Section 6751(b)(1) prohibits the IRS from assessing any penalty against a taxpayer “unless the initial determination of such assessment is personally approved (in writing) by the immediate supervisor of the individual making such determination...” The Tax Court will not sustain a penalty unless the IRS produces evidence that the required personal approval has taken place before the IRS first notifies the taxpayer (typically in either a 30-day letter or the NOD) about the penalty. Section 6751(b)(2) provides an exception to the personal approval requirement for “any...penalty automatically calculated through electronic means.”
Craig S. Walquist and Maria L. Walquist v. Commissioner, 152 T.C. No. 3 (Feb. 25, 2019) (Judge Lauber) was one of two reviewed opinions issued last week that gave the IRS important wins on the scope of §6751. In Walquist the Automated Correspondence Exam (ACE) system hit the taxpayers with a §6662 substantial understatement penalty. No IRS employee even knew about it until after the taxpayers petitioned Tax Court in response to the automated NOD. Thus, there was no supervisory approval as required by §6751(b)(1). The Court decided, however, that the IRS was entitled to the §6751(b)(2) automatic computation exception to the supervisory approval requirement.
At one level, this was an easy case against two unsympathetic taxpayer hobbyists. At another level, however, the decision may create problems down the road because the facts of the case are more modest than the scope of the Court’s language. That tension between facts and language may end up harming other taxpayers ensnared by the IRS automated processes. As usual, you will find the more complete story below the fold.
The Automation Context
Today's lesson requires a basic understanding about the role of automation in tax administration because the administrative context of penalties has been important to how the Tax Court applies §6751.
The Tax Code is studded with sections giving the Secretary of the Treasury, a single human being, broad powers to administer the tax laws. Section 6201 gives the Secretary the power to assess. Section 7301 gives the Secretary the power to collect. And §7601 gives the power to “inquire.” It says that the Secretary shall “direct officers or employees of the Treasury Department to proceed, from time to time, through each internal revenue district and inquire after and concerning all persons therein who may be liable to pay any internal revenue tax....”
If the language now in §7601 seems quaint, that is because it is. It is pretty much the same language Congress first used over 150 years ago in the Revenue Act of 1862. That Act is the foundation of the modern Internal Revenue Code and created the tax administration agency we now know as the IRS. There, §7 directed “...the several assistant assessors to proceed through every part of their respective districts; and inquire after and concerning all persons...liable to pay any...tax.....” Quaint.
While these words have remained largely unchanged for 150 years, the administrative context in which they operate has undergone a radical shift. Even when the words were written, no one thought “the Secretary” would do all the work. The work was instead delegated to subordinate officers and employees. It still is. You can find the current web of delegation orders in the IRM Part 1.11.4.
All those delegation orders, however, are from one human to another human. And until the mid-1940's humans did all the work. Starting in the 1940’s, however, the methods the IRS used to “inquire” started to change. Over time they have changed radically from human inquiry to computer inquiry. For those interested, I give the historical details in “Theory and Practice in Tax Administration," 28 Va. Tax Rev. 227 (2009).
The most salient point of the transformation from human inquiry to computer inquiry is a shift in decision-making from post-informational individualized evaluation to pre-informational bulk evaluation. That is, under the human model, an individual IRS employee makes a judgment about an individual taxpayer based on information the IRS employee has already seen and evaluated. Did the taxpayer fail to properly report all payments received as income? Were deductions proper? After evaluating that information, the IRS employee might propose a penalty.
Under the computer model, however, judgments are pre-programmed by IRS administrators who instruct programmers how to write the computer code. Rather than arriving at decisions based on information, computer administration makes judgments based on operating presumptions applied to a lacuna of information. Once the assumptions lock in, taxpayers must later try to undo the consequences.
The example we find in today’s lesson is the Automated Correspondence System (ACE). This computerized inquiry is triggered by a mismatch between a taxpayer’s Form 1040 and information returns. The operational presumptions built into the computer programming are that (1) information returns are accurate and (2) taxpayers who do not respond to the automated requests for information are noncompliant. The two operating assumptions then result in a default proposed assessment. The National Taxpayer Advocate (NTA) explained it well in her 2008 Annual Report to Congress on pages 248-249:
“Once the IRS engages the batch system, cases move through the examination process automatically. Each step in the process has a pre-established period programmed into the system. Files are not created or examiners assigned to the cases until the IRS receives and controls a taxpayer’s correspondence. If a taxpayer fails to furnish the requests documentation precisely within the prescribed period, the case automatically moves to the next phase in the process. ... Because the batch system automatically processes a case from its creation through the issuance of a Statutory Notice of Deficiency and subsequent closing, the IRS has effectively eliminated the need for human involvement in every case in which a taxpayer does not reply in a timely fashion.”
