DescriptionThe transaction-level implementation of a continuous auditing framework requires the analysis of exceptions and anomalies. Scholars used various data analytics tools to contribute to the evolution of the framework. This doctoral dissertation advances the existing literature with different risk aggregation methods, similarity-aware examination, and a confirmed instance-based feedback loop. The first chapter identifies the “risk position” of a transaction in the “risk space” and uses different distance metrics to calculate the aggregate risk score for the transaction. While variations of Minkowski distance allow users to balance the risk calculation from multiple filter values, the utilization of Mahalanobis distance considers the relationships among filters and adjusts for correlations during the aggregation of filter values.In the second chapter, I propose using similarity constraints during the audit selection procedures to avoid selecting accounting items that represent similar risks which may cause suboptimal use of some audit resources. I introduce two relevant objectives to the audit analytics literature: risk maximization and similarity minimization. Following these objectives, I develop two algorithms: the Skipper and the Stretcher, each prioritizing one of the objectives while holding the other objective bound by a user-defined threshold.
In the third chapter, I contribute a feedback loop method to the existing literature on continuous audit that learns and updates the boundaries of the high-risk subspaces by analyzing the confirmed instances from the investigation of exceptions and outliers. Additionally, this feedback loop also recommends filters that might be missing from the transaction verification module. To evaluate the model, I run a simulation of a multiperiod continuous audit framework with a proposed feedback loop model, learning and adjusting the filters of the transaction verification throughout nine periods of data. This model allows auditors to update filter thresholds, add new filters or remove filters that became obsolete.