After the windup in my prior blogpost, here is more of the pitch concerning the paper itself, organized by its 4 main parts.
1. Why progressivity matters – Arguably, this section title should change to better tip off to readers its content and main takeaway. It argues that distribution is what matters, not “progressivity” within the contours of a tax system that is merely a subset of the broader fiscal system, and shows how the tax policy literature has moved over the decades in exactly this direction.
In earlier decades, the literature tended to discuss “ability to pay” and how it related to how progressive the tax system ought to be. For example, declining marginal utility (DMU) was understood as supporting rising absolute tax burdens as (say) one’s income increased. If DMU rose fast enough, one would want average tax rates (ATRs), not just absolute tax liabilities, to rise with income, and this is how progressivity was typically defined.
As the paper notes, one problem in the “ability to pay” era was the lack of a clear basis just how fast (if at all) ATRs should rise with income. But another problem was that, as I noted in the prior blogpost, ATRs could be misleading absent a budget-neutral and indeed revenue-neutral comparison. One couldn’t ignore the spending side in making informative distributional assessments.
Next in the tax policy literature, as the paper notes, came Blum and Kalven’s famous early-1950s piece, The Uneasy Case for Progressive Taxation. After a whole lot of very sharp argumentation, Blum and Kalven famously conclude that the case for progressivity is best grounded in the aesthetic view that inequality is “unlovely.” Viswanathan rightly notes that this presupposes focusing on overall distribution – although Blum and Kalven emphatically that one simply can’t do a thing about the incidence of government spending, which instead should be treated for analytical purposes as if the revenues, rather than being used in any way, had simply been thrown into the sea.
That aspect of the Blum-Kalven analysis is now a bit dated, but mainly because the world has changed. They had in mind a bunch of tedious and unilluminating debates in the literature of the time about who benefits more from, say, police and national defense spending: the rich because they have more property to defend? The poor because the rich could hire private armies if they had to? While that point about the difficulty of allocating certain types of public goods spending remains valid, the changes in the federal budget since the early 1950s – e.g., the rise of Medicare and Medicaid – means that a lot of things on the spending side now can be meaningfully allocated.
Anyway, though, the Viswanathan paper rightly notes that Blum and Kalven are already, despite that disclaimer, moving towards a whole-budget view. Optimal tax theory (OTT), which it discusses next, then goes the rest of the way in that direction.
2. Definitional ambiguities – This very useful section of the paper includes the contribution it makes that I suspect will be best remembered: the introduction of the term “progressivity base.” Defining such a thing is an application of devising vertical distributional measures more generally.
There is no one right answer to what the progressivity base for distributional assessments (including those of tax progressivity) should be. But we can see the concept at work, albeit without the name and sometimes without a clear expression of what is going or or why, in such circumstances as the following: (a) the Joint Committee on Taxation uses “expanded income” to measure the progressivity of given tax legislation, (b) NGO or other private-sector analysts use income or wealth to assess the progressivity of, say, property or consumption taxes.
If I had to list some of the more plausible progressivity base candidates, I would include at least: (a) expanded income (and at the limit Haig-Simons income, (b) wealth, (c) wealth plus the present value of expected future earnings, and (d) lifetime income including inheritance. A key issue here, of course, is the choice of time frame, e.g., snapshot versus a longer period.
Anyway, in a standard ATR computation the progressivity base supplies the denominator, and the numerator is something like taxes paid or accrued or borne economically. (Again, there are time frame issues here.) But I think the paper could call out at least one dubious entry in the field that it mentions. The Cato Institute argued recently that the 2017 tax act was progressive because poor people had a larger percentage reduction in tax liability than rich people. Thus, if a poor person’s liability went down from $1 to zero (i.e., by 100%), while Jeff Bezos’ liability went down by only 99.9% (e.g., from $100 million to $100,000), this would be a progressive change.
With all due respect to Cato – which not only does some good work, but indeed (on a personal note) once was kind enough to publish an article penned by me – I don’t think this was their best moment. The change in my tax liability as a % of everyone’s tax liability is simply not a good enough measure of a given tax enactment’s effect on me – especially absent a full-budget, budget-neutral comparison – to be plausible as the numerator here. Its use in that instance smacks of trolling – as in, “we like the 2017 tax act for other reasons, but just to blow some smoke in the air we’re going to use the only measure we can think of that yields the result we want.” So I think it would make sense for the article to call them out on this, in the friendly hope of encouraging better work from them the next time.
