Model-Based Digit Analysis for Fraud Detection Overcomes Limitations of Benford Analysis

Benford Analysis is a statistical method used for detecting financial fraud. It compares the distribution of digits in data with the Benford Distribution. But there are often disadvantages ranging from uncomfortable rates of false positives up to total inapplicability of the method. We identified the inaccurate fit of typical data to the Benford Distribution as reason for these deficits. So we propose to use adaptive distributions of digits instead. For that we introduce a procedure which derives the distribution of digits from a ``model´´ for the distribution of data. The term ``model´´ means an abstract distribution which reflects basic properties of the data. This paper identifies different models and analyzes their relevance and performance. We show that model-based Digit Analysis provides a more reliable and more generally applicable tool for fraud detection to auditors.