Speakers: Dr. Daniel Manrique Vallier, Dr. Andrew Womack, Lei Ding, Miguel Pebes Trujillo, Department of Statistics, Indiana University
Title: The Deloitte & Touche Project for Statistical Decision Analysis in Auditing
Abstract: Audits attempt to discover financial errors and improprieties. Standard auditing practices, which have not advanced substantially since the 1980s, rely heavily on random sampling and lack principled ways of incorporating additional information (e.g., historical data, auditor intuition) into audits. The Project for Statistical Decision Analysis in Auditing, a collaboration between the accounting form of Deloitte & Touche and Indiana University, seeks to advance the development of new auditing methods, leading to a new generation of auditing practices.
An audit results in a statement that expresses the extent of the auditing firm's belief in the accuracy of a collection of financial records. In the language of statistics, such a statement describes a subjective probability, a hallmark feature of the Bayesian approach to statistical inference. Bayesian inference provides an intellectual framework in which multiple sources of information can be combined to update prior information about a given proposition, e.g., that the collection of financial records contains no substantive errors. This framework depends on modeling the sources of information as probability distributions, then using the mathematical properties of probability to draw conclusions about the propositions of interest. While several papers in the 1980s noted the relevance of Bayesian inference to the concerns of auditing, there have been no serious attempts to implement Bayesian methods in the field.
To develop a Bayesian approach to auditing, it is necessary to construct a mathematical model of auditing procedures and the phenomena they investigate. Doing so is a highly nontrivial undertaking that requires the substantial involvement of highly trained statisticians with special expertise in developing such models.
