Tadayyon Chaharsoughi, Marjan2019-02-272019-02-272018https://hdl.handle.net/10182/10504The New Zealand financial markets experienced a horrific loss over a three year period (from 2006 through to 2009). This was due to the non-bank deposit taking sector, which resulted in the failure of over 45 finance companies. This categorical failure resulted in losses over NZ$3 billion, impacting between 150,000 and 200,000 depositors (in a country where total share market capitalisation is approximately NZ$75 billion). This event inevitably raised concern over the quality of financial reports, audits and the role of trustees in the New Zealand financial system. While several scholars have examined the internal and external factors which caused this financial crisis in an attempt to attribute blame or responsibility, this study investigates the underlining variables that contributed to the financial companies’ failure, from three years ahead, to achieve the best prediction model. The study uses a sample of 35 failed finance companies and covers the period of 2005 – 2009. It compares these companies with the same number of non-failed companies, matched based on asset size. In particular, it studies their financial and non-financial variables for three consecutive years before failure. Logit and hazard models are used in order to identify a suitable model for prediction. The models were developed based on three different variable categories. The first model incorporates only CAMELS financial variables. While the second model uses only Agency-related variables, the third model is expanded to amalgamate both CAMELS financial and non-financial variables. The study examines the accuracy of the models by testing through out-of-sample data that has been collected from failed financial institutions after 2010, a period characterised by the implementation of new regulations. Error table, AUC, Gini, KS and H measures are used to assess the performance of the models. The overall results of this study suggest that there is sufficient evidence to support the claim that both financial and non-financial indicators are inferior among failed finance companies when compared to non-failed companies, for three years before failure. The results strongly support the argument that the integration of both groups of variables increases the performance of the model in predicting failure in both the logit model and the hazard model. Therefore, the combined model performs better than the models built solely on financial or non-financial variables. The study also verifies that the overall accuracy of combined models is more than 85%, with the highest accuracy of 89.70% at three years before failure, in the logit model. Furthermore, the lowest level of false negative and false positive, and the AUC, Gini, KS and H statistics, which measures the discriminant power of the models, show the combined model at three years before failure has the best discriminative power.enhttps://researcharchive.lincoln.ac.nz/pages/rightsCAMELS ratingcorporate governancecorporate failurefinancial institutionshazard modelThesisANZSRC::1501 Accounting, Auditing and AccountabilityANZSRC::1502 Banking, Finance and InvestmentQ112938380