Forecasting audit opinions on financial statements: statistical algorithm or machine learning?


This paper examines the applicability of different algorithms in forecasting the audit opinion on the financial statements of listed companies in Vietnam. We collected data from 492 enterprises listed on the stock market from 2016 to 2020 with 2460 observations, of which 154 observations have audit reports that are unqualified opinions, accounting for 6.26%. We use logistic regression algorithms, decision trees, and random forests. We consider two research models to assess the influence of factors, including groups of financial factors, factors belonging to the Board of Directors, and other factors on the audit report with an unqualified opinion. For model machine learning algorithms, the data is divided into two sets of Training and Testing with a ratio (of 80:20). The Testing dataset is used to evaluate the effectiveness of the predictive model. The results show that the audit opinion of the previous year has the most significant influence on the audit opinion, followed by profit after tax on equity, the ratio of receivables to revenue, and the business size. In particular, the ability to accurately predict the total non-acceptance audit opinion reaches 97% for the random forest algorithm. This study contributes to the current literature by examining which algorithm is appropriate for predicting the auditor's opinion. Furthermore, this research adds empirical findings to the literature on audit reports to make the financial statement audit process more efficient.

DOI Code: 10.1285/i20705948v17n1p133

Keywords: Audit opinion; Decision tree; Financial statements; Machine learning; Logistic Regression; Random forests

Full Text: pdf

Creative Commons License
This work is licensed under a Creative Commons Attribuzione - Non commerciale - Non opere derivate 3.0 Italia License.