SOURCE:

Faculty: Management Sciences
Department: Accountancy

CONTRIBUTORS:

Egbunike, F.C;
Ekwueme, C.M;

ABSTRACT:

The main objective of the study is to compare the predictive accuracies of four bankruptcy prediction models for Nigerian manufacturing firms. The study specifically developed a model for bankruptcy prediction using Genetic Algorithm (GA) and compared its performance with two traditional techniques (Logit and Discriminant models); and, one artificial intelligence technique (Neural Network model). The study also evaluated whether the predictive accuracy of GA model improved from the inclusion of corporate governance variables. The study is multi-theoretical; and, anchored on three theories: theory of natural selection, anthropomorphism and agency theory. The study was conducted from a positivism paradigm. The study made use of ex-post facto research design. The population for the study comprised one hundred and sixty four (164) quoted firms on the Nigerian Stock Exchange (NSE) as at end of 2017 financial year. The sample was determined as sixty-six (66) companies after exclusion of firms in financial services, natural resources and oil & gas sectors. The study employed a variant of non-probability sampling method the purposive sampling technique in chosen the sample. The data for the study were obtained from secondary sources. The study used four techniques for prediction of bankruptcy; logit, discriminant, neural network and genetic algorithm models. The McNemar test was used to compare the accuracies of the models. The results showed significant difference in the classification accuracies of the GA (96.94%; 97.85%) compared with the logit (93.4%; 93.6%), discriminant (91.1%; 90.9%), and neural network (92.2%; 94.4%) models. The accuracies of the models were slightly improved upon by the inclusion of selected governance variables. The practicality of using the GA in selecting the best set of predictors was also confirmed. Based on this, the study recommends deployment of GA in determining the best set of predictors for manufacturing firms. The study also recommends the use of an alternative model, such as the logit or probit models in benchmarking the performance of machine learning model.