Gamage, Framework for Optimal Choice of Tax Instruments, part 2

Here is Part 2 of an adapted version of my notes for yesterday’s session discussing David Gamage’s “A Framework for Analyzing the Optimal Choice of Tax Instruments.”

2.         THE PAPER’S ARGUMENT FOR USING MULTIPLE TAX INSTRUMENTS

Again, the paper suggests using all of a labor income tax, VAT, capital income tax, wealth tax, legal rules, plus perhaps more (subject to administrative costs) to address distributional issues.  But the argument is based on avoidance costs – not on how distributional outcomes should relate to underlying “true” information about, say, individuals’ ability, earnings, savings, or gratuitous transfers.

One can explain the paper’s main argument in two complementary ways.  First, suppose that potentially useful tax instruments tend to have a rising marginal cost of public funds (MCPF) as we start using them more.  MCPF is basically deadweight loss per dollar of revenue raised, adjusted for the social value or disvalue that is associated with a given outcome’s effect on after-tax distribution.  As you use any instrument more, the ratio of deadweight loss to revenue will tend to increase.  For example, doubling the tax rate may tend to quadruple the distortion.  Likewise, increasing audit rates, if the potential audits are ranked by expected payoff, means adding ever less fruitful new targets.  Keeping distribution constant, the aim is to equalize the ratio of deadweight loss to revenue raised at the margin for all instruments.

Second, and more distinctively, the paper distinguishes what it calls single-instrument vs. multiple-instrument tax avoidance.  Here the focus is on taxpayer actions to reduce tax liability.  This could involve, for example, reducing your labor supply, altering your mix of consumption or investment choices, or paying lawyers and accountants to work their magic for you.

Tax avoidance presumably has rising marginal costs per dollar of tax avoided.  For example, working (and thus earning) just a bit less than you’d prefer, taxes aside is not so bad, but working and earning a lot less starts to get really painful.  Or, the tax planners get the low-hanging fruit first, but then it keeps getting costlier to avoid the next dollar of tax liability.

With a continuous function for tax avoidance, you keep going until you disvalue at exactly $1 the cost of avoiding another $1 in taxes.  Let’s call that the satiation point.

Suppose there is just 1 tax instrument with a 40 percent rate on whatever it might be (wages, Haig-Simons income, consumer spending, wealth, etc.).  Any time you lower the tax base by $1, you save 40 cents of tax.  So you reach the satiation point when it costs you 40 cents (in utility terms) to get a dollar out of the tax base.

Now suppose instead there are 2 tax instruments, each with a 20 percent rate.  Multi-instrument avoidance reduces them both, so the government’s shift to two distinct tax instruments has no effect.  But suppose some avoidance techniques work on only one of the two systems.  For example, a loss-creating tax shelter works against the actual income tax, but not against a typical VAT.  Then you reach the satiation point at 20 cents, instead of 40.

Where only single-instrument avoidance is feasible, multiple tax instruments turn tax avoidance into a costly retail exercise, rather than a cheap wholesale exercise.  The result, less tax avoidance, is unambiguously good within the paper’s assumptions.  What is more, absent an offsetting concern about tax overhead costs, there would be no obvious limit to how many separate systems this analysis suggests that we should have. 

In my view, this analysis logically holds together.  However, the paper’s interpretation of it as suggesting more specifically that we should use all of a labor income tax, VAT, capital income tax, wealth tax, etc., puts me in mind of the following three issues:

(1) Information versus evidence – Note two distinguishable types of issues in tax system design.  First, what is relevant information that should affect tax liability? Second, how as a technological matter do we get at that information?  What evidence can we use to measure it?

By information, I mean something that we want to affect tax liability.  Examples might include labor income and, more controversially, returns to saving.  By evidence, I mean something that we use to get at the underlying information. 

David cites a paper by Roger Gordon and Soren Nielson that considers a VAT and cash-flow income tax that are assumed to have identical behavioral and distributional consequences when accurately applied, but involving different means of avoidance.  You avoid the VAT through cross-border shopping.  You avoid the consumed income tax by concealing receipts.  As is well-known in the tax policy literature, either receipts (sources of income) or consumer spending (uses of income) can be employed towards measuring consumption.  Gordon and Nielson show that it’s plausible one would want to use some of each instrument, so that taxpayers can’t avoid the entire tax along either tax planning dimension.

In this model, the VAT and the consumed income tax both seek to rely on the same information (i.e. how much was consumed), but use different evidence.  By contrast, a realization-based income tax, a VAT, and an estate & gift tax (to name three well-known types of tax system) differ in the information that they treat as relevant, not just in the evidence they use.  The VAT bases liability on consumption, the income tax aims to reach work plus saving, and the estate and gift tax aims at the making of gratuitous transfers (which implies underlying work and saving).

In theory, there is a right answer as to the extent to which tax liability should depend on these different types of information.  That right answer ought to be implemented as to the overall tax system, without regard to separate instruments.  (And yes, I recognize that the relevant information might be viewed merely as evidence of something else still, such as ability or opportunity or expected marginal utility of a dollar).

The mark-to-market income tax, VAT, and estate & gift tax may also, quite distinctly, use different types of evidence that invite different avoidance mechanisms.  For example, the realization-based income tax encourages strategic trading (hold the winners, sell the losers).  For the VAT, there’s unreported cross-border shopping.  The estate and gift tax obviously has a slew of avoidance techniques of its own, but also permits the tax authority to take a one-time in-depth view of everything that the taxpayer has on hand at death.

Suppose you like the different evidence but not the different information.  For example, you might like taxing estates as a backup to the income tax (both for fraud problems and the holding of appreciated assets).  But that would imply a different design for the estate tax than if you actually want tax liability to depend on bequests as an end in itself.

One last point about information: the existing income tax deliberately uses some types of information other than about income.  Examples include the itemized deductions for home mortgage interest and charitable contributions, along with the exclusion for employer-provided health insurance.  While the paper groups these items with other tax avoidance, they are in fact meant by the policymakers to affect tax liability.  In principle, if allowing them is the right policy, we would want to hold constant when using multiple instruments.

The paper notes that, in practice, such items may often be bad policy, and that the predilection to use particular ones seems oddly instrument-specific.  For example, VATs but not income taxes commonly offer reduced tax rates for food.

Is using more instruments likely to increase or reduce the political tendency to make bad choices regarding the use of information to affect tax liability?  The answer to this question is certainly far from clear.

(2) What is an instrument? – The paper defines “tax instruments” as any policy variable that a government might adjust to raise revenue or affect distribution.  Sometimes, however, “instrument” seems to connote instead formally distinct systems.  But you can have multiple instruments in the same system.  An example would be diversifying income tax audit selection techniques to reflect, say, both omissions suggested by counter-party reporting and “lifestyle audits” of people who appear to be doing quite well but haven’t reported commensurate income.

(3) What is tax avoidance? – Taxpayers may want to reduce reported tax liability whether doing so leads to a better or worse measure of the true underlying information.  Consider, for example, documenting legitimate business expenses, or making sure that losses will be deductible, notwithstanding the capital loss limitation, in instances where they truly are representative of overall economic results (i.e., one doesn’t have offsetting unrealized gains).  In principle, while taxpayers’ incentive to reduce tax liability may be too “strong” from a social standpoint (Louis Kaplow has an article on this), the socially optimal incentive is greater than zero.  The paper’s analysis of tax avoidance is premised on its being “incorrect” (not necessarily legally, but in relation to underlying information that is of distributional interest), and the correctness issue makes this more complicated, raising once again the issue of what information we consider relevant.