Again, the operational presumption at work here is that all taxpayers who do not respond timely are noncompliant. Worse, even timely taxpayer responses do not work if the IRS employee cannot enter the response data in the computer system in a timely way. If the data is not timely entered, the system “purges” the case and immediately sends out the next document in the series, such as the NOD. The new mail-out then triggers a new response deadline and a new place for response data to be entered. That is, the computer design requires a second response. No human examiner is assigned to the case unless and until the taxpayer responds in the right way within the right time frame for the computer to recognize that the taxpayer has, indeed, responded to the particular automatic mail-out! Thus, “the automated process limits the ability of taxpayers to engage in a meaningful dialogue with tax examiners, to ask questions about the process and the issues, and to resolve problems that invariably arise during the course of an examination.” NTA 2008 Report to Congress at 249.
As part of this automated process, the ACE programming computes and adds on penalties whenever the criteria for §6662 is met (generally where the program has calculated an understatement of tax greater than $5,000 or 10% of the tax assumed, on the basis of information returns, to have been the right tax to report). Taxpayers may thus very well receive an NOD that proposes penalties without any IRS employee even knowing that until the taxpayer files a petition in Tax Court and the case gets routed to Appeals.
Properly conceived, the ACE result is still human judgment just as it would be if a taxpayer’s return is matched by the eyes of an employee who then reviews the return and decides whether to add a penalty. The difference is that the human judgment in ACE is built into the program’s code and is not an individualized judgment about a particular taxpayer but is a judgment about a group of taxpayers: those who have a mismatch and don’t respond.
Section 6751 and Graev
When courts interpret a statute they start with text, then often move to statutory context. Not so for §6751. Judge Lauber explained in his concurrence in Graev v. Commissioner, 149 T.C. No. 15 (December 2017), that the statutory phrase “initial determination of such assessment” is an oxymoron. That means that text of the statute is useless to its interpretation and the Court must look to context. And the context that has been important to the Tax Court in interpreting §6751 is the statute's perceived role in tax administration. I call that administrative context. That is, the Court has looked to see where in the tax administration process it makes the most sense to give force and effect to §6751's requirements.
The Tax Court has changed how it interprets §6751. Before Graev, the Tax Court looked at the term "initial determination of such assessment" and focused on how the word assessment" fit into the tax administration context. Tax Court proceedings are pre-assessment. The pre-Graev interpretation was that taxpayer complaints about §6751 in a Tax Court case were premature because the assessment had not happened yet. Judge Holmes gives a lucid explanation of the former reasoning in his Graev concurrence.
The Graev Court, however, focused on the words “initial determination” and how those words fit into the tax administration context. After all, penalties are proposed to be assessed as part of the audit process and are included in the NODs that the Tax Court is charged to redetermine. Judge Lauber's concurrence in Graev explains how the Tax Court majority was focusing on this aspect of tax administration because of language in the legislative history of §6751. He wrote that “Congress was concerned about the bigger picture: It desired to prevent rogue IRS personnel from using penalties as leverage to extract concessions from taxpayers.” Once one focuses on that part of the administrative context, the phrase “initial determination” becomes important and allows taxpayers to force the IRS' to prove it complied with §6751 before it sent out the NOD and not just some later time before the formal assessment.
One cannot say that the Graev Court’s interpretation is wrong. Nor can one say it is right. What one can say is that it's shift in focus is unfortunate, for many of the reasons Judge Holmes presciently presented in his concurrence. In particular, it buys into the assumption underlying the statute that tax administration consists of IRS employees interacting with taxpayers. The personal review requirement assumes that human employees are involved in determining penalties. It imposes the requirement to prevent bad behavior by people. What happens when a taxpayer is unable to break out of the computer interaction? Is there such a concept as rogue computers?
