Something Wrong with the Framing of CEO Pay in Steven Kaplan's Feldstein Lecture? Larry Mishel Makes a Good Case That It Is
I have lots of respect for both Steve and Larry, but it seems to me that Larry wins the first set, 6-3:
I compare the average estimated pay of S&P 500 CEOs to the average adjusted gross income (AGI) of taxpayers in the top 0.1 percent of the income distribution. Figure 2 shows that average estimated pay for S&P 500 CEOs, relative to the average income of the top 0.1 percent, is about the same in 2010 as it was in 1994…. Over the last twenty years, then, public company CEO pay relative to the top 0.1 percent has remained relatively constant or declined. These patterns are consistent with a competitive market for talent. They are less consistent with managerial power. Other top income groups, not subject to managerial power forces, have seen similar growth in pay.
Mankiw, Kaplan, CEO Pay and the Defense of the 1 Percent | Economic Policy Institute: Greg Mankiw, in his defense of the top one percent (pdf), notes that “the key issue is the extent to which the high incomes of the top 1 percent reflect high productivity rather than some market imperfection,” and quickly turns to a discussion of CEO pay. Mankiw’s got a point—so let’s discuss whether or not CEO pay simply reflects compensation for ‘talent’ and productivity. Mankiw does not present any evidence on whether CEO pay reflects high productivity: rather, he offers an argument that corporate governance is not problematic, using research by University of Chicago business school professor Steve Kaplan as his evidence. In fact, the chief claim that CEO pay tracks that of other talented workers also comes from Kaplan, who has a paper (not yet public) in the forthcoming Journal of Economic Perspectives issue along with Mankiw’s contribution and a paper from me and my colleague Josh Bivens.
In this post, as promised in a prior one on Mankiw’s data claims, I draw on the evidence presented in our paper to show that CEO pay has grown far faster than that of other very high wage earners (the top 1/1000th) and that the CEO advantage relative to other very high wage earners has grown more than the college wage premium. We also demonstrate that Kaplan’s own data series shows the same pattern. A fair-minded review of these data, in our view, leads to the conclusion that the spectacular growth of CEO pay does not simply, or even primarily, reflect the market for talent, or some imagined increase in CEO productivity….
The first benchmark is the one Kaplan employs: the average household income of those in the top 0.1 percent developed by Piketty and Saez (xls). The second is the average annual earnings of the top 0.1 percent of wage earners based on a series developed by Kopczuk, Saez, and Song and updated in the State of Working America (Mishel et al. 2012)…. The wage benchmark seems the most appropriate one since it avoids issues of household demographics—changes in two-earner couples, for instance—and limits the income to labor income (i.e., it excludes capital income). Both the ratios and log ratios clearly understate the relative wage of CEOs since executive pay is a nontrivial share of the denominator, a bias that has probably grown over time….
CEO compensation grew from 1.14 times the income of the top 0.1 percent of households in 1989 to 2.06 times top 0.1 percent household income in 2010, Kaplan’s metric to measure CEO pay relative to that of other highly paid people. CEO pay relative to pay of top 0.1 percent wage earners grew even more, from 2.55 in 1989 to 4.70 in 2010…. Is this a large increase? Kaplan’s Feldstein lecture (page 4) concluded that CEO relative pay “has remained relatively constant or declined.” Kaplan’s CATO paper (page 14) finds that the ratio “remains above its historical average and the level in the mid-1980s.” It may seem odd that the conclusions differ but it isn’t: Kaplan made a (still unacknowledged) computation error in the Feldstein lecture that was corrected in the later paper.
In both papers Kaplan concludes that CEO pay seems to track that of other high earners but never provides any metric…. The figure below, also from my paper on CEO pay, does so, by putting these series in historical context, presenting the ratios back to 1947. Kaplan’s ratio of CEO pay to top household incomes in 2010 (2.06) was nearly double the historical (1947–1979) average of 1.11, a relative gain roughly equivalent to each CEO adding the total income of a top 0.1 percent household to their relative pay. CEO pay relative to top wage earners in 2010 was 4.70 in 2010, 1.54 higher than the historical average of 3.08 (adding the wages earned by more than 1.5 high wage earners to CEO relative pay). As the data in the table show, the increase in the logged CEO pay premium since 1979, and particularly since 1989, far exceeded the rise in the college-high school wage premium…. Other CEO pay series, such as the one produced by Frydman and Saks (pdf) would show an even larger gain. These metrics indicate that CEO pay relative to that of other high earners has grown considerably….
If CEO pay growing far faster than that of other high earners is a test of the presence of rents (‘imperfections,’ in Mankiw’s terms), as Kaplan has suggested, then we would conclude that today’s executives receive substantial rents. So should Mankiw. The stark increases in CEO compensation do not simply, or even primarily, reflect an increase in their contribution to productivity.
I note that since Kaplan's data ends in 2010 the claim that "over the last twenty years… public company CEO pay relative to the top 0.1 percent has remained relatively constant or declined" refers to a comparison of 2010--when Kaplan's headline number is 2.06--to 1990--when a weighted average of the 1989 and 1993 numbers presented by Mishel would give us an estimate of 1.24, and a quadrupling of the gap does not seem to be properly framed as "has remained relatively constant or declined".
Steve would, if pressed, I believe, say two things:
The rise in the point estimate of CEO pay in 2010 relative to high earners since 1994 or 1993 or even 1990 is small relative to the enormous variation showed by CEO pay over time, and so, taking refuge in the high-ground fortress provided by the null hypothesis, it is impossible to say that the rise in CEO pay relative to high earners from 1.56 or 1.24 to 2.06 since 1993 is statistically significant.
CEO pay before 1993 all the way back to the 1930s and the rise of the CIO was inefficiently low due to the disequilibrium created by union threat--directors found it better to cap the amount of cash provided to CEOs and pay them instead in enormous amounts of soft-dollar compensation (e.g., the RJR/Nabisco auto racing team just because the then-CEO liked to play with fast cars). It was the breaking of private-sector unions during the Reagan-Bush I administrations allowed CEO pay to resume its natural level relative to other top earners by 1993, and so it is relative to 1993 (and perhaps 1929) that we should assess CEO pay.
I very much hope that the JEP editors have made Steve grapple both with the statistical-vs-economic significance issue and with the why-1993-as-the-baseline issue in his forthcoming paper…