Performing financial benchmarks in today's information-rich society can be a daunting task. With the evolution of the Internet, access to massive amounts of financial data, typically in the form of financial statements, is widespread. Managers and stakeholders are in need of a tool that allows them to quickly and accurately analyze these data. An emerging technique that may be suited for this application is the self-organizing map. The purpose of this study was to evaluate the performance of self-organizing maps for the purpose of financial benchmarking of international pulp and paper companies. For the study, financial data in the form of seven financial ratios were collected, using the Internet as the primary source of information. A total of 77 companies and six regional averages were included in the study. The time frame of the study was the period 1995 2000. A number of benchmarks were performed, and the results were analyzed based on information contained in the annual reports. The results of the study indicate that self-organizing maps can be feasible tools for the financial benchmarking of large amounts of financial data.
There are many parties interested in the financial performance of a company. Investors want to find promising investments among the thousands of stocks available on the market today. Managers want to be able to compare the performance of their company to that of others in order to isolate areas in which the company could improve. Creditors want to analyze the company's long-term payment ability, and auditors want to assess the accuracy of a company's financial statements. Financial analysts want to compare the performance of a company to that of others in order to find financial trends on the markets. A tool commonly used by these parties is financial competitor benchmarking (Bendell, Boulter & Goodstadt, 1998).
The purpose of financial competitor benchmarking is to objectively compare the financial performance of a number of competing companies (Karl f, 1997). This form of benchmarking involves using quantitative data, i.e., numerical data, usually in the form of a number of financial ratios calculated using publicly available financial information. The information required for these comparisons can commonly be found in companies' annual reports.
The problem with these comparisons is that the amount of data gathered quickly becomes unmanageable. Especially with the advent of the Internet, access to financial information is nearly infinite. This has led to a situation, faced by many managers and investors today, in which the amount of data available greatly exceeds the capacity to analyze it (Adriaans & Zantinge, 1996).
A possible solution to this problem is to use data-mining tools. Data-mining tools are applications used to find hidden relationships in data. One data-mining tool that could be particularly suitable for the problem in this case is the self-organizing map. Self-organizing maps are two-layer neural networks that use the unsupervised learning method. Self-organizing maps have been used in many applications. By 1998, over 3,300 studies on self-organizing maps had been published (Kaski, Kangas, & Kohonen, 1998). Today, this figure is over 4,300 (Neural Networks Research Centre, 2001). Most applications of self-organizing maps have dealt with speech recognition, engineering applications, mathematical problems, and data processing (Kaski et al., 1998). Some examples of more recent research papers include cloud classification (Ambroise, Seze, Badran, & Thiria, 2000), image object classification (Becanovic, 2000), breast cancer diagnosis (Chen, Chang, & Huang, 2000), industrial process monitoring and modeling (Alhoniemi et al., 1999), and extracting knowledge from text documents (Visa, Toivonen, Back, & Vanharanta, 2000).
Self-organizing maps group data according to patterns found in the dataset, making them ideal tools for data exploration. Kiang and Kumur (2001) compared the use of self-organizing maps to factor analysis and K-means clustering. The authors compared the tool's performances on simulated data, with known underlying factor and cluster structures. The results of the study indicate that self-organizing maps can be a robust alternative to traditional clustering methods. A similar comparison was made by Costea, Kloptchenko, and Back (2001), who used self-organizing maps and statistical cluster analysis (K-means) to compare the economic situations in a number of Central-East European countries, based on a number of economic variables. The authors found the self-organizing map "a good tool for the visualization and interpretation of clustering results."
However, although many papers on self-organizing maps have been published, very few studies have dealt with the use of self-organizing maps in financial benchmarking. An example of the application of neural networks for financial analysis is the study by Mart n-del-Br o and Serrano-Cinca (1993). These authors used self-organizing neural networks to study the financial state of Spanish companies and to attempt to predict bankruptcies among Spanish banks during the 1977 85 banking crisis. Another example is the aforementioned study by Costea et al. (2001).
In our research group, we have conducted several studies on using self-organizing maps for benchmarking and data-mining purposes. Back, Sere, and Vanharanta (1998) compared 120 companies in the international pulp and paper industry. The study was based on standardized financial statements for the years 1985 89, found in the Green Gold Financial Reports database (Salonen & Vanharanta, 1990a, 1990b, 1991). The companies used in the experiment were all based in one of three regions: North America, Northern Europe or Central Europe. The companies were clustered according to nine different financial ratios: Operating profit, Profit after financial items, Return on Total Assets (ROTA), Return on Equity (ROE), Total Liabilities, Solidity, Current Ratio, Funds from Operations, and Investments. The ratios were chosen by interviewing a number of experts on which ratios they commonly used. The objective of the study was to investigate the potential of using self-organizing maps in the process of investigating large amounts financial data. Eklund (2000), Eklund, Back, Vanharanta, and Visa (2001), Karlsson (2001), Karlsson, Back, Vanharanta, and Visa (2001), and str m (1999) continued assessing the feasibility of using self-organizing maps for financial benchmarking purposes.
Back, Sere and Vanharanta (1997) and Back, str m, Sere, and Vanharanta (2000) are follow-up studies to the 1998 paper. The principle difference is that maps for the different years were trained separately in Back et al. (1998), while a single map was used in Back et al. (1997) and Back et al. (2000). Moreover, in Back et al. (2000), the data was from 1996 1997 and collected from Internet. The results showed that a single map makes it easier to follow the companies' movements over years. The results of the studies also gave further evidence that self-organizing maps could be feasible tools for processing vast amounts of financial data.
The purpose of this study is to continue to assess the feasibility of using self-organizing maps for financial benchmarking purposes. In particular, in analyzing the results, we will assess the discovered patterns by putting more emphasis on interpreting the results with existing domain knowledge. This chapter is based on the findings of Eklund (2000).
The rest of this chapter is organized as follows: the next section describes the methodology we have used, benchmarking, the self-organizing map, and choice of financial ratios. Following that, we present the companies included in the study, and then discuss the construction of the maps. The last two sections of the chapter present the analysis of the experiment and the conclusions of our study.