Egbunike, Francis Chinedu PhD1 Department of Accountancy, Nnamdi Azikiwe University firstname.lastname@example.org
Anachedo, Chima Kenneth email@example.com and Echekoba, Felix Nwaolisa
PhD3fn.firstname.lastname@example.org Department of Banking and Finance, Nnamdi Azikiwe University
Ubesie, Cyril Madubuko PhD4 Department of Accountancy,
Enugu State University of Science & Technology email@example.com
Purpose: The study compares the predictive accuracies of two bankruptcy prediction models for Nigerian manufacturing firms. Specifically, the study compares a model for bankruptcy prediction developed using Genetic Algorithm (GA) and an artificial intelligence technique (Neural Network model).
Design/methodology/approach: Utilising a variant of non-probability sampling, the purposive sampling technique used in the study selects a final sample of sixty-six (66) companies after excluding firms in the financial services, natural resources, and oil & gas sectors. The study relied on secondary data from annual financial statements. The study used two techniques for prediction of bankruptcy: the neural network and genetic algorithm models. The McNemar test was used to compare the accuracies of the models.
Findings: The model results showed significant difference in the classification accuracies of the GA (96.94%; 97.85%) compared with the neural network (92.2%; 94.4%) models. In other words, the GA model outperformed the neural network model in corporate bankruptcy prediction. The inclusion of selected corporate governance variables showed slight improvements in the accuracy of the models.
Originality/value: The results demonstrate the practicality of using GA in selecting the best set of predictors in a different context with regulatory and institutional regimes. The current study from a positivism paradigm is also among the first to utilise the concept of Genetic Algorithm in bankruptcy prediction of firms in Nigeria. Limitations: Authors have suggested that the use of existing models is limited by the conditions in which they are developed (Zelenkov, Fedorova, & Chekrizov, 2017). Therefore, the development context of the GA model may limit its applicability to other sectors, more so the use of GA with different classification models would produce varying results.