NYU Tax Policy Colloquium, week 11: Owen Zidar on capital gains revenue estimating, part 2

 My prior post discussed the most important takeaways from the work that Owen Zidar presented yesterday at the NYU Tax Policy Colloquium: namely, that raising capital gains (CG) rates might raise a whole lot more revenue than official revenue estimates suggest. Here I shift to less directly consequential rumination regarding issues that the work either directly or indirectly raises:

1) Labor supply elasticity versus capital gains realization elasticity - While both labor supply and CG realizations can respond to tax rates, they involve very different types of choices. Supplying additional labor to the marketplace for additional compensation reflects an important real life choice. One typically has only a fairly limited ability to re-time it. What one chooses to do is likely to be strongly influenced by the “primitives” in one’s choice framework.

Labor supply elasticity may vary sharply between groups. For example secondary earners and older workers (those nearing retirement age) have typically had far labor supply higher elasticity than other workers. Important changes over time as to these groups – and as to women’s labor force participation, whether or not they are “secondary earners” in a particular household – may cause one to expect overall elasticity changes over time. So might changes in the labor market, e.g., the rise of the “gig economy” and therefore of part-time work. But absent any such things, one might have a presumption of relative stability in labor supply elasticity, because it’s strongly influenced by “primitives” that one might expect to be fairly stable.

CG realization is very different. It’s merely a swap between two things of presumptively equal value: an asset of some kind, and in the most typical case cash.

Why would one even bother to do this? Suppose you hold appreciated stock. Three main reasons why you might choose to sell it for its market value are (a) to get your hands on actual cash, (b) to rebalance your portfolio and/or adjust your risk exposure, and (c) because you believe it’s actually worth less than the market price, hence want to get out for that while you still can.

But as Joseph Stiglitz noted in a famous, and quite prescient, 1983 article, with complete capital markets it would NEVER be necessary to sell, as a practical matter, even presuming any of the above motivations. E.g., you can borrow instead of selling, and use derivatives to adjust your risk exposure and/or to short anything that you believe is currently overvalued.

Code section 1259, first enacted in 1997, admittedly limits the extent to which you can achieve the economics of a sale without having it treated as a sale for tax purposes. But its scope is unclear, and it unambiguously permits some degree of sale economics replication – you just can’t go quite all the way.

An obvious conclusion from all this is that one would be wholly unsurprised to find that CG realization elasticity was significantly higher than labor supply elasticity (as, indeed, a huge body of empirical work confirms). But also: we clearly have less of a good understanding of what drives CG realization behavior, and of how it relates to presumptively stable underlying “primitives.” This may tend to weaken any premise that CG realization elasticity should generally be expected to remain stable over time. It may instead respond strongly to degrees of market completeness, how section 1259 is working, aspects of investor mood and expectation formation that we don’t understand very well, etc.

This supports the conclusion that relying on past work concerning CG realization elasticity may illuminate the current world less than one would have liked. I’d tend to consider, say, 1990s work on labor supply elasticity more predictive of today’s world than 1990s work on CG realization elasticity. And this in turn, to my mind, adds plausibility to the paper’s finding higher CG realization elasticity than studies from decades ago had found (although the paper uses data from over a long period).

2) Ten-year estimating period - revenue estimates typically go just 10 years forward. But under the paper’s model of TP behavior, this results in underestimating the revenue that would be derived from a CG rate increase. Again, a key point in the paper is that, even if I hold rather than sell today because the CG rate is high rather than low, there is still a decent likelihood of my selling the asset in the future (rather than holding it until death), even at the current rate. That prospect should be included in revenue estimates – although it’s not clear how extensively the JCT does so – but even a ten-year estimate will miss it insofar as the delayed sale lies more than 10 years in the future.

Should this be taken into account? Of course it should, since it has a real present value today. And the question isn’t just one of whether revenue estimates should run more than 10 years forward. Budgetary conventions should be our servants, not our masters. And a 10-year cutoff does less to distort one’s vision when all years are typically the same, than when out-years systematically differ from in-years. In the latter set of cases, ad hoc adjustments can help to reduce the distortion of reality that the ten-year cutoff risks imposing on us. This has been done before in other types of cases (e.g., ignoring “off-budget” net positive current year Social Security cash flows, given the expected long-term shortfall).

3) Section 1014 (step-up in basis at death) - Replacing section 1014 with a realization at death rule for appreciated assets (along with taxing the appreciation when property is transferred by gift) would certainly be expected to raise both current CG revenues and the revenue-maximizing CG rate. It would mean that realization was always a “now or later” choice, rather than being potentially “now or never.”

CG realization is also currently tax-deterred by the time value of money, which causes deferral to reduce the liability’s expected present value. But with interest rates being as low as they are today, deferral isn’t worth a great deal, and the timing issue becomes more like that under the “new view” of deferral in the corporate and international settings. In those settings (and that of CG realization if the interest rate is actually zero), deferral does nothing to reduce the PV of one’s liability. Instead, its only advantage is to preserve the option value of awaiting a decline in the applicable tax rate.

With repeal of section 1014, legislative change would be the only remaining mechanism for delay’s reducing the applicable CG tax rate. True, that is potentially a potent reason for delay, but less so than if one is certain of reducing the rate to zero, merely by holding until death.

The paper’s analysis suggests, however, that section 1014 actually matters less than we had thought. If people are more frequently selling a few years down the road than we had thought, and less frequently holding until death, then “now or later” matters more than we had thought, and “now or never” less.

A further implication might be that section 1014 repeal would both raise less revenue than we had thought, and have a smaller impact than we had thought on CG realization elasticity. Not because the extra revenue disappears – but because more of it was already within our reach, even with higher rather than lower CG rates.

4) How close to the revenue-maximizing CG rate should we be? - Normally, one wouldn’t want to be all that close to the peak of the Laffer curve for a particular tax instrument, because just short of the peak there is a very high ratio of marginal deadweight loss to marginal revenue.

But on the other hand, suppose one agrees with Diamond-Saez (and also Saez-Zucman) that marginal welfare weights at the very top of the income distribution should effectively be zero. Then one might be fine with applying the revenue-maximizing rate to individuals at the very top, be it with respect to CG or anything else.

This is a familiar point in the literature (whether or not one agrees with the Diamond-Saez-Zucman view) but I would add a point pertaining to the difference between labor supply and CG realization. Specifically, the deadweight loss imposed on people at the very top by high rates may have very importantly different characteristics in those two settings.

If the labor income rate on someone at the top of the distribution is too high, suppose that he or she therefore substitutes leisure for work. By contrast, if the CG rate on such an individual is too high, suppose that he or she substitutes continuing to hold appreciated assets for selling them. These two responses may have very different social welfare implications.

Substituting leisure for work might reduce the negative externalities from high-end wealth concentration. For example, as Bob Frank has noted, leisure generally isn’t a positional good, whereas wealth-funded material consumption often is. So the deadweight loss (from substituting leisure for work when, taxes aside, one would not have done so) is offset to a degree by a reduction in negative externalities.

Now consider holding rather than selling one’s appreciated assets, and thereby incurring deadweight loss because one wanted (all else equal) to adjust one’s overall investment position. One is still just as wealthy as one was before, and it’s not obvious how this would similarly involve reducing negative externalities from high-end wealth concentration.

All else equal, such thinking might support, at the top of  the distribution, a CG rate that was lower than the labor income rate even if the relevant elasticities at the same. Also, it shows why one might conceivably favor a labor income rate at the top that was above the peak of the Laffer curve (insofar as the revenue loss was offset by reducing negative externalities), but it seems less likely that one would want to do this with respect to CG realizations.