 The Largest Manufacturers

In Figure 4-2, a visualisation of the largest manufacturers in this study is presented. This
benchmark is international, hence companies from different countries are benchmarked
against each other. This means that some accounting differences might influence the
resulting output map.


F [4-2. T .

As Figure 4-2 shows, Nokia (No. 6) is the only mobile phone manufacturer that
experiences a steady increase in financial performance. Nokia is situated in Group B
during the years 1995-96, and took a leap into Group A
1
in 1997-99 (blue arrow). The
reason for the success is a constant increase in all of the selected key ratios. Judging by
the feature planes, Nokia was experiencing excellent profitability in 1999, and very
good liquidity and solidity. Only the value of Receivables Turnover was somewhat
lower than for the previous years.

Motorola (No. 54), on the other hand, shows an almost steady decrease in performance.
In 1995 the company was situated in Group A
2
, but has since then moved into the
slightly poorer of the middle groups (red arrows). This is probably due to increasing
competition on the telecommunications market. Examining the financial statements of
Motorola reveals, for example, that net income has been decreasing steadily since 1995.
Studying the selected key ratios shows that Motorola exhibits very good Equity to
Capital but, on the other hand, its profitability has decreased during the last four years.

15
The third of the major mobile phone manufacturers, Ericsson (No. 3), is firmly situated
in Group B during the years 1995-99 (purple circles). Ericsson is one of few of the
selected companies that display high Receivables Turnover ratios. Furthermore, the
Operating Margin and ROE ratios are excellent. The values in liquidity and solidity are
slightly poorer.

It is also interesting to take a look at the performance of the Finnish mobile phone
manufacturer Benefon (No. 1). During, three consecutive years (1995-97) the company
was firmly situated in Group A
2
, experiencing great liquidity, solidity, and profitability.
In 1998-99, the company experienced the effects of both the Asian crisis and the
Russian crisis, crashing into the poorest group, Group D (orange arrows). During these
years, profitability is terrible and Interest Coverage is disastrous. However, liquidity and
efficiency are good, and Equity to Capital is excellent.

Sony (No. 88), on the other hand, has experienced a slight increase in their
performance. In 1995 the company was situated in the poorest group, but eventually
ended up in the slightly better middle group, C
1
, in 1999 (turquoise arrows). What is
interesting about Sonys performance is that they do not seem to experience any effects
of the Asian crisis, except for a slight backtracking in 1999. One reason for why Sony is
not experiencing any greater effects of the crisis could be the fact that they are a large
international company, thus the Asian market comprises only a small part of their
operations. Overall, Sony is performing much like the average Asian company.

Lucent (No. 51) is a company with similar performance to Ericsson, situated in the
same neurons for the years 96-98. In 1995, the company was situated in Group D,
probably as a result of the fact that they were in their first years of operations. In 1996,
they moved into Group B, and have been situated in the same area since then (yellow
arrow). In 1999, Lucent experienced an increase in all of the selected key ratios, except
Receivables Turnover. At this stage, Lucent had excellent ROE and good Operating
Margin, ROTA, Equity to Capital, and Interest Coverage ratios. As a result of this, a
small movement towards Group A
1
is shown on the output map.

Analysing this map reveals that most of the manufacturing companies are situated either
in the middle groups or in Group B. Only a few companies manage to place themselves
in Group A
1
or A
2
. Similarly, only a few companies have been placed in the poorest
group, amongst them Benefon. Most of the Asian companies are situated close to or in
Group D. This is likely to be a consequence of the strict accounting methods in Asia.

4.3 The largest Service Providers

In this section, the largest service providers are benchmarked against each other. Many
of the service providers are new, growing companies that have recently started selling
their services on the market. As a result of this, many of the companies are experiencing
low profitability. This is probably a result of the high cost of developing their own
infrastructure, i.e. building the telecommunications network, hiring personnel, and
developing new services. Therefore, it was presumed that many of the companies would
be situated in the group of poorly performing companies.

16

F [4-3. T .

