Regression as an Analytical Procedure: Are Adjustments for Autocorrelation Really Necessary?

Abstract

Auditors who use regression analysis when performing analytical procedures may be exposed to the problem of autocorrelation, especially if small-interval time-series data are used. The existence of autocorrelation may bias the standard error of the estimate, which is used in calculating the upper precision limit. This limit determines the acceptance or rejection of an account balance as not materially misstated.

Using real-world data and the STAR decision rule, the results of this study suggest that transforming autocorrelated models increases effectiveness (capability of detecting material errors) only slightly, if at all. The results, however, did indicate that efficiency may increase significantly (reduced Type I errors), and that the more severe the autocorrelation, the more likely that efficiency will increase. Therefore, auditors have the most to gain in terms of audit-time savings when models are more severely affected by autocorrelation since fewer incorrect rejections requiring unnecessary additional audit attention would occur with successful transformation.