Machine Learning Techniques in Determining the Dynamics of Corporate Financial Distress: An Empirical Treatment and A Comparative Analysis of Financial and Non Financial Micro Data of the Turkish Private Sector

ABSTRACT Recent financial crises and especially large corporate bankruptcies, have led bank managements and financial authorities to follow and monitor both financial and real sector risks, and to focus on firm failures. Bank of International Settlements, has therefore, taken the decision to include the necessity for banks to employ internal rating systems among BASEL II criteria. Thus, risk assessment and internal rating systems criteria would be made operational by the individual European Union banking systems, by the end of 2007, and January 2008 in Turkey, at the latest. Financial and operational information of the firms, makes up the input to the risk analysis. This information can be aggregated to portray the sectoral trends, and/or focused upon on a firm basis to understand firms’ financial behaviours. Finance theory summarizes firms’ risks under financial distress and firm failure. There have been a myriad of works under these two headings, particularly in the United States, after the Great Depression. While early studies have focused upon the differences in the financial ratios of financially sound and failed firms, especially with the advances in computing capacity, the last two decades have witnessed an increasing use of machine learning methods in the failure prediction. Therefore, machine learning methods can be considered as having great potential in failure prediction and being good candidates as decision aids for policy-making. This study considers financial distress and firm failure on theoretical grounds, gives a compact but elaborate explanation of machine learning schemes, and analyzes the results of these schemes run with data obtained from the database of Real Sector Data Division of the Central Bank. Cost sensitive learning was given special attention in the analysis. Keywords: Financial Distress, Firm Failure, Machine Learning, Classification and Cost Sensitive Learning. xiii