Tax policy colloquium, week 4: Gerald Auten’s Income Inequality in the United States, part 2

My prior blog post noted that, while the Auten-Splinter (AS) paper, despite its clear merits, seems in tension with the vast anecdotal evidence (and other empirical studies) suggesting that there has been a substantial rise in U.S. high-end inequality over the last three decades, there are ways of getting past the initial head-scratching. Let me start with the question of why high-end inequality might matter, which relates to how one might try to measure it for different purposes, and then turn to a few of the particular empirical issues in the debate, along with the relationship between AS and the most recent work in the same area by Piketty, Saez, and Zucman (PSZ).

AS seek a “broad and consistent income measure” for purposes of measuring high-end inequality. For the most part, they look to market measures of income earned by different households and individuals, and at taking national income and trying to allocate it to people, although this is influenced by issues of what we might reasonably infer in some circumstances about underlying utility. But why would one care about this?
An initial point, and I think closest to their purpose, is as follows. If one thinks in terms of a national “pie” of income that the people in the country divide somehow, then it is of interest what percent of it is held (or “eaten”) by people at the very top. This might implicate concerns about distribution or vertical equity. It also might be a relevant input to thinking about the marginal utility that might be gained and lost by changing the distribution. So let’s say the following:

Reason 1 for caring about high-end inequality is that we care about the distribution of the total.

Reason 2 is that we draw inferences from it about marginal utility. All else equal, if the rich are richer and the poor poorer, the marginal utility gain from transfering a quantum of consumption from the former to the latter might be expected to increase.

But this does not exhaust the possible reasons for concern about high-end inequality. AS mention, at the start of their paper, three further issues (developed in earlier economic literature) that might be raised: “Increased inequality could be an indicator of greater concentration of political power and increased rent-seeking … or a result of increases in the bargaining power of top earners for compensation.” Hence, they note, three particular ills might accompany rising high-end inequality.  Let’s call these Reasons 3 – 5, given the two I noted above. In AS’s words, they consist of:

Reason 3: “decreasing institutional accountability due to concentrated power,”

Reason 4: “decreasing economic efficiency due to rent-seeking,” and

Reason 5: “stagnating middle-class wages due in part to shifts in relative bargaining power.”

Let me add to the list two more possible ills from rising high-end inequality:
Reason 6: If people care about relative consumption (hint: they do), there might be what Robert Frank calls “expenditure cascades” radiating from the top down, potentially reducing subjective welfare for people at all levels.

Reason 7: Research by Richard Wilkinson and Kate Pickett powerfully suggests that greater inequality is correlated with, and apparently causes, increases in social gradient ills, ranging from violence to drug abuse to alcoholism to suicide to other stress-related health problems. Even the rich are affected. This matters not just for its own sake (social gradient ills are bad), but as a diagnostic indicating broader negative effects on people’s subjective wellbeing in more unequal societies.

This list is not necessarily exhaustive. One may wish to add other reasons, or delete those listed above that one deems not to be serious concerns. But here’s the thing: If you still have several reasons on your list, they are unlikely to motivate identical measures of high-end inequality for purposes of assessing them.

Here is a simple example. AS address the question of whether we should think of, say, healthcare benefits provided through Medicare, Medicaid, and employer-provided health insurance as providing benefits that are worth their cost to the recipients. This clearly must be addressed with regard to Reason 1 – it tells us about the value of the “pieces” of the “pie” that people in different households are getting.

But how much does it affect, say, Reason 4 or Reason 6? Do we think there’s less rent-seeking at the top if poor people are getting good rather than mediocre value from Medicaid? Really no, except insofar as prevalent rent-seeking might increase the likelihood that they might be handed bad healthcare. Do we think there will be less by way of “expenditure cascades” that are mainly a product of visible public Veblenesque consumption? People aren’t competing over cancer and diabetes treatment in quite the same way as they might be over their homes, weddings, vacations, food consumption, and air travel.

One reason AS find that high-end inequality hasn’t risen as much as one might otherwise have thought is that people lower in the distribution, due to advances in medical technology, are getting far more expensive – but also better – healthcare than previously. PSZ note in their work that wage increases among the bottom 50% have been substantially eaten up by rising healthcare costs. These are two different ways of saying the same thing. The AS way of putting it might be better in relation to Reason 1, but it’s conceivable that all of Reasons 2 through 7 (to the extent affected at all) are better aligned with PSZ’s framing.

Here’s another tough measurement issue that both AS and PSZ have had to wrestle with. Say I save during my working years for my retirement years, perhaps through tax-favored retirement saving that I accumulate upfront and withdraw/use at the back end. How should we measure this?

