Christine here, posting incognito.
Special Trial Judge Diana Leyden issued a nondesignated order in April 2018 that signals interesting reading and litigation ahead on another Graev-related issue: when is a penalty “automatically calculated through electronic means” so that supervisory approval is not required? Thanks to Carl Smith for alerting us to this order in Triggs v. Comm’r. This is not a new legal issue, but it is newly salient in deficiency litigation since the Tax Court’s December 2017 ruling in Graev III (first blogged on PT here).
read more...When a tax return omits income or overstates credits, an accuracy-related penalty might be imposed under section 6662(b). At issue in Triggs are the negligence penalty in section 6662(b)(1) and the understatement penalty in section 6662(b)(2). There are several other grounds for accuracy penalties in section 6662, which are not discussed here.
The Graev and Chai litigation concerned section 6751(b)(1), which generally requires written supervisory approval before the IRS assesses a penalty. However, section 6751(b)(2) sets out two exceptions to that requirement.
Paragraph (1) shall not apply to—
(A) any addition to tax under section 6651, 6654, or 6655; or
(B) any other penalty automatically calculated through electronic means.
Section 6651 contains the failure to file and failure to pay penalties. Sections 6654 and 6655 establish estimated tax penalties. The application of these penalties is mechanical and automatic, as many of us have explained repeatedly to indignant clients. Taxpayers can contest the 6651 penalties (and sometimes the 6654 penalty) by showing that their failure to comply was due to reasonable cause and not willful neglect. However, this burden is on the taxpayer.
That brings us to section 6751(b)(2)(B). What else counts as a “penalty automatically calculated through electronic means”? Which of the accuracy-related penalties might be considered sufficiently automatic, and under which circumstances? Can a determination of negligence ever be “automatically calculated”? Les pointed me to the authorities discussed in Saltzman & Book 7B.24, at n 889.1.
Since at least 2002, the IRS has taken the position that supervisor approval is not required if a penalty is assessed entirely by an automated computer program. SCA 200211040 addresses Automated Underreporter (AUR) assessments of the negligence and substantial understatement penalties. It acknowledges a lack of legislative history to guide the analysis. The memo concludes that
assessment of a penalty qualifies as one calculated through electronic means if the penalty is assessed free of any independent determination by a Service employee as to whether the penalty should be imposed against a taxpayer.
The key fact according to the memo is that the taxpayer must respond to a proposed AUR assessment before any IRS employee looks at the matter and exercises judgment. This may seem surprising as to the negligence penalty, since to find negligence there must be a determination of whether a taxpayer made a reasonable attempt to comply with law and whether they exercised ordinary and reasonable care in doing so. See Treas. Reg. § 1.6662-3(b)(1). The SCA memo reasons, however, that
programmed into the computer are uniform factual criteria under which the computer will automatically propose a negligence penalty; when a taxpayer, for a second year, fails to report income reported on third party information returns, the programmed determination is that the taxpayer has not exercised ordinary and reasonable care in the preparation of his return. When the computer program automatically assesses the penalty on the basis of this mechanical determination, we believe that it qualifies as an exception to the general rule requiring written supervisory approval.
I find this rationale troubling. Hopefully IRS personnel also aim to use uniform factual criteria when deciding whether a negligence penalty is warranted. Very complex decisions can be made by computers if they are programmed to do so, but programming incorporates the biases and errors of human programmers. A computer does not necessarily arrive at a legally correct determination. There is a wealth of debate on this issue in other contexts, including criminal sentencing. As journalist Matthias Spielkamp puts it, “algorithms make no value judgments – except the ones designed by humans.” The AUR negligence logic tree may not be as complex as the algorithms that send people to prison or deny them credit, but the difference seems to me one of degree. As tax compliance has become more automated, and as algorithms have become more sophisticated, it is hard to identify any penalty decision that could not in theory be made by a computer program.
Troubling or not, the IRS applied the reasoning of SCA 200211040 in 2014 to the frivolous tax filing penalty under section 6702, in CC 2014-004. As in the AUR function, the IRS had programmed its computer system to apply predetermined criteria for assessing the frivolous return penalty, and no human employee reviewed an individual assessment unless the taxpayer responded.
