Information transfer in analytical procedures: A simulated industry knowledge-management approach

Analytical procedures in auditing are diagnostic tests performed by auditors to assess the reasonableness of recorded accounting numbers. Effective and efficient sets of analytical procedures result in reduced cost and risk to auditors and in increased reliability that accounting numbers are stated in accordance with the Generally Accepted Accounting Principles. This study extends the existing research in analytical procedures by allowing for learning from contemporaneous information transfers among peer companies. Prior research examined the prediction performance of analytical procedures using various statistical techniques and temporal disaggregation. In general, researchers found that the performance of analytical procedures improved when (1) a greater number of predictor variables were used, (2) more sophisticated models were exploited, and (3) more frequency data points were included within the prediction model.; In this study, the prediction performance and error detection of analytical procedures is examined within a framework of information sharing. Within each industry, peer companies are selected for each audit client by using a multi-step ranking procedure. Contemporaneous information is then transferred among these peers and is included in the prediction models. The prediction performance and error detection are evaluated using statistical regression. Furthermore, this study examines the prediction performance of peer models using artificial neural networks.; This study demonstrates that transferring contemporaneous peer specific data across the industry being audited improves predictions and error detection. It is documented that, by using this approach, many of the industries included within this study experience a significant increase in prediction accuracy. However, while, the use of peer data results in a substantial decrease in type II errors, an increase in type I errors is observed. The inclusion of contemporaneous peer data is shown to be especially beneficial for improving predictions for high growth companies and results indicate that there are potential benefits associated with the use of monthly prediction models in comparison to quarterly prediction models. Finally, predictions generated by using various artificial neural network architectures are found to be inferior to the predictions generated by statistical regressions.