Detecting Fraud in the Data Using AID, Part Two


In the Jan./Feb. 2004 issue we examined the basics of automatic intervention detection – an advanced computer-based tool. Here we describe the methodology of a research study conducted to explore the effectiveness of AID. 

We developed a research study to evaluate whether automatic intervention detection could be used successfully to distinguish among companies with reported fraudulent data and those with no such reports. We selected eight companies which had been identified in the general media as having engaged in financial fraud. We downloaded newspaper articles from NEXIS outlining the specifics of each fraud. Table 1 lists the eight fraud firms and the type of fraud in which they were reported to have engaged. 


Company  Nature of the Fraud  
Cendant Inflated earnings and improper use of reserves
Con Agra Improper revenue recognition of fictitious sales
Enron Failure to disclose liabilities and improper recognition of revenue
Grace Improper use of reserves to facilitate income smoothing
McKesson Premature recognition of sales revenue
Rite Aid Recognition of fictitious vender credits
Sunbeam Fictitious sales and improper use of reserves
Waste Management Improper revenue recognition
Table 1    


We pair-matched each fraud firm with a non-fraud firm classified within the same SIC code. If data availability permitted the identification of multiple non-fraud firms, we randomly selected two such pair-match firms (if available). The fraud firms and the pair-matched non-fraud firms are listed in Table 2.  


  Fraud Firms  Non-fraud Firms
Pair-Match 1
Non-fraud Firms
Pair-Match 2
1 Cendant Advance Tobacco Products Competitive Technologies
2 Con Agra Sara Lee Classica
3 Enron Mercury Air Group World Fuel Service
4 Grace Great Lakes Chemical None
5 McKesson Bergen Brunswig None
6 Rite Aid Drug Emporium None
7 Sunbeam Decorator Industries None
8 Waste Management Rich Coast Wastemasters
Table 2      


We pbtained financial statements from the COMPU-STAT Annual Industrial File. Since this is an exploratory study, we downloaded data for all available data items. However, examination of the data on a firm-by-firm basis revealed that certain data item fields frequently were missing. We eliminated these fields because their data couldn't be compared across firms. The 45 remaining items are defined in Table 3.  

For full access to story, members may sign in here.

Not a member? Click here to Join Now and access the full article.