The Fed’s cost gauge cannot tell the difference between a raise that comes out of profits and one that comes out of production.
The Fed’s cost gauge cannot tell the difference between a raise that comes out of profits and one that comes out of production.
N ew Chairman Kevin Warsh has stopped reading the instrument the Federal Reserve has used to judge wage pressure for decades. “In tracking underlying inflation,” he said at Jackson Hole on August 28, “wage growth has not proven a reliable indicator of future inflation for a very long time.” The models underneath that judgment have not been changed, and that is an important distinction. Three of his colleagues dissented in July in favor of raising rates, and on Wednesday the Fed’s policy committee increased them unanimously — citing inflation and saying nothing about wages in either the statement or the chairman’s opening remarks.
Warsh is right to drop the wage gauge. But the reason is more specific than the usual observation that wages lag prices. The instrument that the Fed’s models use to read cost pressure cannot tell two kinds of pay raise apart, and the two have different consequences for inflation.
Here is the conventional formula: In the workhorse models, the “labor share of income,” or the fraction of output paid out in wages, stands in for the real cost of economic production. When labor’s share rises, the model reads rising cost, and rising cost means that price will follow. That shortcut has been in the literature for a quarter century and in the Fed’s briefing books for nearly as long.
Let’s start with what the gauge gets right. The labor share is measured as unit labor costs against final prices; unit labor cost is the wage divided by output per hour. So, when a pay raise is backed by rising output per worker, the labor share does not move, and the gauge correctly stays quiet, since no cost pressure needs to be recovered in prices. That is the standard account of how real wages are supposed to grow, and the conventional formula already contains it. Productivity is not the problematic signal. The problem is when pay increases are funded by something else the gauge cannot see.
A working paper published in May by Takushi Kurozumi and Willem Van Zandweghe at the Federal Reserve Bank of Cleveland shows what that something is. Inputting the actual way that firms set wages splits the labor cost in two: the true marginal cost of production, and the markdown gap between what a worker adds to output and what the worker is paid. Employers with wage-setting power pay less than the marginal worker produces — and this gap is not small. Research in the American Economic Review in 2022 put it at roughly 35 cents on the marginal dollar in the average American manufacturing plant. When this gap changes, the labor share of income also changes without affecting the marginal cost of that labor.
The marginal cost of an extra hour is higher than what a worker is paid for that hour. That’s because the next hour of work means raising the wage for every hour already on the payroll, so what it truly costs is the extra hour plus the overall raise. The gap refers precisely to that add-on: at the wage a firm actually chooses, those 35 cents are the distance between what the next hour of work is paid and what it truly costs. If hourly pay rises as the add-on shrinks, then marginal cost can stay put.
Consider what happens when jobs are plentiful: Wages and the labor share rise as the markdown gap shrinks. The wage gauge lights up. But two quite different things could have produced that reading, and the series shows the same number regardless. The increase may have come out of the markdown gap, moving pay closer to what the worker already adds to output — in which case the firm pays more and earns less per unit, marginal cost has not increased, and the gauge produced a false alarm. Or the pay increase is larger than the gap can fund, which means marginal cost has increased and the gauge is correct to light up. The instrument reports these two scenarios as the same event.
Now run the model the other direction. When employer wage-setting power deepens over a decade, and the gap widens, labor’s share falls, the wage bill per unit falls with it, and the marginal cost doesn’t move at all. A number that is loud in one direction and silent in the other is not a reliable cost measure. It’s a bargaining-power measurement wearing the costume of a cost-growth measurement.
And notice how the Fed uses the final number. When the labor share rises, it is read as cost pressure and a reason to tighten. But when it falls for 40 years, nobody argues that the Fed has been too tight for decades. The share of non-financial corporate income paid out in wages is at a low of about 55 percent against roughly 64 percent of the last three decades of the 20th century, while the profit share has gone from about 11 percent to nearly 20 percent. If that series were really the cost measure the models treat it as, that would be the largest disinflationary signal in modern American history, and it would have to be argued with rather than passed over. But the signal is passed over because everyone knows, informally, that it measures something other than cost. The models have simply not caught up.
In his Jackson Hole address, Warsh said that the Fed will “endeavor to construct more reliable models and more robust rules to guide policy decisions.” The rule needs a cost term that separates a shift in market power from a shift in cost, which the Cleveland Fed paper shows can be done within the Fed’s own apparatus. What that would take is narrower than it seems. The markdown is not a new statistic anyone has to start collecting. It’s a term the models currently hold as fixed instead of estimating it along with everything else.
The Cleveland Fed authors do exactly that, inside an otherwise standard model, and the plant-level work gives an independent read on the size of the gap in manufacturing. It is worth knowing what the current models do instead. When the inflation they estimate does not match actual inflation, the difference is absorbed by a residual that is added to the equation and left unexplained. The Cleveland Fed authors find that once the wedge is carried properly, that residual is no longer needed. A monetary rule built on the present wage-cost term is therefore reacting, in part, to a number that nobody can interpret.
The thing to watch now is whether Warsh’s inflation-frameworks task force decides that the cost term should carry that wedge, or keeps treating the labor share as if it were the marginal cost. The stakes are not academic. Every rate hike justified by a wage signal that turns out to be a markdown shrinkage will needlessly destroy jobs to correct a mismeasurement. And 35 cents on the marginal dollar is a big number to have left out of the Fed’s cost accounting.
Warsh has stopped reading the broken gauge. The next step is to replace it with one that’s accurate.