شناسایی اهداف مالکیت در صنعت بانکداری اتحادیه اروپا: استفاده از روش چند معیاری

In this paper we develop classification models for the identification of acquisition targets in the EU banking industry, incorporating financial variables that are mostly unique to the banking industry and originate from the CAMEL approach. A sample of 168 non-acquired banks matched with 168 acquired banks is used over the period 1998–2002, covering 15 EU countries. We compare and evaluate the relative efficiency of three multicriteria approaches, namely MHDIS, PAIRCLAS, and UTADIS, with all models developed and tested using a 10-fold cross validation approach. We find that the importance of the variables differs across the models. However, on the basis of univariate test and the results of the models we could state that in general after adjusting for the country where banks operate, acquired banks are less well capitalized and less cost and profit efficient. The results show that the developed models can achieve higher classification accuracies than a naïve model based on random assignments. Nevertheless, there is fair amount of misclassification that is hard to avoid given the nature of the problem, showing that as in previous studies for non-financial firms, the identification of acquisitions targets in banking is a difficult task.

The purpose of this study is to evaluate the performance of multicriteria decision aid (MCDA) prediction models developed specifically to identify acquisition targets in the banking industry, an area that is relatively under-researched1. Of approximately 30 papers that we can identify in the literature which utilised one or more methods for the prediction of acquisition targets, all but one (Pasiouras & Tanna, 20062) have focused on samples of firms drawn from the non-financial sectors (i.e. manufacturing, retail, hospitality, etc.) and excluded banks from their analysis. One reason for their exclusion is the unusual structure of banks' financial statements suggesting that certain bank-specific characteristics distinguish them from other corporations (Bauer & Ryser, 2004). In line with Pasiouras and Tanna (2006) the present paper utilises financial variables that originate from the CAMEL3 approach in developing prediction models that distinguish acquired from non-acquired banks, based on a sample of commercial banks covering 15 EU countries4 (the former EU15). Most of the past studies have used multivariate statistical and econometric techniques such as discriminant analysis (e.g. Barnes, 1990 and Stevens, 1973) and logit analysis (e.g. Barnes, 1998, Barnes, 1999 and Powell, 2001) and only more recently the parametric nature and the statistical assumptions/restrictions of those approaches have led researchers to the application of alternative techniques such as artificial neural networks (Cheh, Weinber, & Yook, 1999), rough sets (Slowinski, Zopounidis, & Dimitras, 1997), recursive partitioning algorithm (Espahbodi & Espahbodi, 2003) and multicriteria decision aid (MCDA) (e.g. Doumpos, Kosmidou, & Pasiouras, 2004). Some of these studies focused on the search of the best predictive variables (e.g. Bartley and Boardman, 1990, Cudd and Duggal, 2000 and Walter, 1994) and others on the search of the most effective empirical method for the development of the prediction models (e.g. Cheh et al., 1999, Doumpos et al., 2004, Espahbodi and Espahbodi, 2003 and Slowinski et al., 1997). The present paper has two overall objectives that cover both categories mentioned above. First, it aims to jointly investigate the efficiency of three MCDA techniques. Second, it attempts to reveal the factors that contribute in the identification of acquisitions targets. The major advantages of the MCDA over the traditional techniques are that they do not make any prior assumptions5 about the normality of the variables or the group dispersion matrices (e.g. discriminant analysis), and that they are not sensitive to multicollinearity or outliers (e.g. logit analysis). Furthermore, MCDA techniques can easily incorporate qualitative data, while they are also very flexible in terms of incorporating preferences of the decision-maker. The rest of the paper is as follows. In Section 2 we first describe our sample of EU commercial banks and explain the cross validation procedure for developing and validating the models6. We then provide a detailed discussion of the financial variables representing bank-specific characteristics that we consider appropriate for the identification of bank acquisition targets. The three MCDA approaches are described in Section 3, while Section 4 discusses the empirical results. Finally, Section 5 presents our concluding remarks on the use of the MCDA models for identifying acquisition targets and highlights some issues for further research.