An excellent example of the earlier mentioned hypothesis, that some of the service
providers would be situated in the worst group, is the company Orange (No. 23).
Orange is a fairly new service provider that is still in the phase of developing their own
infrastructure. Examining the map reveals that Orange is situated in the poorest group,
Group D, during the years 1995-98 (brown circle). Unfortunately, data for 1999 was not
officially published on the Internet at the time this study was carried out. By studying
the selected key ratios, it becomes clear that profitability is low, even though
improvements have been made. During the last years, liquidity is below average and
solvency is terrible. The reason for the low solvency and the ROE ratios is that the
figure of retained earnings
1
is negative, a figure as large as -1.1 billion! This suggests
the possible acquisition of Orange, since it is clear that they cannot continue to operate
in the same fashion, as they have to date. Actually, in May 2000, France Telecom
acquired Orange. This is probably also the reason for why financial information for
1999 has not been published as of today.

The acquisition of Orange makes France Telecom (No. 18) the second largest service
provider in Europe, with 22 million customers. France Telecom is a Group C
1
company,
although they are situated close to both Group B and Group A
1
(white arrows). France
telecom shows rather similar values in the selected key ratios over the years, with high
values in Operating Margin, ROE, and Equity to Capital ratios.

For Sonera (No. 7), everything looks nearly as bright as for Nokia, situated in Group C
1

in 1995, and entering Group A
1
in 1996-97 (turquoise arrows). In 1998, Sonera
experienced a slight backtracking and moved into Group A
2
, but eventually returned to
the best in class group in 1999. This slight backtracking becomes obvious when

1
( )
17
examining the financial key ratios closer; in 1998 the Operating Margin ratio decreased
slightly while the Interest Coverage ratio increased significantly. In 1999, the Operating
Margin ratio increased while the Interest Coverage ratio decreased.

Vodafone (No. 31) is constantly situated in the better groups. In 1995-96 they were
firmly situated in Group A
2
, and moved into Group A
1
in 1997 (yellow arrows).
Analysing the financial key ratios slightly closer reveals that they showed a slight
increase in ROE, but instead, a decrease in solidity. In 1998, Vodafone moved into
Group B, much due to a continuous decrease in solidity, although they experienced an
uninterrupted increase in ROE. In 1999, the ROE ratio continued to increase, at the
same time as solidity improved, resulting in the company re-entering Group A
1
.

British Telecom (No. 14) is another English service provider that shows some
movement. This company is situated in Group C
2
during the years 1995-96, moving into
Group A
1
in 1997-99 (pink arrows). Examining the financial ratios closer reveals that
British Telecom shows a slight increase in all financial ratios, during the years 1995-99.

When studying the American companies, it is obvious that AT&T (No. 37) is a Group B
company, situated in the same group during the years 1995-98. However, they show
somewhat decreasing performance, moving into Group C
1
(red arrows) in 1999.
Actually, this slight decrease seems to be characteristic for larger American companies.
Apart from AT&T, Motorola also, as mentioned earlier, experienced somewhat poorer
performance, and Sprint too backtracks into the poorest group during the years.

One thing that is very interesting is the clustering of the two companies Telia (No. 10)
and TeleNor (No. 9). These two companies negotiated possible merger with each other
for quite a long time, but eventually they broke off the negotiations. What is interesting
is that these two companies seem to have almost identical financial performance. Both
companies are situated in the same neuron for the years 1995-96 and 1998, and next to
each other during the other years in Group C
2
(green circle). Examining the selected key
ratios reveals that Telia shows slightly better profitability and liquidity, while TeleNor
shows slightly better solvency and efficiency.

Another interesting observation is the performance of NTT Docomo (No. 83), which is
the largest service provider in Japan. NTT actually managed to improve their financial
performance slightly during 1997-99 (blue arrows). When examining the financial key
ratios a little bit closer it becomes clear that NTT has managed to increase not only their
profitability, but also liquidity and solidity during the Asian crisis. One reason for this is
the dramatic increase in their customer base, which has occurred due to new and
popular services. At the same time, as NTT has improved, Japan Telecom (No. 77) has
moved towards the poorer end of the map. This is probably a result of the Asian crisis,
as well as customers and market share lost to NTT.
18
4.4 The Northern European Companies

In the following section, the Northern European companies will be benchmarked
against each other. It is likely that the accounting principles will not have a large impact
on this benchmark, since the companies in this region has similar laws and principles.
Figure 4-4 visualises the movement over the years for the Northern European
companies included in the research.

