Evaluation of financial failure of firms with bayesian multinominal logistic regression analysis: An application of manufacturing sector

This thesis aimed to predict the failure of the manufacturing industry firms operating in Borsa Istanbul. To do this, 157 manufacturing sector firms operating in Borsa Istanbul in 2018 were discussed. The reason for choosing the manufacturing industry in practice is that the manufacturing industry is one of the main sectors that contribute to sustainable economic growth. The manufacturing industry sector, which has a critical role in determining the national income per capita, has critical importance in the development of countries as it also affects other sectors along with technological developments. Moreover, the manufacturing industry is the crucial subsector of the industrial sector with the added value and employment it provides to the country's economy. A successful manufacturing industry plays a key role in keeping the country's economy alive and contributing to economic development. For this reason, the financial success of manufacturing industry firms has been frequently discussed and various models have been developed in the finance literature. The most well-known of these models is the Z-score model developed by Altman (1968). In the study, the dependent variable has been divided into three categories using the interest coverage ratio and the equity ratio. Thus, with the multinominal logistic regression model chosen in accordance with the multicategory dependent variable created, the financial failure of the firms was tried to be predicted one year in advance with the help of the financial ratios in the balance sheet and income statements. By using the financial ratios of the previous year, it is aimed to forecast the situation of the firms that will continue their financial success in the next year and will have difficulties in terms of flow or stock. Therefore, in the study, two-stage estimation has been used by classical multinomial logistic regression method and Bayesian multinomial logistic regression method which is thought to make more successful estimations in small samples. Various prior distributions have been put into account for Bayesian multinomial logistic regression models and Metropolis-Hasting algorithm and Gibbs sampling, which are Markov Chain Monte Carlo methods, have been used. After that, the correct classification rates of the estimated models have been compared. According to the findings, the highest correct classification rate, 82.17%, is obtained from the Bayesian model constructed with the Cauchy distribution, and it has been observed that this model produced more successful predictions than the alternative classical and Bayesian multinomial models.