Comparative performance analysis of artificial neural network optimization algorithms on the estimation of bank return on equity

It is always important to measure the performance and profitability of the banking sector, which is the most important part of a country's financial system. Thanks to performance measurement, banks can evaluate the competitive situation, growth potential and risk, and they can be more successful to continue their activities. In the literature, artificial neural networks (ANN) have rarely been used for evaluating economic performance. Therefore, in the present thesis, it is aimed to examine the use of ANN in depth to estimate the performance of banks. The aim is to estimate returns on bank equity by using ANN calculation methods for given a set of financial parameters. The thesis was carried out by considering all deposit banks in Turkey. Quarterly data for 11 years of 6 external and 8 internal parameters, which are considered as independent variables, were used. The dataset was created based on information of Banking Regulation and Supervision Agency, the Banks Association of Turkey, Turkey Statistical Institute and the Central Bank of the Republic of Turkey. The success of various ANN optimization algorithms in predicting the bank return on equity, which is regarded as the dependent variable, has been compared. As a result of the numerical experiments conducted, it was seen that all the ANN methods predicted the return on equity with an accuracy of over 80%. Also, it has been observed that the best ANN method is different for each bank. Comparison with linear regression method showed that ANN methods are more successful.