ON THE PERFORMANCE OF ARTIFICIAL INTELLIGENCEMETHODS FOR FAILURE PREDICTION: EVIDENCE FROMISTANBUL STOCK EXCHANGE

The prediction of business failure is a widely studied subject in financialliterature. Many earlier studies on this topic employed statistical methods such asmultiple discriminant analysis, logit and probit topredict corporate failure usingpast financial data (especially the ratio data). However, there has been a recentsurge in academic interest in the use of artificialintelligence (AI) methods topredict financial distress. Numerous studies documented that AI methodsoutperform traditional methods. Majority of these studies used data fromestablished markets, the number of studies on emerging market data is ratherlimited and only a handful of studies employed Turkish data for analysis. Thisstudy aims to contribute to the literature by applying the artificial neural networksto predict deletions from Istanbul Stock Exchange (ISE) National 100 Index. Thesample is constructed using the quarterly fundamental data of the companies listedin this index the period between January 2008 and December 2012. We employedNeural Networks (NN), logit and probit to predict deletions from index onequarter before they have occurred. Results show that although the logit providesslightly better in-sample predictions, all of the methods fail to identify deletions inthe out-of-sample periods