The Effectiveness of Expectation Models in Recognizing Error Patterns and Generating and Eliminating Hypotheses While Conducting Analytical Procedures

Abstract

This study analyzes the hypothesis generation and elimination capability of analytical procedures using Structural, Stepwise, Martingale, and ARIMA (Auto Regressive Integrated Moving Average) expectation models. We seed 14 errors into 36 complete sets of actual monthly financial statements from three companies. The bivariate pattern of differences, along with the structure of the accounting, business process, and economic system, are used to analytically determine (hypothesize) the most likely cause of the error. We then test the capability of these expectation models to generate correct hypotheses or to eliminate incorrect hypotheses. Positive and negative testing approaches, founded on multivariate normal theory, are examined. From a hypothesis generation perspective using the positive testing approach, the results indicate that the Structural and Stepwise models, yield lower effectiveness risks (type II error) than the ARIMA and Martingale models, with the edge going to the Structural model. From a hypothesis elimination perspective using the negative approach, the results indicate that the Structural and Stepwise models yield lower efficiency risks (type I error) than the Martingale and ARIMA models, with the edge going to the Stepwise model. This study provides strong evidence to support the use of the structure of an organization's business processes (McCarthy 1982; Bell et al. 1997), its associated accounting system, and economic structure to build an expectation model. Moreover, the joint consideration of errors is found to be superior to the marginal approach advocated by Kinney (1987). The results presented here have the potential to assist auditors in directing audit efforts.