AbstractFinancial distress is a critical social and economic problem that affects innumerable businesses the world over. Consequences of such an occurrence can go beyond the business owners and stakeholders – as was evident in the 2008 Global Financial Crisis (GFC), it can lead to a much larger macroeconomic calamity. Therefore, having the power to predict – and hence aid businesses from failing, has the potential to save not only the business, but whole economies from collapsing. This research’s academic contribution is to advance the field of Financial Distress Prediction (FDP) by tackling this issue from multiple angles – each being explored in a separate chapter – including: industry-specificity, index development, Islamic banking, variables affecting bankruptcy, class imbalance in data-sets, and Large Companies (LCs) vis-à-vis Small and Medium Enterprises (SMEs). This was achieved through utilising cutting-edge machine learning techniques, such as: Artificial Neural Networks (ANNs), Decision Trees (DTs), Random Forests (RFs), and Stochastic Gradient Boosting (SGB); and comparing their outcomes with results achieved from using well-established benchmark statistical techniques, such as: Multivariate Discriminant Analysis (MDA) and Logistic Regression (LR).
Two major databases were used in this thesis to extract more than 60 explanatory variables derived from financial statement data pertaining to thousands of existing and failed Australian and international companies across various industries in the marketplace. The extracted data were used to test for the validity and predictive power of the developed statistical models. The results in Chapter 3 empirically showcase that industry-specific models are superior to a one-size-fits-all model. The chapter also presents the most important variables in predicting financial distress pertaining to each industry. The results in Chapter 4 show that all FDP models built using machine learning techniques outperform a model built using the traditional LR statistical technique. Chapter 5 reveals that FDP models built using a data-set via the Synthetic Minority Oversampling Technique (SMOTE) outperform those using a standard data-set that is imbalanced. Chapter 6 presents a series of novel and user-friendly FDP indices that provide a standardised score for companies according to their success or distress potential. Chapter 7 explores the differences between conventional and Islamic banking, then proceeds to build FDP models using machine learning techniques, each with a different measure of Islamic banks’ financial distress. The aim was to present the most important variables in forecasting financial distress relating to Islamic banks. Chapter 8 creates FDP models using machine learning techniques on data-sets comprised of LCs and SMEs that are listed on the Australian Stock Exchange (ASX). These models are then compared with models that were built using data that have been SMOTEd, in order to establish the empirically superior FDP model, as well as outlining the most important variables in determining the successes or failures of SMEs and LCs.
The multifaceted approach used in this dissertation contains many important practical contributions, including: aiding lenders in accurately determining the economic viability of providing loans to prospective borrowers, offering investors with invaluable insight on their existing and/or potential investment, enabling governmental agencies to monitor businesses with high chances of bankruptcy, and providing managers and decision makers with invaluable insight to be used in conjunction with their expertise, in order to install proactive measures to mitigate the chances of falling into financial distress. These benefits have the potential to assist whole economies from falling into a recession as a result of increased business failure.
|Date of Award
|12 Oct 2019
|Kuldeep Kumar (Supervisor) & Adrian Gepp (Supervisor)