non-parametric methods

The purpose of this research is to measure the efficiency of selected Iranian banks from 2010 to 2016. We estimate efficiency, productivity and efficiency changes under the Bayesian setting. In this research, we consider the Non-Performing Loans (NPLs) of banks as bad outputs, while good outputs are loans, off balance sheet assets and securities. Also, variables such as number of employees, bank capital, fixed assets and deposits play inputs role. Results show that the productivity changes of selected Iranian banks are positive over the period of the study, which is mainly due to the improvement in technology like e-banking services, while efficiency changes continued to be negative over the same period. It is observed that the average efficiency for banks is 85 percent. The Karafarin bank has the largest efficiency among selected banks while Refah Kargaran bank efficiency has the smallest one. Results also show that, during the specified period, private banks have a few better efficiency estimate in comparison with the efficiency of state owned banks.

This study aims to investigate the effects of accounting and non-accounting indices on financial distress prediction and also to compare parametric and non-parametric methods. Therefore, the sample consists of 211 distressed firms selected by special distress criteria and 211 healthy firms listed in Tehran Stock Exchange during 2006-2015. This study applies 32 accounting and 20 non-accounting indices and uses 2 parametric methods including Logistic Regression and Multivariate Discriminate Analysis and 7 non-parametric methods including Support Vector Machine, Artificial Neural Network, Decision Tree (with 4 algorithms) and Bayesian Network to predict financial distress. The results show that the models extracted from accounting indices have significantly more predicting accuracy than those from non-accounting indices, and adding non-accounting indices to the models based on accounting indices does not significantly increases their predicting ability. Also, since the average predicting ability of non-parametric methods is more than parametric ones, this difference is not statistically significant.