A comparative study between logistic regression and artificial neural network models for bankruptcy prediction on companies listed on Bombay stock exchange in the Indian market

1 Introduction 10
2 Literature Review 12
3 Research Goal 18
3.1 Main Research Goal 18
3.2 Research Hypotheses Statement 18
3.3 Research Questions 18
4 Methodology 20
4.1 Data 20
4.2 Data Analysis 24
4.2.1 Standardization 24
4.2.2 Logistic Regression 25
4.2.3 Artificial Neural Network 27
5 Findings 30
5.1 Evaluation score 30
5.2 Precision 31
5.3 Recall 31
5.4 F1-score Test 32
6 Conclusion 36
6.1 Final Interpretation 36
6.2 Limitations 38
6.3 Future scope and application 38
7 References 39
8 Appendix 41
Купить

Bankruptcy is a state of insolvency wherein the company or the person is not able to repay the creditors the debt amount. Bankruptcy prediction is of importance to the various stakeholders of the company such as creditors, investors as well as employees. One can argue that bankruptcy can have a huge impact on society, whether it be in its immediate surroundings or a global one.
Since the development of bankruptcy prediction models by Altman and Ohlson using the statistical empirical analysis methods in the 1980s, research in the field of bankruptcy prediction has only moved forward. There have been many methods developed and used across the industries. Some of the more common methods are the Altman Z score (Altman, 1968) and the Ohlson’s logit regression model (Ohlson, 1980). Some with the advancement of technology utilize the application of software-based computations, such as Machine learning and Artificial Intelligence. Each model has its own advantages and own limitations. The research in the field of Bankruptcy prediction has always been an attractive spot for comparative studies. For a large part of time the research has been dominated by Logit and Probit regressions. However, in recent decades that torch is seen marching with the Machine learning techniques.
Development of Neural networks such as ANN, CNN, Deep learning and artificial intelli-gence have research moving away from statistical methods of empirical analysis (Salchenberger, et.al 1992). Here the question arises whether machine learning is the true answer to bankruptcy prediction model development? or do the statistical models such as logit regression still command a significant position? And can it be applied in every situation? in any type of economic condition and markets across the globe? It is necessary that we develop methods to identify firms that might run a risk of going bankrupt. The development of a model that is easily compatible yet efficient in terms of accuracy will be beneficial for all the stakeholders of a business. It is also imperative that the method is simple, applicable across industries and consistent in predicting bankruptcy.
There have been many studies in the past regarding the efficiency of prediction models. Attempts to find the best prediction model have been satisfactory but none of have been very successful. Moreover, most of these studies have been on a global scale and concentrate more on firms that are huge multinationals. Thus, previous attempts and studies have lacked comprehen-siveness. None of them provide a single viewpoint on which prediction method is best suited for an industry or which method is superior and why. The available comparison of different bank¬ruptcy prediction models in most of the studies deal with similar economic conditions and thus do not account of different markets or varying macro-economic conditions. Moreover, there is a con-siderable research gap in bankruptcy prediction studies considering the Indian market.
The purpose of this research is to study the performance of major bankruptcy prediction models by applying them to companies in the Indian market. This thesis involves comparing the performance of the logistic regression model and artificial neural network model in predicting bankruptcy for Publicly listed companies. This study incorporates data collection, date transfor-mation and data analysis pathway. The data is collected in form of 27 financial ratios spread across 5 years prior to the company’s bankruptcy. Data sample also includes financial healthy companies in equal numbers to the bankrupt ones. The data is divided into five data sets each corresponding to respective years prior to bankruptcy. Since the data is being collected from Publicly listed com-panies, delisting due to liquidation is considered as the definition of bankruptcy for the purpose of this study. The financial data is standardized and then used to train and develop the models. Once developed a classification report was generated to access their prediction capabilities and then compared. MS excel was used for data collection and all other analysis were carried out using the Python programming language in the Jupyter Notebook of the Anaconda software with relevant functions such as scikit-learn, TensorFlow, and keras.
Купить