3. Calculational ambiguities – The article notes that, in assessing the numerator (tax burden), it would be desirable to adjust taxes paid to reflect determinations of incidence and deadweight loss, insofar as this was sufficiently feasible.
This is a very good point, although its feasibility in particular cases will vary. But to give a simple example, showing how both the numerator and the denominator may change, suppose that $100 taxable bonds pay $10 of interest, while otherwise similar but tax-exempt $100 municipal bonds pay $8 of interest, reflecting a $2 implicit tax. I gather that the JCT, under its “expanded income” measure, would include the $8 in the denominator. In principle, however, one should at least arguably include the $2 implicit tax in the numerator, and $10 in the denominator.
What about deadweight loss? E.g., X would have earned me $10 of income and generated $2 of true surplus (i.e., doing what I needed to to earn this income would have generated disutility that I value subjectively at $8.) But because the tax on it is $3, I don’t do X, and earn zero instead of $10
True, we are unlikely this sort of question in practice, due to real world measurement difficulties, but in principle it should be addressed theoretically.
In practice, all of deadweight loss, externalities, and internalities may be very important to how we think about particular tax provisions or proposals, including distributionally. The difficulty of incorporating them into an ATR measure should not lead us to forget that they are important.
Here’s a real world practical case. Suppose a soda tax is borne mainly by poorer people but improves their welfare, because they reduce their consumption of unhealthy glop. The revenues will show up as regressive, the lost subjective surplus from reduced consumption won’t show up, and the increased welfare if people’s health improves also doesn’t show up. This is important stuff to think about, whether or not we can make any progress in deciding how to use it in an ATR measure.
A further topic that the paper identifies is macroeconomic effects. For example, suppose we accepted Kevin Hassett’s 2017 claim that cutting the US corporate tax rate would raise workers’ incomes by an average of $6,000 per household. I personally think that this claim was wildly wrong, but if it was indeed reasonably expected to happen, and within the relevant timeframe, then of course one should consider it.
There is a further procedural issue of whether we should allow such claims to affect respectable computations, if we fear that they will be misused. But perhaps the experience of the JCT in keeping the measurement of “dynamic” revenue effects within a reasonable and respectable range, by using plausible and internally consistent models, offers a favorable precedent for allowing macro estimates to be used here as ell.
A final big issue here is timeframe. For example, if a VAT was enacted in the US, the expected current year liability effect on people with a lot of wealth would be a whole lot smaller than the effect on the present value of their expected long-term liabilities. So time frame, along with the distinction between transition and permanent effects, is an issue that requires attention and clear statement (possibly along with alternative presentations).
4. Improving progressivity assessments – Among the issues this section of the paper currently discusses is earmarking, as in the case where one has a budget-neutral enactment because it raises revenues that are linked to a particular spending use. Earmarking is a tricky subject, both because there are multiple issues raised by its political economy and other merits, and because its reality in substance may be hard to gauge. For example, Social Security payroll tax revenues are ostensibly earmarked to pay for Social Security benefits, but are they really? Suppose Congress notionally “includes” a current year Social Security surplus when deciding how big a deficit for a given year to tolerate (i.e., they look at least implicitly at the unified, not the on-budget, deficit measure). And suppose future benefit payouts turn out not to depend so much on the stated measure of the Social Security Trust Fund. Then the earmarking’s degree of reality may be open to fair debate.
Another thing that this section might do is offer guidelines for real-world tax progressivity measures that are done either by the Joint Committee on Taxation and other government entities, or by NGOs such as the Tax Policy Center, Cato, and, say, the folks at Penn-Wharton should they decide to enter this area (as they are certainly qualified to do, and my apologies if I am overlooking that they have already done so). One might lay out a set of alternative best practices – e.g., you could reasonably do any or all of A, B, C, or D, but you really shouldn’t do E or F – without being overly restrictive or narrowly ideological. And one might also note how underlying theoretical constructions (e.g., noting incidence and deadweight loss issues, even if they can’t be measured properly) might lead, in some circumstances, to systematically different results.
Clearly, addressing all the issues that I find interesting here would take multiple papers, and they surely don’t all belong in this paper. But it already offers a very welcome advance, and it would be great to see follow-up papers, whether by Viswanathan or anyone else, that push the analysis further.