Auditors’ decision-making aid for going concern audit opinions through machine learning analysis

Abstract

Prior going concern studies often use regression techniques. Such techniques do not often examine the complex intertwined relationships between factors and therefore have limited value as a decision process aid. However, this study overcomes these limitations by employing a hierarchical machine learning method, a decision tree model, to discover potential interactions to create an understandable decision aid. This research explores the complex interactions between many factors that hold information about the auditors’ decision process. The findings also suggest that an indicator variable for a low return on equity (ROE) contained relevant information about the going concern decision, as well as indicator variables for low current ratios, a low stock price, and several new interaction variables. Through a “white box” machine learning method, this study discovers economically and statistically significant indicator variables, rules, and interaction variables to improve the understanding of the external audit decision process and to produce a usable decision aid for auditors and investors. Moreover, the simplicity and informative “white box” nature of decision trees makes this method a good approach both in future research and in practice to understand decisions and to produce decision aids.