Tax policy colloquium, week 4: Gerald Auten's Income Inequality in the United States, part 1

Yesterday at the colloquium, Treasury economist (at the Office of Tax Analysis) Gerald Auten presented his paper, co-auithored with David Splinter of the Joint Committee on Taxation, entitled Income Inequality in the United States: Using Tax Data to Measure Long-Term Trends.

This paper is an important entry in the ongoing empirical debate among economists regarding the rise of high-end inequality. I think it's worth offering a bit of the backstory here.

My view, and that of many others, has been that the rise of high-end inequality in the U.S. (and elsewhere) in recent decades is both unmistakable and highly consequential. Let me offer here just a few anecdotal examples off the Internet, illustrating this pervasive (I would argue) phenomenon:

--A 2016 NYT article, entitled "In an age of privilege, not everyone is in the same boat," describing how consumer businesses increasingly direct their efforts towards extremely high-end consumers, creating sectors that feature a "money-based caste system."

--A 2015 WaPo article, entitled "In 2016 campaign, lament of the not quite rich enough," describing how, in the days when Jeb Bush appeared to be riding high, campaign finance had moved in such a way that presidential campaigns had little interest in courting mere multi-millionaires. One needed to be a billionaire to really get all those lap kisses and special access.

--A pair of delightful 2014 blog posts by economists, the first called "Inequality in the Skies" and the second "Fly Air Gini," describing how the allocation of seat space inside commercial airplanes has become more high-end skewed. (This despite the fact that the real high-flyers, or top 0.1 percent, aren't to be found in business class or even first class - they are using their own personal or corporate jets. The influience they nonetheless have on commercial flights is related to what my recent High-End Inequality Colloquium teaching colleague, Robert Frank, calls "expenditure cascades," involving the radiation downward through the wealth distribution of influence from high-end consumption patterns.)

Anyway, it seems to me verging on undeniable that this thing at the top has happened. Widespread agreement about the reality and importance of this phenomenon helped to trigger the extremely high-profile reception of Thomas Piketty's 2014 tome, Capital in the 21st Century.

The Piketty book, and its underlying research (much of it conducted with Emmanuel Saez, and subsequently also with Gabriel Zucman), had a lot going for it. It took a fresh look at previously under-utilized data, it reached important conclusions that seemed clearly correct, and Piketty even made the book more fun by discussing the novels of Austen and Balzac.

But there were a couple of things to create potential unease about this work. First, there was reason to question Piketty's causal theory for the rise in high-end inequality, which he based on "r > g" - the excess of the rate of return over the rate of growth in the economy, causing capital-owners to hold an ever-increasing share of the overall pie. Not only were there some soft spots in the underlying theoretical reasoning, but it seemed clear that, in the U.S. at least, the rise of high-end inequality was driven primarily by rising high-end labor income inequality.  (Joe Bankman and I wrote about Piketty's work, in part critically, here, in connection with an interdisciplinary NYU-UCLA conference on the book that Piketty attended.) Ought the problems with the underlying theory support broader skepticism about the work - or was it simply a causal dispute? It was hard to know.

Second, while I am not myself an empiricist, I began to hear suggestions from various economists - straight-up folks without intellectual axes to grind - that they were uneasy about some of the details in the (at that stage mainly) Piketty-Saez work. Reading the footnotes, in effect, seemed not to inspire universal confidence. And there are people to whom it sometimes seemed that debatable interpretive choices - although inevitable, given the data challenges in the enterprise - seemed to have a tendency to lean towards making the estimates of rising high-end inequality higher, rather than lower.

Yesterday's colloquium presenter, Gerald Auten, got into the debate as a consequence of his in effect carefully reading the footnotes in connection with IRS tax data. Among the things he noticed was that the Piketty and Saez ("PS") methodology seemed to show a huge uptick in high-end inequality right around the enactment of the Tax Reform Act of 1986.

