Our objective was to determine whether digital analysis could be effectively used by the Internal Revenue Service (IRS) to identify potentially erroneous and/or fraudulent fuel tax claims. Digital analysis is a technology designed to find anomalies in corporate data. It is based on a mathematical theory known as Benford’s Law that established the digital patterns normally found in authentic and unmanipulated data. Digital analysis has been successfully used in conjunction with Benford’s Law to identify financial fraud in both the government and private sectors.

To test the feasibility of using digital analysis on fuel tax claims, we obtained the assistance of the IRS’ Office of Excise Taxes in Washington, DC, the IRS’ District Office Research and Analysis (DORA) Division in Cincinnati, Ohio, and the IRS’ Excise Research function in Covington, Kentucky. A DORA statistician applied Digital Analysis Tests and Statistics (DATAS) software to the IRS’ national database of approximately 178,000 fuel tax claims filed in 1997 and 1998. The statistician’s analysis determined that the refund amount of the claims filed by about one-third of the taxpayers who had filed 20 or more claims each year did not conform to the predicted digital patterns. This indicated that digital analysis could be used with Benford’s Law as a potentially effective tool for identifying fraudulent claims and/or productive examination cases.

The statistician selected a sample of 204 taxpayers for examination by IRS excise tax agents. Each of the taxpayers had filed a minimum of 20 claims. The sample included taxpayers whose claims "passed" Benford’s Law (i.e., conformed to the expected digital patterns) and taxpayers whose claims "failed" Benford’s Law. The sample included taxpayers from 29 of the 33 IRS district offices that existed prior to the IRS’ reorganization in October 2000. Details of the sampling plan are presented in Appendix I.

As of December 13, 2000, the IRS’ excise tax agents in these districts had completed reviews of the claims filed by 191 of the sampled taxpayers. The scope of our work was limited to an evaluation of these examination results. We did not assess the technical quality of the reviews.

This test of digital analysis was originally initiated as part of our 1999 audit of the processing of fuel tax credits. The final report for that audit was issued on March 30, 2000. We are reporting the digital analysis test results separately due to the need to wait until the IRS substantially completed its reviews of the sampled fuel tax claims. This review was performed in accordance with the President’s Council on Integrity and Efficiency’s Quality Standards for Inspections.

Major contributors to this report are listed in Appendix II. Appendix III contains the Report Distribution List.

The federal government levies excise taxes on the sale, use, and inventory of various types of goods or services, including fuels such as gasoline and diesel. Excise taxes may be refunded when taxed fuel is subsequently used for tax-exempt purposes or when the fuel is taxed a second time.

Taxpayers who paid excise taxes on tax-exempt or double-taxed fuel may request a refund from the IRS by filing a Claim for Refund of Excise Taxes (Form 8849). These taxpayers are frequently wholesale distributors of gasoline and registered ultimate vendors of diesel fuel who buy the fuel at a price that includes the excise tax and subsequently sell it at a tax-excluded price to state and local government entities, non-profit educational organizations, or farms for farming purposes. Taxpayers who meet certain dollar requirements may file multiple claims throughout the year rather than waiting until the end of the tax year. The refund claim amount is calculated by multiplying the number of gallons of fuel purchased by the applicable fuel tax rate. These rates vary for each type of fuel.

During Calendar Year (CY) 1998, the IRS processed almost 90,000 claims and refunded approximately $500 million in fuel taxes. Although all Forms 8849 are reviewed prior to payment by revenue agents or tax auditors in the Centralized Excise Program Section at the Cincinnati IRS Center, the IRS has no specific data-driven programs for selecting claims for examination by excise tax agents in the field.

Fraud control is an area of high importance and concern in the IRS’ efforts to promote voluntary compliance. The General Accounting Office (GAO) has reported that, while the IRS has detected several scams involving refunds of gasoline or diesel fuel excise taxes, it does not know how extensive this fraud may be. The GAO stated that the incentives to obtain fraudulent excise tax refunds are great because of the rising fuel tax rates.

Digital analysis is apparently not an effective method for identifying potentially erroneous and/or fraudulent fuel tax claims. This conclusion is based on the results of the IRS’ reviews of claims that were selected by using digital analysis techniques. The completed reviews resulted in the assessment of additional taxes against only 8 percent of the sampled taxpayers. The taxpayers whose claims did not conform to Benford’s Law did not have a significantly higher change rate or adjustment amount than the taxpayers whose claims conformed to Benford’s Law.