In this study we developed MCDA classification models for the identification of acquisition targets of commercial banks operating in the EU. The sample consisted of 336 banks operating in the EU, of which 168 were acquired between 1998 and 2002. Eight variables most of which originate from the CAMEL model (and representing seven potential motives for banks acquisitions) were selected for inclusion in the models. Since the sample was drawn from 15 EU countries, the individual banks' ratios were transformed to industry-relative ratios (by dividing the values of the variables of the individual banks with the corresponding average values of the commercial banking industry in the country where the banks operated). The models were developed using three multicriteria decision aid techniques, namely MHDIS, PAIRCLAS and UTADIS. A 10-fold cross validation approach that allows the maximum use of the available data while it ensures the proper evaluation of the models, was used to develop and validate the models. The ability of the models was assessed by comparing their classification accuracies, in terms of the percentage of banks correctly classified in each group. Such models can prove useful to managers who are interested in a decision tool that could allow them to identify potential candidates among a large set of banks and proceed to a more detailed examination of the ones that are closer to the typical profile of an acquisition target. Furthermore, as Curry (1981) mentions, bank regulators might be interested in the development of such models, which could be useful in forecasting the degree of competition in the market. Obviously, the efficiency of the model, depends not only on the specifications and flexibility of the classification technique, but also on whether acquired banks have unique characteristics that distinguish them from non-acquired banks. Our results indicate that the characteristics that can be useful in identifying the acquisition targets may differ across the techniques used to develop the models. However, this is not surprising and has been the case in past studies as well. One possible explanation is that, although all methods are using the same information (i.e. in terms of the data set and variables employed) and they have the same objective (i.e. correct classification), each one of them processes the information differently, due to differences in the procedures for solving the problem. For example, while UTADIS and PAIRCLAS develop only one utility function characterizing all banks, MHDIS develops two functions, each describing one of the two groups. Furthermore, UTADIS solves one linear programming while MHDIS solves three mathematical programming formulations (i.e. two linear and one mixed-integer). Finally, PAIRCLAS operates on the basis of pairwise comparisons rather than the preference disaggregation approach. On the basis of the contribution of the criteria (in terms of weights) in the three models and the univariate results we could conclude that after adjusting for the country where banks operate, non-acquired banks were better capitalized. We also found evidence to support the inefficient management hypothesis, as acquired banks were characterized by lower profitability and less efficiency in expense management. Liquidity and loan activity also appeared to be important in characterizing acquired banks in the MHDIS model, while GROWTH was more important in PAIRCLAS. Turning to the classification ability of the models, the average results over the 10 replications in the validation set showed that all models performed better than a naïve model based on random assignment to outcomes based on prior probabilities (i.e. 50% in an equal sample). Nevertheless, there is fair amount of misclassification, which is hard to avoid given the nature of the problem. The highest overall classification accuracy was obtained by MHDIS (65.7%), followed by PAIRCLAS (63.8%), and UTADIS (61.6%). However, it should be noted that the superiority of any classification procedure may be context or sample specific (Espahbodi & Espahbodi, 2003). Nevertheless, the non-parametric MCDA approaches have certain advantages over the parametric approaches, in that they do not require any assumptions and can easily incorporate qualitative variables, which leads us to conclude that they can be considered as a reliable alternative to the traditional statistical techniques. One potential shortcoming of the study as it is the case in many classification problems in finance (e.g. bankruptcy, credit risk) is that the usefulness of the model might be limited to the countries for which it was developed19 (i.e. EU-15). In our case, the problem might be even more serious since, as previously mentioned the nature of acquisition activity and the dominant motivations for acquisitions may differ across countries. However, it should be emphasized that the overall framework (i.e. variables selection process, selection of classification techniques, development and evaluation process) can be easily adopted while using data from other countries (e.g. US, Asia) by re-estimating the criteria weights in our models. Future research could extend the present study towards various directions such as the inclusion of additional non-financial variables (i.e. ownership type, manager's experience or technological capacity), the testing of the usefulness of the models for other countries, the employment of and comparison with alternative techniques (i.e. multidimensional scaling, neural networks, etc), and the combination of MCDA and other techniques into integrated models.