The companies Benefon, Ericsson, Nokia, Sonera, Telia and TeleNor have been
benchmarked earlier and will not be examined in detail in this section. Instead, more
emphasis will be placed on their regional competitors.


F [4-4. T N E .

Regarding the improvements in the performance of Sonera it is also interesting to see
how the local competitor Helsingin Puhelin Yhtit (HPY) is performing. This company
is also known as Elisa Communications, and owns Radiolinja, which is the second
largest provider of mobile services in Finland. Like Sonera, HPY (No. 4) has
experienced a continuous increase in its financial performance, much in the same rate as
the usage of mobile phones has increased. HPY was situated in Group C
1
in 1995-97 but
moved into Group B for the years 1998-99 (yellow arrows). This movement is a result
of an increase in the key ratios ROTA, ROE, and Interest Coverage, even though HPY
shows a slightly lower Operating Margin ratio than for the first three years.

Doro (No. 2) is a Swedish company specialised in telecommunications equipment. This
company too has experienced an almost continuous increase in financial performance.
Except for a slight backtracking in 1996 within the Group C
1
, Doro have moved into
Group B where they are firmly situated in the same area as Ericsson (blue arrows). The
main reason for this movement is a heavy increase in the ROE ratio, but also a
continuous increase in the other profitability ratios can be found.

19
Tele Denmark (No. 8) is firmly situated in the same area during the years 1995-99,
currently situated in Group A
1
, close to both Group B and Group C
1
(yellow arrows).
During this period of five years they experienced hardly any larger changes in their
financial key ratios.

NetCom (No. 5) is a rather new service provider from Sweden that also operates in
Denmark, Norway, and in Europe. Known brands are Comviq and Tele2. The fact that
this company is a rather new service provider is also visualised in the financial figures,
NetCom is situated in Group D for all years included in this study (red arrows).
Actually, the company has shown a positive financial result for the last two years, but
terrible values in the ROE ratio are still placing them in the worst group.

An important question arises whether the Return on Equity is such a suitable financial
ratio for this type of comparison of telecommunications companies. In many cases a
company with negative profitability for several years will also have negative equity,
because of the cumulated retained earnings (losses). This results in a phenomenon
where the ROE ratio turns into a very high positive ratio, since a negative number
divided by another negative number becomes a positive number. This does not give a
fair view of the companies performance. In this report this problem has been dealt with
by altering this number into a negative one in the case where such a problem exists.
This is not recommended but in this research it gives a more accurate view of the
companies financial performance. Since rather many companies, mostly service
providers, had this problem, the conclusion is that this financial key ratio is not very
usable in this type of financial research. The findings of Chapter 3 shows that the ROE
ratio is one of the most valid key ratios used in this report; it is also the ratio most
widely used by the telecommunications companies in their financial statements.
However, the extreme values and the great variations in this ratio make it possible that
too much weight will be put on it when training the self-organising map.

4.5 The Continental European Companies

In the following section the Continental European companies will be benchmarked
against each other, this benchmark will include both manufacturers and service
providers. The Northern European companies will not be included since a benchmark
for these was conducted earlier. In this benchmark of the Continental European
companies some companies are included with missing information for either 1995 or
1999. Therefore, these companies will appear for four different years.

20

F [4-5. T C E .

The Continental European companies seem to be spread rather wide over the map, with
companies in all groups. There is no astonishing clustering of the companies. A few
companies and their respective movement will be evaluated in more detail.

Alcatel (No. 12), a French manufacturer of telecommunications equipment, shows a
rapid increase in financial performance for the years 1995-98 (dark blue arrows).
Located in Group D in 1995 Alcatel moves into Group B in 1997-98, primarily because
of a rapid increase in the ROE and Interest Coverage ratios. In 1999 Alcatel falls to
Group C
1
, again mostly due to a heavy decrease in the values in the ROE and Interest
Coverage ratios.