Through December 13, 2000, IRS excise tax agents had completed reviews of the fuel tax claims filed by 191 taxpayers selected for examination using digital analysis techniques. The completed reviews found that the claims filed by 92 percent of the taxpayers were correct as filed.

Additional taxes totaling $246,833 were assessed against the remaining 8 percent of the sampled taxpayers. The completed reviews found no evidence of fraud. Further, neither the change rate nor the adjustment amount was significantly higher for those taxpayers whose claims did not conform to the digital patterns expected by Benford’s Law.

Analysts and statisticians from the Excise Research and DORA functions provided the following potential reasons that digital analysis was not an effective approach for identifying erroneous fuel tax claims:

• Most claimants were registered ultimate vendors and are subject to IRS reviews every 2 years.
• Many claimants operate in a controlled environment. Fuel purchases are generally delivered by tanker load or by tank wagons. Therefore, the quantity of fuel ordered and delivered would generally involve a consistent number of gallons. Thus, there would be a high probability that the first, second, and, possibly, the third digits of the claim amounts would be repetitive and, therefore, would not follow the expected digital patterns according to Benford’s Law.
• Claimants tend to file periodically. Having a reasonably constant volume of fuel sales would lead to smaller deviations in the refund amounts of the claims than would be expected by Benford’s Law.
• Claimants might have some "business" reason for filing a claim when the total amount reaches a certain dollar level. Therefore, these claims would not follow Benford’s Law.

Digital analysis is apparently not an effective tool for identifying inaccurate fuel tax claims. The IRS’ reviews of sampled claims produced only an 8 percent change rate and found no fraud. Further, the no change rate and the adjustment amounts were not significantly higher among taxpayers whose claims "failed" Benford’s Law than among those taxpayers whose claims "passed" Benford’s Law.

The Internal Revenue Service’s (IRS) District Office Research and Analysis (DORA) Division, Ohio District, developed the sampling plan presented in the following table to test whether digital analysis could be effectively used to identify erroneous fuel tax claims. The sample identified 204 taxpayers for examination by IRS excise tax agents and was designed to ascertain the difference between categories of claimants.

The Excise Research function at the Cincinnati IRS Center assimilated audit packages that contained the original claims, an audit checksheet and procedural guide, and other information for each taxpayer. These packages were forwarded to field excise tax groups for examination. The completed examinations were returned to the Excise Research function for review and tabulation.

Gordon C. Milbourn III, Associate Inspector General for Audit (Small Business and Corporate Programs)

Philip Shropshire, Director

William E. Stewart, Audit Manager

Michael R. Van Nevel, Senior Auditor

Commissioner N:C

Commissioner, Large and Mid-Size Business Division LM

Deputy Commissioner, Large and Mid-Size Business Division LM

Deputy Commissioner, Small Business/Self-Employed Division S

Acting Director, Internal/External Stakeholders, Small Business/Self-Employed Division,

Compliance Policy S:C:CP:I

Director, Compliance, Small Business/Self-Employed Division S:C

Director, Communications, Small Business/Self-Employed Division S:COM

Chief Counsel CC

Director, Legislative Affairs CL:LA

Director, Office of Program Evaluation and Risk Analysis N:ADC:R:O

National Taxpayer Advocate TA

Office of Management Controls M:CFO:F:M

Audit Liaisons:

Commissioner, Large and Mid-Size Business Division LM

Commissioner, Small Business/Self-Employed Division S

This appendix presents detailed information on the measurable impact that this evaluation of digital analysis for selecting fuel tax claims for examination had on tax administration. These benefits will be incorporated into our Semiannual Report to the Congress.

Increased revenue/revenue protection – Actual; the Internal Revenue Service’s (IRS) reviews of the fuel tax claims filed by taxpayers selected for inclusion in the test of digital analysis resulted in the assessment of additional taxes totaling $246,833; (see page 4)

The information on the amount of additional taxes assessed was provided by the IRS’ Excise Research function in Covington, Kentucky, which was responsible for tabulating the examination test results.

The response was removed due to its size. To see the response, please go to the Adobe PDF version of the report on the TIGTA Public Web Page.