Measuring it on an accrual basis rightly captures that the individual who accumulates retirement saving during her working years is better off than one who consumes the same amount that year but doesn’t have anything left over for retirement saving. The saver preferred to save – why should she be treated as worse-off than if she had spent more on current consumption in lieu of saving for retirement.

But if you measure consistently on an accrual basis, then the person who is living off her savings once she retires appears, in the measure, to be poor, even though she isn’t. And double-counting it for both periods leads to a falsely high measure of lifetime income.

To add in a factor that both AS and PSZ leave out for methodological reasons (since they are doing annual snapshots), what about life expectancy differences? If I have $X (and let’s suppose it’s a tidy sum) at retirement, perfectly self-annuitize, and spend it all before I die, was I better-off in the scenario where (a) I correctly forecast that I would live a very short time and hence blew through it rapidly, than (b) in that where I correctly forecast that I have many years left, and prudently spread it over a longer period?

Annual scenarios make me look worse-off in (b) than (a), whereas the truth is the other way around. Even under a pure accrual method, which misclassifies me as poor in my retirement years because the retirement savings have already been counted during the build-up years, I appear to be poor for more periods, by reason of my living longer, so I make high-end inequality look worse even if in fact I’m at the top throughout in any real sense.

It certainly matters to the assessment of high-end inequality that the life expectancy gap between rich and poor has been growing in the United States. But if we go back to Reasons 1 through 7, it doesn’t matter identically for all of them! It clearly makes #1 look worse, as assessed from a longer than merely annual perspective, it has ambiguous effects on #2 (we could really benefit poorer people if we enable them to live longer, but if we take the discrepancy as given then having a longer lifespan increases the marginal utility of a dollar of lifetime income), and, as to the rest, who knows (or at least, it would require more analysis).

So we have complex and ambiguous issues, not reducible to being assessed via a single-bullet measure of current annual shares of the pie. Plus, you could have exactly the same such measure in Society 1 as opposed to Society 2 – thus equalizing them in terms of Reason 1 – yet the two societies might differ substantially with regard to Reasons 2 through 7. Hence the point, for example, that even if high-end income concentration in 1960 was closer than we thought to being similar to that today, its adverse effects may have been less, e.g., because people were less inclined to use their wealth aggressively in either politics or Veblenesque status competition through conspicuous consumption.

One last point before I close: How does the AS versus PSZ standoff look today? The latter have shifted from Piketty-Saez approach of starting from taxable income and moving up (effectively critiqued by AS in various respects) to an approach of starting with a measure of national income and allocating it down.

AS find that the top 1% have a national income share of 10.2%, whereas PSZ come out at 15.7%. This is still a sizeable different – albeit, not necessarily large enough to motivate fundamentally different bottom line viewpoints about whether we have a problem from rising high-end inequality.

PSZ report that their finding 5.5% more of national income concentrated at the top arises as follows:

1) 1% comes from PSZ’s including retirement savings as it accrues, rather than when it is later spent. As noted above, there is clearly some merit to this position (although its upside of more properly measuring the income of workers who save comes with a downside of overestimating the poverty of retirees who have saved).

2) 0.4% comes from assuming that deficits will be funded half through tax increases and half through spending cuts, rather than all through tax increases as AS assume. I find the PSZ scenario more plausible, although either alternative raises the question of how we should think about the current distributional implications of future policy changes that have not been adopted yet and that remain deeply unpredictable.

3) 0.2% comes from differences in how they treat married couples. This is a very interesting and intricate subject – we spent a lot of time on it at yesterday’s AM and PM colloquium sessions, but I lack the time and space to cover it properly here.

4) 0.3% comes from only AS’s removing young and dependent filers from the lists, on the view that their low income measures may be wholly uninformative if, say, they are living with affluent parents.

5) 2.6% comes from PSZ’s treating unreported income as concentrated at the top, whereas AS allocate it ratably to reported income. I see a number of strong reasons (although, again, no time or room to run through them here) for leaning heavily towards PSZ on this issue.

6) 1% comes from other methodological and data source differences between AS and PSZ.

Suppose one were to run a totally back-of-the-envelope scoresheet on this. For example, say we gave PSZ half of #1, all of #2, half of #3, none of #4, and all of #5. Then, even if we gave them none of #6, we would still have raised the AS estimate of a top 1% income share of 10.2% by 3.6%, to 13.8% – almost two-thirds of the total difference. Plus, I’ve suggested that some of the reasons why the top 1% income share isn’t higher – e.g., because people at the bottom do at least get some medical treatment that is not only costly but also valuable – might not matter so much for Reasons 2-7, even if quite relevant to Reason 1.

I think this plausibly puts us in a scenario where concern about high-end income inequality that we believe to be both high and rising is by no means rebutted or shown to be an “illusion” by AS’s nonetheless valuable and admirable contribution to the debate.