It is interesting to see that Philips (No. 24) has a similar movement to Alcatel. Philips is
situated in Group C
1
in 1995, but backtracked close to Group D in 1996 (red arrows).
Examining the financial key ratios reveals that Philips encounters a slight decrease in
the Operating Margin and ROTA ratios, but the decrease in the ROE ratio is significant.
In 1997 Philips financial performance improved greatly, taking a leap into Group B,
resulting from an increase in profitability, liquidity and solidity. In 1998 Philips
improves a little bit more, within Group B, despite a decrease in both the Operating
Margin and ROTA ratios. However, the reason for the improvement can be found in the
ROE ratio, which increases dramatically. In 1999 Philips takes a dive into Group C
2
,
despite a remarkable increase in the Operating Margin and ROTA ratios. The reason for
this backtracking is, not surprisingly, a significant decrease in the ROE ratio. Philips is
an excellent example of how the ROE ratio dictates the company movement on the self-
organising map in this research.

Siemens (No. 28) is clearly a Group C
1
company, situated in the same group for all of
the years 1995-99. Evaluating the financial key ratios reveals that Siemens shows good
profitability, excellent liquidity, and good solidity. In 1999 Siemens improves its
financial performance slightly, moving a few neurons closer to Group B (yellow
21
arrows). This movement is probably due to a somewhat increasing ROE ratio, but the
other ratios are also on a good level.

Sagem (No. 27) is another French manufacturer benchmarked in this research. Sagem is
actually one of the companies included with missing information for the year 1995.
Furthermore, the figures for interest expenses could not be found in the annual
statements. Therefore, the values used in this research are estimated values. This should
not have so much impact on the result since not very much weight was placed on the
Interest Coverage ratio by the self-organising map. In 1996 Sagem was situated in
Group C
1
, and moved into Group A
2
in 1997 (turquoise arrow). This movement is
mainly a result of increased liquidity and solidity.

The final company to be benchmarked in this section is Deutsche Telekom (No. 17), the
largest service provider in Germany. Deutsche Telekom is situated in Group B in 1995,
but moves back to Group C
1
in 1996-99. The reason for this movement is a significant
decrease in the profitability ratios, especially in the ROE ratio. For the rest of the years
the figures remains on the same level, with small variations.

4.6 The American Companies

In the next section the American companies will be benchmarked against each other.
The assumption was that most of the American companies would be classified as
middle to good companies, because of the easy accounting principles. As the Figure 6-7
below illustrates, not many of the companies can be found in the middle groups; instead
most of them are situated either in the worst or the better groups.



F [4-6. T .

Most of the companies in the Group D are service providers. As stated earlier, the poor
performance is probably a result of the high costs of entering the market. One example
22
is MCI WorldCom (No. 52), who in 1995 in fact was situated in Group B, with good
financial ratios, and fantastic values in the ROE ratio. In 1996-99 MCI WorldCom fell
to Group D (red arrows) as a result of poor profitability and especially very poor values
in the ROE ratio. In 1999 MCI WorldCom actually had very good profitability, but the
ROE ratio was still very poor, resulting in the company remaining in Group D.

One manufacturer situated in Group D is Audiovox (No. 38) in 1995-96. In 1997 they
jumped into Group B (blue arrows), which is a result of an improved financial result. In
1998 Audiovox experiences a decrease in most of the financial key ratios, and in the
ROE ratio the decrease is significant. The result of this is a backtracking into Group C
1
.
Analysing the annual statements reveals that no direct decrease in profitability occurs.
However, in 1997 they sold some investments, which made the financial result positive.
In 1999 Audiovox re-enters Group B due to an increase in all of the financial key ratios.

As mentioned earlier both Motorola and AT&T show slight backtracking, which is also
the case for the service provider Sprint (No. 60). In 1995 Sprint is situated in Group C
1
,
and is moving into Group A
1
in 1996 due to an increase in most of the selected key
ratios (yellow arrows). In 1997-98 Sprint moves back into Group C
1
, and towards Group
D. This is mainly a consequence of a decrease in the profitability. In 1999 the financial
result was actually negative, resulting in entering the Group D.

