Comparing Dissimilarity Measures: A Case of Banking Ratios

The aim of this paper is twofold. Firstly, to discuss a clustering of a given set of the European banks into groups based on their performance during 1999–2013. Secondly, to compare different dissimilarity measures and to determine which of them suits best for clustering banking ratios. Six ratios that reveal profitability, efficiency, stability and loan portfolio quality of the banks were used. The similarity/dissimilarity between banks was estimated using measures that are based on time series or functional data properties. Two dissimilarity measures that are not commonly used in the literature are proposed and two measures are extended from univariate into multivariate case. The results of our study show that there is no dissimilarity measure which would provide the best clustering results for all ratios. However, dissimilarity measures based on functional data properties in many cases outperfomed measures based on time series properties. The choice of the number of clusters is not that clear. According to different banking ratios, it is found that banks could be grouped into 6–12 clusters.