Some of you may be screaming “that's what the automatic computation exception is for!?!?” Hang on. I’ll get there when I get to the Court’s holding. But first let’s see the facts, because they are important to the lesson.
The Facts of the Case
Mr. and Ms. Walquist got snagged by the computers. They had filed a return for 2014, reporting wages and other gross income items totaling $94,000 but they left off $1,215 of unemployment compensation received from the state of Minnesota. Meanwhile, the they took a imaginary deduction of over $87,000, giving it the give-away label of “Remand for Lawful Money Reduction.” Uh-oh.
The IRS has a hard time catching erroneous deductions. If the only problem with the Walquists’ return was a single defective deduction entry, it would likely take a human to catch it. Sure, it might get kicked out by a large Discriminant Inventory Function (DIF) score, but that would still take a human reviewer to spot the deduction as the reason to send the return to Exam. See IRM 184.108.40.206 for details. Or you can read this EA's blog post on DIF scores.
But the IRS is GREAT at catching unreported income, like the $1,215 unemployment compensation that the Walquists failed to report. Minnesota sent the IRS a 1099-MISC. When the IRS computer systems “saw” that mismatch it kicked the return into ACE. And once in that system, the computer ignored the weird deduction and dinged the Walquists for the entire income line and calculated an understatement of tax of almost $14,000.
I confess I do not know why ACE ignored the deduction in proposing the deficiency. I welcome anyone to comment on that. I suspect that the ACE programming is designed to disallow all deductions except the standard deduction. If that is correct, however, it again shows how pre-informational decision-making is built into computer programming. It is still a human judgment being made here, albeit one exercised by instructions to a computer programmer to create an algorithm and not one made in reaction to actual information. But I don’t really know.
I do know, however, that no human touched the return and the calculated understatement of tax was so large that it triggered a §6662 substantial understatement penalty of almost $2,800 that appeared on both the 30-day and 90-day letters. ACE was programmed to add that penalty. There was no evidence that the Walquists attempted to respond to the IRS directly, and thus, as Judge Lauber points out, “the penalty determined in the notice of deficiency was not reviewed before issuance of that notice by any human IRS examiner.”
The Walquists are not sympathetic taxpayers! They filed a Tax Court petition filled with long-winded and long-rejected tax protestor arguments, thus proving once again that P.T. Barnum knew what he was talking about. By the end of the case, the Tax Court ended up walloping them with an additional $12,500 §6673 penalty. That’s pretty astonishing, considering how careful, cautious, and considerate the Tax Court generally is about imposing §6673 penalties on taxpayers.
Even the most unsympathetic taxpayers, however, deserve to have the IRS comply with the law in general and with §6751 in particular. And so the Tax Court was obliged to look at whether the IRS violated §6751 when its computer program automatically added the §6662 substantial understatement penalty to proposed deficiency and then issued the NODs without any human IRS employee signing off on the proposed imposition of the penalty.
Here's where I discuss the automatic calculation exception. Section 6751 contains two exceptions to the personal approval requirement. The first comes in §6751(b)(2)(A), which says that the approval of a supervisor is not required for “any addition to tax under sections 6651, 6654, or 6655.” The second exception comes in §6751(b)(2)(B), which says that the approval of a supervisor is not required for “any other penalty automatically calculated through electronic means.”
The Tax Court held that the second exception applied in this case. It used this sweeping language: “Because the penalty was determined mathematically by a computer software program without the involvement of a human IRS examiner, we conclude that the penalty was “automatically calculated through electronic means.” Thus, the NOD was not subject to the personal approval requirement.
One finds similarly sweeping language in the Court’s summary of the case: “Penalties determined under I.R.C. sec. 6662(a) and (b)(2) by an IRS computer program without human review are “automatically calculated through electronic means” within the meaning of I.R.C. sec. 6751(b)(2)(B) and thus are exempt from the written supervisory approval requirement of I.R.C. sec. 6751(b)(1).”
The true lesson in this case may be more modest than that the sweeping language suggests. Two aspects of the case should give readers pause before they take the sweeping language literally to apply to all computer-generated penalties. First, the Tax Court relies upon a weak analogy between the §6662 penalties and the sections 6651, 6654, and 6655 additions to tax. Second, the taxpayers in this case did not successfully respond to the computer notices. Therefore it is difficult to say what the result might be for taxpayers who are able to successfully respond to computer notices and thus trigger the intervention of a human IRS employee.
