Sunday, July 10, 2011

Reverse Engineering The IRS DIF-Score

What would you do if you were a statistician audited by the IRS? Dr. Amir Aczel, Professor of Statistics at Bentley college and author of the 1995 book How To Beat The IRS At Its Own Game chose to reverse engineer the IRS' secret audit selection formula (the DIF score) using quantitative methods and publish it to the world.

The Secret Formula: How The IRS Picks Audits

The IRS uses a quantitative method called Discriminant Analysis to identify the 'underreporters' from the normal returns, driven largely by the following details.
  • Schedule A Ratio: They'll audit you if your schedule A (Itemized) deductions are more than 44% of your income.
  • Schedule C Ratio: They'll audit you if your ratio of schedule C (Business) deductions is more than 63% of income.
  • Schedule F Ratio: They'll audit you if your ratio of schedule F (Farm) deductions is more than 67% of income.
  • Audits are 4x more likely if you own a business and 2x if you own a farm.
  • The Obama administration has focused on high earners: 
  • Occupation affects your audit likelihood 
    • 22% of business and personal services companies are audited every year.
    • 16% of building contractors are audited every year.
  • Returns filed after April 15 aren't audited. 

Debunking Myths

 Dr. Aczel refuted a number of IRS myths in addition to reverse engineering the IRS audit formula.
  • Debunked IRS Omniscience: The IRS previously held that the formula was complex and included countless variables known about your life.
  • Debunked Annual DIF Score Updates: The IRS formerly stated that they modify the selection algorithm yearly but Aczel could find no detectable changes. 
  • Highlighted Politically Motivations Rather Than Economic: At the time of writing, Congress was considering budget cuts to the IRS. The Commissioner tried to prove the agency's value by tripling the number of audits, albeit by focusing on lower payout cases.

Game Theoretic Interpretation of the Audit Process

  • Entrapment: Before reviewing the case, the Auditor may offer to settle for half of your claimed deductions. They are trained to make this offer for two reasons:
    • You may settle solely to avoid the hassle, but the IRS will interpret this as an implicit admission of guilt. Accordingly, they will audit you repeatedly throughout your life.
    • Following the case to completion will halve your tax obligation (at least). So the auditor doesn't sacrifice anything by making this offer.
  • Auditors Don't Want Cases With Taxpayers That Appeal: IRS agents select cases from a pool chosen by the DIF Score. "The auditor's performance is judged by the IRS based on the number of cases closed with the taxpayer's agreement." If you settle quickly and pay a fine, other agents will gladly audit you in the future. Your best deterrent is to fight for every deduction, let the process fully run its course, escalate it to a supervisor and use the appeals process because it makes you an unattractive target for future years.
  • Always Appeal. (85% Settled With An Average 40% Reduction In Tax) Always appeal the Senior Auditor's decision. The appeals officer is 'judged by how many cases he or she settles with the taxper, rather than on backing up the original auditor's report.'
  • Post-Appeal. (Another 85% Are Settled Pre-Trial) Even after the appeals process, 85% of the post-appeals cases are closed before they get to trial. Why? The IRS' 'legal resources are scarce'.


I love how the learnings completely reverse the negotiation. Firstly, taxpayers are empowered to confront the auditor with justifications for the item which flagged them. Secondly, this information empowers taxpayers to avoid audits by delaying their tax filing until August, rebalancing deductions to different Schedules or by providing thorough explanations for items that will be flagged by the IRS. Lastly, QM removes the myth of omniscience and proves that the audit payoff is counter-intuitive, because you save more money the longer you continue the process (despite any comments from the auditor). It is probably the only example in government relations where you should ignore their recommendations (they're not mandates) and just consistently escalate the issue.

These findings are also disconcerting because they point to an ineffective detection mechanism for tax evasion. It means that someone with no legitimate deductions but who claims 20% won't be audited, while someone with a legitimately high number of claims will be audited. Looking at this from the systems perspective, the IRS is probably chasing the same group of legitimate high-deducters year after year while completely missing the filers with smaller illegitimate claims. One such example was mentioned:
"During one of my radio interviews, on WJJD in Chicago, a lady called in and described on the air how every single year she is audited by the IRS. She and her husband, apparently, make a modest living, and don't file any 'complicated' tax forms. They don't own a business or a farm; their charitable contributions are moderate; they don't have a home office. But they have 21 children. She described how every year her husband has to go downtown to meet with the IRS after receiving the letter asking them to come for an examination."
Do they really need to validate this every year? Wouldn't they eventually realize that the dependents are all verified?

Lastly, am I the only one shocked at the level of subjectivity in these proceedings? "Every year, Money magazine asks a group of 50 top tax experts to prepare the return of a hypothetical individual, and they then publish the comparison of the experts' results. In all the years this has been don, not a single time have even two experts agreed on the exact amount of tax due!"

How to Beat The I.R.S. At Its Own Game. By Amir Aczel, Ph.D. Four Walls Eight Windows Publishing, 1995. New York, NY.

"Predictors Of Unreported Income: Tests of Unreported Income (UI) DIF Scores." by Dennis Cyr, Thomas Eckhardt, Lou Ann Sandoval, and Marvin Halldorson. Internal Revenue Service Research Conference.
June 11, 2002.