Combining hidden Markov models with probabilistic Bayes networks to conduct business forecasting and risk simulation

This research integrates business analytics with enterprise risk management to help firms implement a decision support war-room. In particular, this study explores the hidden states of the small- and medium-sized enterprises (SMEs) which financial performances tend to intensively rise and fall. In practice, managers need to know what indicators can be good predictors to predict bankruptcy and what actions should be taken before bankruptcy? Thus, a novel framework is presented to achieve the following goals: (1) hidden Markov model is constructed to relate firms’ hidden states (healthy, risky, and sick) to observable variables (ROA—return on asset, EPS—earnings per share, and NPBT—net profit before tax). (2) Ensemble learning (EL) like random forest (RF), adaptive boosting (AB), and gradient boosting (GB) is used to conduct business forecasting. (3) Probabilistic Bayes network (PBN) is used to conduct stochastic risk simulation. Data samples collected from various industry sectors (semiconductor, optoelectronics, consumer electronics) are used to justify the validity of this research. Experimental results identify cash equivalents, gross sales profit, and operating revenue can efficiently enhance the SMEs from risky states to healthy firms. In contrast, operating profit margin and net income margin are more effective to improve the SMEs from sick states to risky firms. Managers are suggested to continually monitor and track these key indicators.