The Wage Curve in the United States

How to have your cake and eat it too

In The Making of the Mother of All Economic Booms, I explained that America’s powerful technocrats had discovered that they could have their cake and eat it too: inflation is unlikely to materialize even with tight labor markets because the Phillips curve is dead, but tight labor markets can deliver broad-based growth because the Wage Curve is alive and kicking.

We have previously established beyond a shred of doubt that inflation no longer obeys the Phillips curve because the inflation process has transformed beyond recognition by globalization. What drives US inflation is the slack in the global production system as a whole. That is why the strongest correlate of US inflation is mean inflation in the advanced economies:

The second half of the cake theorem—the Wage Curve—was put on the credit card, so to speak. We have not previously documented the fact that tight labor markets deliver higher wages. In what follows, we will present time-series and cross-sectional evidence for the claim that tighter labor markets are associated with higher wages.

Part of the systematic policy error of the Fed since inflation expectations became firmly anchored on target in the mid-1990s has been to hike in anticipation of inflation on the basis of tightening labor markets (see previous graph). The anti-labor reaction function of the Fed in the neoliberal ‘institutional supercycle’ [Gabor] relative to the postwar one, can be seen from the following graph. The dramatic contrast between the relaxed response function of the postwar Fed and the aggressive response function of the neoliberal Fed can be seen by comparing the left and right panels.

The time-variation of worker compensation is a function of productivity growth and unemployment rate.

The elasticity of wage growth is b = 0.5 (t = 10.6) with respect to productivity growth and b = -0.2 (t = -4.2) with respect to the unemployment rate. This means that, controlling for productivity growth, a 1 percent lower unemployment rate predicts a 0.2 percent faster growth in real compensation per hour.

Not only is overall wage growth faster in tighter labor markets, premia on skill, college degrees, and even race, begin to vanish late in the cycle when labor markets get tight. The Atlanta Fed (how great is their website?) allows us to track different wage spreads that contain information on the dispersion of wage growth. When labor markets began to tighten in 2016, the skill premium in wage growth narrowed.

In the previous two late-cycles, we’ve also seen the racial gap in wage growth reverse. The wages of nonwhite workers grow faster late in the cycle when labor markets tighten.

The salaries of college graduates have grown faster than the wages of high school diploma holders. Again, this college premium in wage growth vanished as labor markets tightened after 2016.

These patterns suggest that a high pressure economy with tight labor markets is a reliable way to deliver broad-based growth. This is the new diagnosis that the technocrats have arrived at since the shock of 2016 concentrated minds on the problem of broad-based growth.

---

In the mid-1990s, a number of workers led by Blanchflower and Oswald established an empirical pattern known as the Wage Curve. This is a nonlinear, negative relationship between local labor market slack and wage levels. This is contrary to standard neoclassical economic theory, which predicts a positive relationship between local wage levels and unemployment rates: unemployment is imagined to be the result of wages exceeding the market-clearing level, so higher wages should be associated with an oversupply of labor (ie, higher unemployment). As it turns out, the exact opposite relationship holds empirically.

The elasticity of wage levels against local unemployment rates is remarkably uniform across dozens of nations and regions that have been investigated. It holds steady, in fact, emerges even stronger, if we work with micro data and control for worker features—more skilled workers being more likely to be found in higher wage locations.

Here we look at data from David Dorn’s website—the infamous China shock dataset. I choose to look at this data because they report unemployment rates and household wage incomes at the commuting zone (CZ) level. This is the right level of analysis for a number of reasons. The most important reason is that people tend to stay put in the same commuting zone when searching for new jobs. They may have children in school. Their spouses may be employed in the same area. They may have elderly parents they want to live within driving distance of. They may have friends that they are loath to part with. They may not want to leave situated communities that give them a sense of belonging like their church or even the local pool bar.

