Machine Learning for Predicting the Procurement of an Audit at Small Private Banks: Is the Decision to Procure an Audit Systematic?

Abstract

Absent regulatory requirements, 64% of small private commercial banks voluntarily procured an independent audit in our dataset. Hence, it can be argued that an independent audit may have a perceived value. In our study, we examine whether the decision to procure an audit is systematic. First, we model and predict the decision to procure a voluntary audit using classification learning algorithms. We find, in particular, that the IBK and Random Forest learning algorithms have the highest cross-validated prediction accuracy of 81% and 80% respectively, thus supporting the claim that the decision to procure an audit is not made arbitrarily, but rather systematically, and therefore is amenable to analysis. Second, we use logistic regression to identify those characteristics that may systematically influence the decision to procure an independent audit. Generally, we find that bank size, profitability, growth, leverage, complexity of operations, and type of ownership may systematically influence the decision to procure an independent audit. In contrast, banks that are less capitalized are less likely to procure an audit.

Keywords: Audit Demand, Data Mining, Machine Learning, Classification Problem, Commercial Banks, FDIC Call Reports

JEL Classification: M40, M41, M49