© 2015 Research Academy of Social Sciences http://www.rassweb.com 135 International Journal of Financial Economics Vol. 4, No. 3, 2015, 135-149 Business Failure Prediction: An emperical study based on Survival Analysis and Generalized Linear Modelling (GLM) Techniques Alhassan Bunyaminu 1 Abstract This study investigated business failure using financial (ratios) and non-financial factors of listed companies on Ghana Stock Exchange. A combination of quantitative and qualitative variables have been used to predict corporate failure. Quantitatively, the study used 19 corporate determinants as predictors of business failure of listed companies on the Ghana Stock Exchange. The qualitative analysis used managerial (non-financial) factors to determine the success or otherwise of a business. An initial sample of 22 companies was divided into 70% estimation sample, 30% holdout sample and the overall prediction for a cumulative three-year data set.The study used the Cox proportional Hazard of Survival analysis technique and Generalised Linear Modelling (GLM) to predict business failure with an appreciable degree of accuracy. To reduce the dimentionality of the initial data space, the study initially used Factor Analysis(FA) by transforming a number of possibly correlated variables into a smaller number of uncorrelated variables called factor components. The study further employed Generalised Linear Modelling (GLM) with its three link functions-the Logit model, the Probit model and the Complementary log-log (Clog-log) function. Among the three link functions of GLM, the logit model provides the highest overall accuracy with the lowest Akaike Information Criteria (AIC): 46.456. The study reveals that among corporate determinants, the most significant variables that appear as consistent indicators of financial distressed companies in the superior model (logit model) are Profitability ratio (Return on total assets) and Leverage ratio (Solvency, Gearing and Interest cover). In connection with non-financial (managerial) factors used in the study, the study finds age and years of experience of managers as significant factors contributing to survival. The study recommends that future research should focus on business failure prediction that is based on tri-dimensional approach instead of the binary classification approach. Keywords: Distressed companies, Non-Distressed companies, Principal Component Analysis, Dichotomous approach, Generalized Linear Modelling, multi-state approach, Survival Analysis. 1. Introduction Business Failure Prediction (BFP) has generated immense interest in research and practice culminating into making deep inroads into the subject matter (Oki, 2004). The investigations of corporate failure prediction research (Altman 1983; Ballantin 1992; D’Aveni 1989; Dugan and Zavgren 1989; Koh and Killough 190; Pech and Alistair 1993; Shumway 2001; Chava and Jarrow 2004; Raza et al., 2011; Bunyaminu and Issah 2012; Bunyaminu and Bashiru, 2014) mostly determine the backrupcy probabilities of a firm using a number of covariates in the form of financial ratios, market related variables or non-fiancial factors such as managerial factors. Previous empirical studies on failure prediction have focused almost exclusively on financial ratio data that are mostly based on quantitative models. However, some scholars question the usefulness of ratio-based business failure prediction models (Alves 1978; Corman and Lussier 1991; Gilbert, Menon, and Schwartz 1990; Bunyaminu and Bashiru, 2014). For instance, El-Zayaty (1986) found ratio models to be unreliable predictors of bankruptcy based on his study of 132 businesses on the subject of corporate failure prediction. 1 Banking and Finance Department, University of Professional Studies, P.O. Box LG 149, Legon, Accra, Ghana