An Assessment on Banking Returns with Data Mining Approach: A Critical Investigation on Deposit Banks in Turkey

Abstract

The latest advances in performance assessment techniques in finance based literature spotlighted the importance of qualitative and quantitative hybrid models for an effective investment decisions. In this study, it has been aimed to analyze the returns of deposit banks in Turkey by utilizing two data mining technique namely Classification and Regression Trees (CART) and Multi-Layer Perceptron (MLP). For this purpose a dataset containing 47 variables of 29 deposit banks in Turkey for the period of 2004–2014 are collected. For choosing the most relevant features for the banking returns, CART method, which provides a variable importance measure, is utilized. For the classification purpose on the dimensionally reduced dataset, MLP method, which is considered a powerful tool, is employed. Results indicate (i) two of the most important independent variables for classifying banks according to ROA is (“Income Before Taxes/Total Assets”) and (“Net Operating Income(Loss)/Total Assets”) having the values of 100% and 93.7% respectively, (ii) two of the most important independent variables for classifying banks according to ROE is (“Net Profit (Losses)/Paid-in Capital”) and (“Income Before Taxes/Total Assets”) having the values of 100% and 86.2% respectively, (iii)using CART is a very convenient way for feature selection step in data mining and MLP is an efficient tool for using classification in finance domain.