1. Weak Analogy. The Tax Court properly gave weight to the word “other” in the statute. That is, the automatic computation exception applies only to penalties that are analogous to the listed additions to tax, the ones in §6651, §6654, or §6655 . The Tax Court noted that each of the additions to tax were statutorily mandated, and that taxpayers could escape the additions only by showing reasonable cause. The Tax Court thought that structure of the §6662 penalty was similar, at least when computer-determined. The computer automatically imposes the penalty and the taxpayer has to respond to avoid it.
The weakness here, of course, is the Court’s focus on the language imposing the addition or the penalty. All penalties have mandatory language! The better focus is to look at what makes the listed additions to tax “automatically calculated.” That is, in what sense are those additions to tax automatic? What triggers them?
When one uses the better focus one sees that the additions to tax are all triggered by easily determined events and are “automatically calculated” in the same way interest is calculated, by the mere passage of time. For example §6651 requires in addition to tax for the failure to file and for the failure to pay and “an additional [amount] for each additional month or fraction thereof during which such failure continues.”
In contrast, the §6662 substantial understatement penalty is triggered only after someone has made a discretionary determination “of the tax required to be shown on the return for the taxable year.” §6662(d)(1)(a)(i). That discretionary determination is what Congress was worried about being improper, not the resulting mechanical calculation of tax. The penalty calculation is as mechanical when done by computer as when done by a human. Both follow the statutory formula but that formula is predicated on a determination of the proper "tax required to be shown" that was not, actually, shown.
he Court goes on to say that if the §6662 substantial understatement penalty were not considered a “penalty automatically calculated through electronic means, it is difficult to conceive what type of penalty would qualify for the statutory exception.” Well, heck, what about §6677? That’s another failure to file penalty, arguably triggered by an event more like the §6651 addition to tax than like the substantial understatement penalty. It even has analogous passage-of-time language to §6651 by imposing an additional “$10,000 for each 30-day period (or fraction thereof) during which such failure continues....”
2. The Facts Limit the Holding. One must be careful not to extend the holding beyond the actual facts of the case. Here, the taxpayers did not respond to the ACE 30-day notice. I would be hesitant to say that the substantial understatement penalty is excepted from the personal approval requirement when an actual human IRS employee puts eyeballs on the taxpayer’s return. Even the IRS recognizes this in IRM 220.127.116.11.2(5). There, the IRS instructs employees that a supervisor must affirmatively approve the ACE penalties before the NOD goes out when a taxpayer successfully gets someone’s attention.
I doubt the Tax Court’s broad language makes that the IRM provision unnecessary. On the one hand, the temptation to read it that way is strong. The Congressional concerns about individual employees trying to browbeat taxpayers by reckless decision-making during audit are not obviously present when automated systems treat every taxpayer the same. The Tax Court pointed this out: "where, as here, a penalty is determined by a computer software program and never reviewed by a human being, it could hardly be considered a “bargaining chip.”
On the other hand, the concerns about individualized abuse may translate to computerized abuse. Congress wrote §6751 as an anti-abuse statute predicated on an outdated view of individualized IRS-taxpayer interaction. The same concerns may indeed translate to automation. A cynic might put it this way: in the old days individual IRS employees put the screws to individual taxpayers; now ex-ante programming decisions put the screws to entire classes of taxpayers. The notion that the computers treat all taxpayers the same is false. Taxpayers who respond are treated very differently than taxpayers who are so unfortunate as to not be able to respond in the precise way in the tight time periods they are given.
And here may be the nut of it: the Walquists were not unfortunate taxpayers; they were unsavory taxpayers. That fact is yet another reason to treat this case with more caution than the sweeping language might suggest. Just as hard cases sometimes make bad law, we must be careful not to let easy cases do the same.
Coda: In the other §6751 case issued last week, the taxpayers made a bunch of poultry-produced fecal-matter arguments. It's another part of the ongoing efforts by the Tax Court to give §6751 a fair and balanced interpretation, given the unfortunate decision in Graev. I will blog the other case next week, unless something else provides a better Lesson From The Tax Court.
Bryan Camp is the George H. Mahon Professor of Law at Texas Tech University School of Law