Analytical Procedures Phase of PCAOB Audits: A Note of Caution in Selecting the Forecasting Model

Abstract

The best-practices execution of PCAOB audits requires the use of Analytical Procedures at the Planning and the Substantive Phases. This often finds the auditor using the standard OLS two-parameter linear regression forecasting model [OLSR] to project account-values from the Planning Phase to balances expected at Year-End so as to effect a variance analysis at the Substantive Phase. This is the point of departure of our study. We examine the practical effect of using the OLSR model in a time-series context of the audit. Specifically, this research report provides information on the use of the OLSR model as the model of choice in the audit context compared to the ARIMA(0,2,2)/Holt model which is usually the standard choice for an exponential smoothing model in the presence of autocorrelation of data in the time-stream; autocorrelation is the usual case for longitudinal series taken in the audit. Results: We find that there are reasons to condition the selection of the forecasting model in the Analytical Procedures context based upon autocorrelation in the data-stream. When the time-stream of data exhibits autocorrelation the OLSR model fails in a statistically significant manner to capture the next or one-period ahead client value at the same rate as does the ARIMA/Holt model. This then has implications for the False Negative Investigation Error.