Qualcomm is an American mobile manufacturer that is performing well, situated in
Group B for all years (pink arrows). Qualcomm shows good values in all financial key
ratios, and for the ROE ratio the values are excellent. Qualcomm seems to have similar
characteristics to Ericsson, situated in the same two neurons for all of the years 1995-
99. This is particularly interesting since Ericsson this summer (2000) acquired the
CDMA wireless infrastructure business from Qualcomm, including its R&D resources,
related assets, premises and personnel.

4.7 The Canadian Companies

In the following section the Canadian companies will be benchmarked against each
other. This is a region consisting of very few companies, which are all service
providers. In a way this benchmark may not be too important, but it is interesting to see
how the Canadian companies differ from the American as well as visualising why the
country average is situated in the slightly worse middle group.

23

F [4-7. T C .

As the Figure 4-7 visualises, almost all of the Canadian companies are situated close to
each other in Group C
2
, with the exception of Bell Mobility (No. 66) and Clearnet (No.
67). These two companies are moving all over the map with great variations in their
financial performance. Nearly all of the other companies are situated within the same
area, with three companies in the same neuron. It seems that most of the Canadian
service providers have very similar characteristics.

Bell Mobility is the leading wireless provider in Canada, and has moved into Group A
1

after a short backtracking into Group D in 1997 (dark blue arrows). The backtracking
into Group D is a result of a negative ROE ratio. This again, is a consequence of a
negative net income in 1997, while all other ratios are positive and good. Once again, it
becomes obvious that the self-organising map has placed most weight on the ROE ratio,
probably as a result of the variations and extreme values found there. Again the
question arises whether the ROE ratio is to be considered as a valid ratio for this type of
research.

Clearnet is a rather newly started service provider, which is suffering from very poor
equity, which leads to terrible values in the ROE and Equity to Capital ratios. The
financial result has been negative for all of the years between 1995-99, placing Clearnet
in Group D (red arrows). The constant decrease in financial performance resulted in
Clearnet being acquired by Telus (No. 71) in August 2000, resulting in the largest
telecommunications company in Canada. Telus is a company with rather firm financial
performance, situated in Group C
2
for the years 1995-99 (yellow arrows) with good
values in the profitability ratios and exceptionally high values in the Equity to Capital
ratio.

24
4.8 The Asian Companies and the Asian Crisis

In the following section the Asian companies will be benchmarked against each other.
One objective of this section is to evaluate whether the Asian crisis has affected the
Japanese telecommunications companies and to what extent. Included in this
benchmark are also two Indian companies and one Australian, but these companies
were probably not affected by the crisis.

It can also be mentioned that recession has existed in Asia for the last ten years. To a
large extent this means that the real effects on the financial performance of the
companies are not visualised in this research. However, the major economic crisis
occurred in 1997-98, and these effects will likely be visualised by the following
benchmark. [Corsetti, 1998]

But although recession has existed, the Asian market is a sleeping giant with a huge
amount of potential telecommunications customers. In Finland roughly 70 per cent of
the population have a wireless phone, but only 5 per cent in Asia. Still there live 3.3
billion people in Asia, representing 58 per cent of the worlds population. The number
of wireless users in Asia is expected to double in the next three years, to 350 million
users [Ihlwan et al, 2000]. In January 2000 there were around 400 million mobile users
in the whole world. When looking at these figures it becomes a little bit more
interesting to look at the financial performance of Asian companies, and to think about
whether known companies like Nokia and Ericsson will penetrate the Asian market.


F [4-8. T .

As the map illustrates most of the Japanese companies are situated in either the middle
groups or the poor group. The Australian company Telstra (No.89) and the Indian
companies Indosat (No. 74) and Videsh (No. 92) are situated in Group A. The reason
for Telstras success is probably that they are such a large service provider in Australia
that the situation resembles a monopoly. Regarding the Indian companies a suspicion
exists that their success could depend on the accounting principles used in India.
25

Kokusai (No. 78) is a Japanese mobile manufacturer situated in Group A
2
for the years
1995-97, showing great values in all key ratio figures. In 1998-99 they clearly
experience the effects of the Asian crisis, crashing into the worst group in 1999 (red
arrows). At this stage they show terrible profitability with low values in all profitability
ratios. For this smaller company, which only has Asia as the market for their products,
the financial crisis was devastating.