Now, as an aside, it is plausible that TRA 86 might have tended over time to boost high-end inequality (it aimed to be distributionally neutral, but only within a 5-year budget window and without looking at the top 1% - it was designed in light of effects on income quintiles). But not quite so immediately. The jump one computed by replicating the PS methodology seemed to rely a lot on (a) the fact that TRA 86 induced a lot of people to shift out of corporate form, so that the same earnings they had previously been generating through C corporations now appeared on their individual returns, and (b) the fact that, under TRA 86 provisions targeting tax avoidance by rich people, more children of rich people now had to file tax returns on which they might have trivial amounts of taxable income. These looked as if they were the tax returns of poor people, but actually weren't. The growth in seemingly "poor" taxpayers made the income concentration at the top look greater than it would have, had the kids (as previously) not filed.

In short, there were technical reasons unrelated to the true growth of high-end inequality why it seemed to grow sharply, under the PS methodology, right around 1986. And if one took that growth away from the estimates, what was left seemed potentially much smaller than PS had been trumpeting.

Auten and Splinter therefore started preparing their own estimates of high-end inequality trends from 1960 through the early twenty-first century, and came up with a much more modest growth estimate. Indeed, the current paper draft concludes that its findings support an "alternative narrative ... [to that in PS] when consistent and broad measures of pre-tax and after-tax incomes are used: changes in the top one percent income shares over the last half century are likely to have been relatively modest."

As I see it, this can put one in a strange place, as a consumer of the empirical debate. Much of what Auten and Splinter (AS) do - such as addressing the anomalies that I noted above - seems clearly correct. What's more, they clearly have no axe to grind, and simply want to get the data analysis right.

On the other hand, it kind of seems like their bottom line conclusion can't quite be correct. I'm reminded of the old joke: "Who are you gonna believe, honey - me or your lying eyes?" It's not just bias or media trends - our not so lying eyes surely have seen a real phenomenon unfolding over time. How could it not be happening after all?

This angle is well-captured in a recent Vox blogpost by Dylan Matthews, entitled "A new study says much of the rise in inequality is an illusion. Should you believe it?" This in the end is not ultimately my take on how to think about the significance of the AS paper, but I will admit that I started there.

Now, as I'll discuss in a follow-up blog entry, there may actually be several good reasons, based purely on data interpretation, for revising the AS "narrative" very substantially towards that in PS. Indeed, the latest version of the latter - PSZ for Piketty-Saez-Zucman - really helps in framing the empirical gap in ways that could well come out sufficiently in their favor to revive their "narrative" relative to that in AS.

Let us suppose, however, that this remains indeterminate. How should one's prior about the compelling, even if anecdotal, evidence of the contemporary rise in high-end inequality affect one's response to the empirical analysis in AS?  Here are two possibilities:

--It can function as a Bayesian prior, influencing how one evaluates uncertainties in the data, even though one must always be open-minded,

--It can motivate considering how the AS analysis and what one has seen happening over the last three decades could BOTH be true. For example, if empirical measures of high-end inequality came out exactly the same in 1960 as today - which would go well beyond their finding merely "modest" high-end growth - might the changes we've observed nonetheless have been happening? I find it very plausible that cultural and behavioral changes since that time could have this upshot. If people at the top, in 1960, didn't flaunt and flex their wealth to nearly the same degree then as they do now, either in consumer markets or in politics, then the changes so many of us have observed could reflect that, rather than (just) material changes in distribution.

I'll develop all this further in the next blog entry (after participating, in a few minutes, in an NYU Law Forum where I and three colleagues will discuss the 2017 tax act with an audience mainly of law students).

But first, a last quick point to sharpen (I hope) interest in the follow-up. High-end inequality is not a single entirely well-defined "thing," nor does it matter for just one reason. For each reason why it might matter, a different mode of measurement might be appropriate. So there is no unitary answer to the measurement issues that are raised by AS, PS, and PSZ.