So, for all sorts of reasons, leaving the commuting zone for a job is costly. Most people are rooted and tend to look for work within the commuting zone. The average American lives 18 miles away from her mother. This endows commuting zones with a degree of autarky as labor markets. Employers have to compete with other employers locally for workers. This is the underlying reason for the Wage Curve.

The next figure displays the US Wage Curve at the commuting zone level. Where the unemployment rate is low, workers have more bargaining power and wages are higher. Where the unemployment rate is high, workers have less bargaining power against local employers and wages are lower.

Since the relationship is nonlinear, we log transform both the predictor and the response to obtain the elasticity of wages against unemployment. We find a robust elasticity of b = -0.28 that is highly significant (t = -11.6). Note that we use heteroskedasticity robust (HC3) standard errors to compute t-statistics.

The estimated elasticity is much larger than the one reported in the literature (-0.1 to -0.2). The reason is that CZ level variation in wages is confounded by more skilled workers being concentrated in high wage places. College graduation rates by CZ predict wage levels, as the next chart shows.

Since we’re working at the commuting zone level and not with micro data, we cannot control for worker features. We can, however, control for human capital differences between commuting zones by adjusting for college graduation rates. The next figure displays the Wage Curve relationship at the CZ level after adjusting for college graduation rates. The elasticity estimate is tempered but remains robust, b = -0.18, and highly significant (t = -7.8). This estimate of elasticity is more in line with estimates from micro data. Along the cross-section of CZs, 1 percent higher unemployment predicts 0.18 percent lower wages after controlling for inter-CZ differentials in human capital as captured by the variation in college graduation rates.

Conditional correlations tell the same story, of course. Pearson’s partial correlation between wages and unemployment controlling for college graduation rate (all log transformed) is r = -0.27, P < 0.0001. Spearman’s rank correlation for the same is rho = -0.26, P < 0.0001.

So, wage differentials between commuting zones reflect differences in both human capital and local labor market slack. Together they explain 36 percent of the variation in wage levels in the CZ cross-section. In particular, the Wage Curve survives adjustment for our proxy of human capital. It is alive and kicking in the United States.

The evidence from the time-variation and the cross-sectional variation across commuting zones is thus consistent with the hypothesis that tight labor markets deliver faster wage growth.

However, a high-pressure economy with a tight national labor market is not enough as a solution to the problem of restoring broad-based growth and political stabilization. In order to really make progress on restoring elite-mass relations, what must also be pursued is a conscious strategy of regional depolarization. This general principle should inform the targeting of public investment: Yellen’s trillions must be purposefully dispersed across cities and states.

What would really help in achieving regional depolarization is our proposal for a standalone public ratings agency coupled with US guarantees for municipal debt. It will save cities and states hundreds of billions and enable trillions in financing. It will also be more democratic than federal investment since it will put local elected leaders in the driver’s seat. That will make public investment more responsive to local needs. Just as importantly, it will empower intermediary institutions between the working-class hinterland and the elite institutions on the coasts.

One of the overlooked problems of the neoliberal institutional supercycle, documented by Bob Putnam, is the erosion of grassroots intermediary institutions and their replacement by Astro-Turf institutions based in New York and Washington, DC. Another is the concentration of research funding, expertise and prestige in elite universities. Both of these have played their part in the erosion of elite-mass relations. In light of these patterns, the most straightforward strategy of regional depolarization is to invest heavily in public universities. This would depolarize the steep ladder of school prestige. And it would temper the concentration of skill, expertise and high incomes in superzips. Saule Omarova has brilliantly suggested that the expertise necessary for the ratings agency should be housed within or next to public universities. We support this idea.

We should also pay more attention to the general strategy of regional depolarization. A public green investment-driven high-pressure economy is not by itself sufficient to restore elite-mass relations.

---

Postscript. I should've put this chart in the original post. It was implied that unemployment rates are a nonlinear function of college graduation rates in the cross-section of 722 CZs. Unemployment is higher in the working class than in the middle and professional classes—when the college graduation rate is increased by 1 percent, the unemployment rate falls by 0.25 percent.