The Triggs case involves the negligence and understatement penalties, asserted in the alternative. The IRS conducted a correspondence examination of Mr. Triggs’s return, but he did not respond until after the notice of deficiency was issued (with the proposed penalty). Like many of the post-Graev cases we have blogged, the trial in Mr. Triggs’s case occurred before the Graev III decision. In March 2018, Judge Leyden invited the IRS to file a motion to reopen the record to submit proof that it had complied with the penalty assessment procedures. The IRS responded by asserting that those procedures do not apply to Mr. Triggs’s case, because his penalty was proposed by the correspondence examination computer system without any human review and therefore falls under section 6751(b)(2)(B). This position is consistent with the reasoning in SCA 200211040 and CC 2014-004, although those memos did not involve correspondence examinations.
However, Judge Leyden points out several IRM sections that seem inconsistent with this position. For example, the order notes:
With respect to correspondence examinations, one provision of the IRM states: “The determination to assert penalties, to identify the appropriate penalties, and to calculate the penalty amount accurately is primarily the examiner’s responsibility. This responsibility remains the same even when examinations are conducted by correspondence.” IRM pt. 4.10.3.16.8 (Mar. 1, 2003).
Judge Leyden therefore orders the IRS to file a supplement by May 31, setting out the legal basis for its position in more detail, and specifically addressing the IRM provisions identified in the order. The supplement will be available at the Tax Court’s offices in Washington, D.C., if any enterprising reader is motivated to obtain a copy.
In 2014, in response to the IRS position on AUR assessments, the National Taxpayer Advocate issued a legislative recommendation that Congress amend section 6751(b)(2)(B) to specify which penalties are not subject to the general requirement, and to specifically exclude the negligence penalty. She has maintained that recommendation for several years, and it is now included in the Purple Book compilation. That recommendation will be even more important if the government maintains its current position in Triggs.
This raises a question that has been vexing me recently, mostly because I think there should be an obvious answer that I just don’t have the skill to find.
Suppose there are two taxpayers, both in a 25% tax bracket. One of them withdraws $18,000 from a retirement account, with no tax withheld. The other withdraws $22,000 from a retirement account, with 20% tax withheld. Neither taxpayer reports the 1099-R income on Form 1040. IRS issues a CP-2000 to both.
The first taxpayer owes $4,500 tax, but no “substantial understatement” penalty because the tax is less than $5,000. The second taxpayer owes $1,100 tax after credit for the withholding, but the computer will also assess a $220 “substantial understatement” penalty because the tax already withheld is ignored.
I don’t think this is what Congress intended, but the language of Section 6662 is foggy enough that a Vulcan programmer in the AUR department could see it that way.
CC 2014-004 says:
“The models use objective data, such as the amount of income tax and withholding reported on a return and the type of Forms used to make certain claims, to determine whether the imposition of a section 6702 penalty is warranted.”
When an IRS data entry clerk embezzles withholding that was reported on a type of Forms used to make certain claims, and objective data (IRS records) show that the legitimate taxpayer didn’t have the withholding they claimed[*], human involvement is no longer necessary to determine that imposition of a section 6702 penalty is warranted.
When a stockbroker reports on Form 1099-B that they withheld 30% of gross proceeds of sales of shares, but the capital gain (or loss) was less than the amount of withholding, human involvement is no longer necessary to determine that the declaration of withholding larger than the amount of income is obviously false and therefore frivolous.
It must have been quite a burden on the IRS to make multiple employees examine the altered records, refuse to contact payers, approve imposition of penalties, and refuse to inform the taxpayer of what the taxpayer’s frivolous position was. What a relief to see that these operations could be automated.
But why this:
“The pre-notification letter advises the taxpayer that an assessment of the penalty is being proposed, and gives the taxpayer a fixed period of time to contest the assessment. If no response is received within the fixed period after the pre-notification letter is sent, the penalty is systematically assessed. If a response is received, then a Service employee from the Frivolous Return Program works the case.”
If a response is received, why continue to burden an IRS employee with responsibility to ignore the response and proceed to, as the case may be, assess the penalty or forget to assess but partially collect anyway? This can be automated too, you know?
[* The IRS doesn’t issue Notice CP-2000 in such a case. That would run the risk of a taxpayer figuring out what’s happening.]