Tax policy colloquium week 3: Natasha Sarin's "Understanding the Revenue Potential of Tax Compliance Investments"

Earlier today at the Tax Policy Colloquium, Natasha Sarin presented her paper (co-authored by Lawrence Summers), Understanding the Revenue Potential of Tax Compliance Investments.

The paper argues that restoring the IRS budget to a 2011-equivalent level, by increasing its budget by $107 billion over the next ten years (mainly for auditing and technological improvements), could increase federal income tax revenues by more than $1 trillion. It criticizes Congressional Budget Office estimates that are more conservative both in looking only at smaller budgetary increases, and in failing to include properly measured (or any) indirect revenue gains (i.e., those from taxpayers other than the ones who are actually audited).

A 10-1 Marginal Revenue Payoff from Increasing IRS Outlays – If we lived in a different world than the one we actually live in, the claim of a 10-1 revenue payoff here would verge on being self-refutingly absurd. So grossly under-funding the IRS would require such gross negligence, verging on deliberate sabotage, that one really wouldn’t expect it. But in fact this estimate a 20-plus year partisan war against the IRS that (at least, among those who are not completely ignorant) has reflected high levels of conscious bad faith. So it isn’t surprising at all.

Even absent the current partisan environment, however, one might expect Congress deliberately to under-fund the IRS. This offers two big advantages to members who are seeking reelection The big money folks in their districts or states don’t like being audited, and are presumably ready to put their money where their mouths are. And, if the IRS performs poorly because it has deliberately been under-funded, the under-funders can grandstand in front of ill-informed voters by holding hearings, complaining about it, etc.

Determining the Revenue Payoff – Again, the paper offers a rough ballpark estimate of $107 billion in IRS budget increases (over 10 years) as raising tax revenues by over $1 trillion. (Charles Rossotti has estimated $1.6 trillion.) This would mainly be backloaded in the 10-year estimating period, because it takes a while to ramp up. And there would be large revenue gains outside the 10-year window.

As noted above, the CBO excludes indirect revenues (much the larger piece) from official estimates, deeming them too uncertain. But this illogically responds to uncertainty with infinite discounting.

Excluding out-year revenues means that, after engaging in no present value discounting within the 10-year budget, one arbitrarily switches at the boundary line to infinite discounting. This is not an intellectually defensible approach.

Why stop at $107 billion over ten years and restoring 2011-equivalent revenue levels? – This may reflect political reality, but otherwise it makes no sense. Suppose one is still earning a 10-1 return on increased IRS outlays at the margin reached after the 10-year increase (which is only about $10B extra a year, after years of budgetary sabotage). Then it would be silly to stop there, although it is true that well-used budget increases might require first ramping up the iRS’s absorption capacity.

In describing how one might think about the question of just how high the IRS budget should go as the revenue payoff presumably (at some point) starts to decline, a useful structure is provided by the marginal efficiency cost of funds (MECF), first described by Slemrod and Yitzhaki in a 1996 article. Roughly speaking, the MECF from a given revenue increase = (Revenue + effect on taxpayers’ deadweight loss) / (Revenue – effect on government administrative costs).

Thus, for example, suppose the government raised $10 in tax revenues, and that the change increased taxpayers’ DWL by $3.50, while also costing the government $1 to collect. Taxpayers would be worse off by $13.50, while the government would have netted $9, so the MECF for this change would be 1.5.

The lower the MECF the better, all else equal. A perfect seamless lump sum tax would have an MECF of 1.0, and a Pigovian tax might come out lower. But real world tax instruments are likely to have higher MECFs that also tend to rise with the marginal use of the instrument.

Now suppose the government gets a 10-1 revenue boost out of increasing the IRS budget. The MECF for a $1 increase in the budget equals (10 + the effect on DWL) over 9. That yields a pretty low MECF unless the effect on DWL is high. But note that increasing IRS audits, while it would increase taxpayer DWL in some respects (such as requiring them to go through the audits), might also reduce DWL in some dimensions. Suppose for example, that it makes taxes, although higher, also somewhat more neutral because it reduces the effective tax preference for cheating and over-aggressiveness. It also might cause some taxpayers to give up the game and reduce costly evasion and avoidance effort.

Plus, it might increase equity in both the vertical and the horizontal dimensions. Given how little we audit high-income taxpayers, greater auditing might in practice make the tax system more progressive, while also increasing equality of tax treatment as between the honest and sketchy at similar pretax income levels.

MECF helps to show why one would want to stop well short of increasing auditing to the point where it broke even budgetarily at the margin. Suppose, for example, that a $1 increase in the IRS auditing budget increases revenues by only $1.01, because we have crossed over so far into the realm of diminishing returns. The MECF from doing this would be ($1.01 + effect on DWL) / .01, which does not look good at all.

Thus, in principle one should stop increasing audit levels (and other aspects of the IRS budget) when the marginal return, keeping in mind equity considerations as well, no longer looks good compared to alternative choices. But we would appear at present to be far, far short of that.

Additional points of interest raised by the paper

1) How would raising $1 trillion over 10 years in this way, mostly from high-income taxpayers, affect the merits of other proposed instruments for increasing progressivity, such as wealth  taxes or mark-to-market taxation? The answer, I’d say, is that this really isn’t an either-or choice. Those instruments should be used if (and only if) they score well enough in a distributional and efficiency-based analysis. Better IRS auditing capacity can cause such instruments to perform better than they otherwise would, however.

2) How should one account for budgetary out-years? The 10-year window, with its infinite discounting for things outside the boundary, is especially disastrous in cases, such as that of ramping up the IRS’s budget and audit capacity, where positive returns lie disproportionately outside the window. Curtailing the period of analysis does less harm when in-years and out-years are fundamentally alike. But it makes no sense to be as myopic as Congress deliberately is under existing budget rules. Infinite-horizon budget forecasting (under a defined set of policies that may not be sustainable) is a valuable intellectual tool, whether or not one wants to deploy it in official budget rules.

3) How should Congressional budgeting account for the revenue gains from increased IRS spending? Ignoring the gross revenue increases is nothing short of idiotic On the other hand, looking purely at the net revenue effect would turn net revenue-maximization into the implicit default, which isn’t quite right either.