OKI (No. 84) is another Japanese company that backtracks into Group D. Starting in
Group B in 1995 OKI first moves into middle group C
1
before crashing into Group D in
1997-99 (yellow arrows). The actual Asian crisis occurred in 1997-98, but for most
companies the effects are visualised in the financial figures for 1998-99. However, it
seems that OKI experienced the effects already in 1997 with a decreasing profitability,
although it is still positive at this stage. This movement on the map is most likely a
cause of a drastic decrease in the ROE ratio. In 1998-99 OKI shows a negative
profitability, decreasing the value in all of the selected financial key ratios.

Matsushita (No. 80) is the second largest manufacturer of electronic equipment in the
world, more known for their brand Panasonic. Matsushita does not experience any
greater effects from the Asian crisis, even though they backtrack slightly in 1999.
Overall, Matsushita is situated in the same two neurons for all five years, in Group C
2

(blue circles). This company shows rather similar values in all of the selected key ratios,
except for the Equity to Capital, they constantly show great values. The reason why
Matsushita is not so much affected by the crisis is probably that they are a large,
international company like Sony.

The second largest service provider in Japan is Japan Telecom (No. 77). This company
obviously experiences the effects of the Asian crisis, performing a steady backtracking
since 1998 (turquoise arrows). Looking at the financial key ratios reveals that the
profitability ratios have decreased considerably while the rest of the ratios seem to be at
the same level as for the years 1995-97.

Samsung (No. 85) seems to be the only company that is able to recover well from the
Asian crisis. Samsung is situated in Group A
1
in 1995, but falls back to Group C
2
in
1996 as a result of a significant decrease in the profitability, resulting in a decrease in
all of the profitability ratios (pink arrows). In 1997-98 the business is even worse,
moving Samsung into Group D with a negative result and very poor values in the
selected key ratios. This backtracking is likely to be an effect of the Asian crisis. In
1999 Samsung is back in business. Great profitability and a strong increase in all key
ratios move them back into Group A
1
.

26
5. Conclusion

Six different clusters of companies were found and analysed in more detail in the self-
organising map; two groups with very good financial performance, one with above
average performance, two average groups, and one with very poor financial
performance. When visualising the companys movement over the years, these clusters
were used to determine whether a company was performing well or poorly. The
movement of a company on the map indicates a change in its financial performance.

Different benchmarking studies were performed over companies in different regions,
and their performance were measured against each other. The best in class companies
were identified, as well as those struggling to stay alive. In these benchmarking studies
the companies movement over the years is visualised and evaluated. The movement of
a company indicates whether a company has performed well or poorly.

The results of the study indicate that the Northern European companies have performed
well, both Nokia and Sonera can be found in the best group. Regularly the Continental
and Northern European companies are performing equally well, what is surprising is
that the Asian companies also seem to have similar financial performance, except for
somewhat decreasing financials during the Asian crisis in 1998-99. The American and
Canadian companies show similar financial characteristics, although most of the
Canadian companies are located in the slightly poorer of the middle groups. The
American companies on the other hand are spread out over the whole map.

One interesting thing is that the worst group, Group D, consists mainly of service
providers from Continental Europe and America, except for a few Japanese companies
during the years 1998-99. The reason for why the service providers usually are
performing poorly financially is probably that most of the companies are very young
and still developing their own infrastructure.

The output map received and analysed in this study indicates that the self-organising
map technique is a useful tool in structuring large amounts of input data, and also to
find patterns that otherwise might be difficult to discover.

7. Acknowledgements

The financial support from TEKES (grant number 40943/99) is gratefully
acknowledged.
27
References

Ambroise, C., Seze, G., Badran, F., and Thiria, S. (2000), Hierarchical Clustering of
Self-Organizing Maps for Cloud Classification. Neurocomputing, Vol. 30, No.
1, pp. 47-52.

Back, B., Sere, K. and Vanharanta, H. (1998), Managing Complexity in Large Data
Bases Using Self-Organizing Maps. Accounting Management and Information
Technologies. No. 8, 1998, pp. 191-210.

