ACCTMIS 7510

Article Questions

Bell, T. B. and J. V. Carcello. 2000. A decision aid for assessing the likelihood of fraudulent
financial reporting. Auditing: A Journal of Practice & Theory 19(1) (Spring): 169-84.
Aids

1. Once again, don’t get lost in the details of the method and analyses. Concentrate on the main conclusions of the article and on the general method used to support the conclusions.

2. Like the Geiger, Raghunandan & Rama article, this study uses (1) both univariate and multivariate analyses and (2) logit (i.e., logistic regression) analysis. You may want to refer to your article 2 questions for descriptions of these.

3. You may skim the two sections on “Comparison of Decision Aid and Unaided Auditor Judgments” on pages 179-182.

Questions

1. On p. 170, the authors discuss the L/W model, which indicates that auditors should consider three components in assessing whether fraudulent financial reporting may occur: Conditions (C), Motivation (M), and Attitude (A).

a. Do you think that these components are best thought of in a multiplicative or an additive fashion? That is, do all three components have to be present in a particular audit engagement in order for fraud to occur?

b. Is the final model developed by Bell & Carcello (Table 4) consistent with a multiplicative model or an additive model (or both)?

2. In Table 3, two risk factors are significant in the “wrong” direction. For each of these factors, speculate as to why the presence of the factor might be associated more with nonfraud than fraud companies.

3. The authors mention (p. 177) that the selection of a cutoff point would be arbitrary in the absence of information about costs of misclassification.

a. Which type of misclassification would be more costly? Why?

b. Roughly estimate the relative magnitude of misclassification costs for an audit. Using your estimate, determine which cutoff point (using the estimation sample) would produce the lowest expected cost of misclassification. (Assume that correct classifications incur no cost.)

(more on back!)

4. An alternative way for auditors to assess the probability of fraud in a company is to employ a probability revision model such as Bayes’ theorem. In this model, RF refers to risk factors (such as the ones employed in the logistic regression model), F refers to fraudulent financial reporting, and NF refers to nonfraud.

P(F|RF) = P(RF|F)*P(F) ,

P(RF|F)*P(F) + P(RF|NF)*P(NF)

where: P(F) =probability that fraudulent financial reporting exists, assuming no knowledge of RF. (I.e., this is the rate of occurrence of fraudulent financial reporting in the population of companies. Also called the “base rate.”)

P(NF) =probability that fraudulent financial reporting does not exist, assuming no knowledge of RF.

P(RF|F) =probability that RF are present given that fraudulent financial reporting exists. This is also called the “hit rate.”

P(RF|NF)=probability that RF are present given that fraudulent financial reporting does not exist. This is also called the “false positive rate.”

P(F|RF) =probability that fraudulent financial reporting exists, given that RF are present. This is the probability that the auditor will use to decide whether to modify (i.e., extend) audit procedures to more actively search for fraud.

a. Using the information in Bell & Carcello’s article, calculate P(F|RF). (To do this, you will have to develop estimates of the RHS variables from the article. Use Figure 1 to develop estimates of P(RF|F) and P(RF|NF). Base your estimate of P(F) on footnote 3.)

b. Speculate why the probability of fraud you calculated in part a. is so much lower than the probabilities implied by Bell & Carcello’s model.