Financial performance analysis of European banks using a fuzzified Self-Organizing Map

Abstract

Due to the recent wave of bank failures, stress tests have been conducted on banks within the European Union. The stress tests, however, only consider the adequacy of a bank's capital ratios, whereas the general financial performance of individual banks is disregarded. In this paper, we use the Self-Organizing Map (SOM) to perform a visual multidimensional and temporal financial performance analysis of European banks. We address several problems concerning financial performance analysis. We deal with the problem of selecting suitable financial ratios by performing dimensionality reduction using Principal Component Analysis. We also deal with difficult data using outlier trimming and normalization techniques, and use the SOM for imputing missing values. We use a decision-framework for choosing the final model, based upon a set of map and clustering quality measures. In addition, we implement a second-level fuzzified Ward clustering for visualization purposes and for assessing the crispness of the solution. The result is a visual SOM model for financial performance analysis of European banks.

Data and dimension reduction for visual financial performance analysis

“…MDS and the SOM have been applied in illustrating the performance of banks on a low-dimensional display: MDS was used for illustrating the performance of Spanish banks 27 and the SOM was used for benchmarking European banks. 28 In addition, while there exist applications of both MDS 29 and the SOM 30 for predicting failures of banks, the SOM has retained its popularity not only in the domain of bank analysis but also in other financial areas. Some examples of financial application areas of the SOM are firm-level analysis of the pulp and paper industry 31 and country-level analysis for financial stability surveillance.…”

“…One common way to represent cluster structures in an SOM is applying a second-level clustering on the prototypes and visualizing it through color coding. 85 This has also been done using fuzzy methods, 28 where soft memberships of prototypes in each cluster can be represented on their own grids or associated with individual data. The strengths and directions of state transition patterns on an SOM can similarly be represented on own grids, such as prototype-to-cluster transition probabilities in Sarlin et al, 86 not to mention using the prototypes as an input for some other predictive methods, such as the neural network in Serrano-Cinca (1996), 87 and visualizing the prediction on the SOM grid.…”

A Framework for State Transitions on the Self‐organizing Map: Some Temporal Financial Applications

“…The number of nodes and clusters were set according to evaluation criteria based upon map and clustering quality measures. For further details of the model, readers are referred to Sarlin and Eklund (). The SOM model, with data for Deutsche Bank and Banco Santander from 2002–2008, is displayed in Figure and its feature planes are in Figure .…”

“…Sarlin and Eklund () extended the SOM to show membership degrees of each time‐series point to each cluster using fuzzy C ‐means clustering. This is suitable for assessing the current state, but says nothing about future transitions.…”

“…The SOM has been shown to be a particularly feasible tool for building visual monitoring systems for financial performance comparison of companies (Eklund et al , ) and for monitoring indicators of country‐level financial crises (Sarlin and Peltonen, ). We show the added value of the transition‐probability framework by applying it to two models in these fields: financial performance comparison of banks (Sarlin and Eklund, 2011) and monitoring indicators of currency crises (Sarlin, ). In addition, we also show examples of how the transition information can be used in practice.…”