Becanovic, V. (2000), Image Object Classification Using Saccadic Search, Spatio-
Temporal Pattern Encoding and Self-Organization. Pattern Recognition Letters,
Vol. 21, No. 3, pp. 253-263.

Bendell, T., Boulter, L. and Kelly, J. (1998), Benchmarking for Competitive Advantage.
Pitman Publishing. London.

Bishop, C. M. (1995), Neural Networks for Pattern Recognition. Oxford University
Press. Avon. pp. 296-300

Chen, D., Chang, R., and Huang, Y. (2000), Breast Cancer Diagnosis Using Self-
Organizing Map for Sonography. Ultrasound in Medicine and Biology, No. 26,
pp. 405-411.

Corsetti, G., Pesenti, P., and Roubini, N. (1998), What Caused the Asian Currency and
Financial Crisis? <http://www.stern.nyu.edu/globalmacro/> (read August 8
th

2000)

Ihlwan, M., Webb, A., and Caragata, W. (2000), Asia Gets Hooked on Wireless.
Business Week, June 19, 2000, pp. 62-64.

Johnson, R. A. and Wichern, D. W. (1997), Business Statistics: Decision Making with
Data. John Wiley & Sons, Inc. The United States of America. pp. 63

Karlsson, J. (2000), Financial Benchmarking of Telecommunications Companies.
Unpublished Masters Thesis at the Department of Information Systems at bo
Akademi University, Turku.


Karlf, B. (1997), Benchmarking i verkligheten. Svenska Frlaget Liv & Ledarskap AB.
Borg.

Kaski, S., Kangas, J., and Kohonen, T. (1998), Bibliography of Self-Organizing Map
(SOM) Papers 19811997. Neural Computing Surveys, 1, pp. 102-350.
<http://www.icsi.berkeley.edu/~jagota/NCS/> (read July 24
th
2000).

Kohonen, T., Hynninen, J., Kangas, J. and Laaksonen, J. (1996), SOM_PAK: The Self-
Organizing Map Program Package. Helsinki University of Technology.
28

Kohonen, T. (1997), Self-Organizing Maps. 2
nd
edition. Springer-Verlag. Heidelberg.


Kohonen, T. (2000), 4311 Works That Have Been Based on the Self-Organizing Map
(SOM) Method Developed by Kohonen. <http://www.cis.hut.fi/nnrc/refs/> (read
October 24
th
2000).

Lehtinen, J. (1996), Financial Ratios in an International Comparison. Acta Wasaensia
No. 49. Vasa.

Raivio, O., Riihijrvi, J., and Mhnen, P. (2000), Classifying and Clustering the
Internet Traffic by Kohonen Network. Proceeding of the ICSC Symposia on
Neural Computation (NC2000) May 23-26, 2000 in Berlin, Germany. ICSC
Academic Press.

Stein, R. (1993), Selecting Data for Neural Networks, AI Expert. Vol. 8, No. 2,
February 1993, pp. 42-47.

Ultsch, A. (1993), Self organized feature planes for monitoring and knowledge
acquisition of a chemical process, Proceedings of the International Conference
on Artificial Neural Networks, pp. 864-867. Springer-Verlag. London

Visa, A., Toivonen, J., Back, B. and Vanharanta, H. (2000), A New Methodology for
Knowledge Retrieval from Text Documents. Proceedings of TOOLMET2000
Symposium Tool Environments and Development Methods for Intelligent
Systems, pp. 147-151.

Databases Used:

Global Corporate Information Service: http://www.gsis.com
Hoovers Online (U.K): http://www.hoovers.co.uk
HUGIN Online International: http://www.huginonline.com
Japan Financials: http://japanfinancials.com
The Canadian Depository of Securities (SEDAR): http://www.sedar.com
U.S. Securities and Exchange Commission (EDGAR); http://www.sec.gov,
http://www.freeedgar.com


























Turku Centre for Computer Science
Lemminkisenkatu 14
FIN-20520 Turku
Finland

http://www.tucs.abo.fi/






University of Turku
Department of Mathematical Sciences



bo Akademi University
Department of Computer Science
Institute for Advanced Management Systems Research



Turku School of Economics and Business Administration
Institute of Information Systems Science