What We Know and Do Not Know About the Natural Rate of Unemployment

The paper organizes its framework around two central ideas: first, that the labor market is characterized by large continuous flows of workers between employment, unemployment, and non-participation, which by itself implies some frictional unemployment; second, that the nature of relations between firms and workers leads to wage-setting that differs substantially from competitive wage-setting, producing a level of unemployment that differs from the efficient frictional level (Introduction, p. 2). They write their general wage relation as W/P = B·g(u, X_s), with the demand (or “labor demand”) wage set by firms as W/P = A·f(X_d); the natural rate u* is defined as the unemployment rate that equates these two wages, g(u*, X_s) = (A/B)·f(X_d) (Sec. 1.3, pp. 7-9, Figure 1).

The paper reviews the matching/bargaining approach (following Diamond 1982 and Pissarides 1990), in which wages result from bilateral Nash bargaining and the relevant labor-market variable is the exit rate from unemployment (hires divided by unemployment) rather than the unemployment rate itself; the efficiency-wage approach (Shapiro and Stiglitz 1984; Akerlof and Yellen 1990), in which firms pay above the reservation wage to limit shirking or for fairness reasons, with the exit rate again playing a similar role; and the competitive approach, under which unemployment at the low-skill end mainly reflects workers’ own reservation wages relative to poor labor market opportunities, so the relevant variable becomes the non-employment rate rather than unemployment narrowly defined (Sec. 1.2, pp. 3-6).

The paper shows that an increase in productivity (A) shifts up the demand-wage relation proportionally, and that the natural rate is unaffected only if the reservation wage (B) rises in step, shifting the supply-wage relation by the same proportion; they argue this condition “need not hold as a matter of logic, but… all appear plausible,” since unemployment benefits and other social-insurance components have “roughly increased in line with wages” over the last 50 years in the U.S., and aspiration wages – even if slow to adjust – should eventually track realized wage growth (Sec. 2.1, pp. 9-11). They note, however, that this long-run neutrality does not preclude dynamic effects: a slowdown in productivity growth, if aspirations are initially anchored to the old, higher growth rate, can raise the natural rate for a protracted transitional period (p. 11).

Using a matching-model framework with the exit rate from unemployment on the horizontal axis, the paper shows firing restrictions produce a lower-productivity, lower-flow, longer-duration “sclerotic” labor market, but argues this does not necessarily translate into a higher unemployment RATE, because both the exit rate (H/U) and the flow of hires (H) fall, leaving an ambiguous net effect on U (Sec. 2.3, pp. 12-14, Figure 2). They state plainly: “In words, higher firing restrictions lead to a sclerotic labor market, in which average productivity is lower, the flows of workers are lower and the duration of unemployment higher; but, contrary to the claims of the OECD report, they do not necessarily lead to a higher rate of unemployment.”

Estimating an augmented wage equation that nests the standard macroeconometric specification (which imposes no long-run level relation between real wages and productivity/unemployment) against the paper’s supply-wage theory (which implies such a relation should exist), the point estimate of the error-correction coefficient λ on 1970-95 annual U.S. data is actually negative, -0.11 (standard error 0.05), “suggesting the absence of any long-term level relation between the real wage and the unemployment rate” (Sec. 3, Table 1, pp. 15-16, 18). The authors note this is consistent with other researchers’ estimates, including Grubb (1986), though Turner and others (1996) find a small positive value (0.08) for the United States.

Using regressions of log state wages on lagged log wages, state unemployment rates, and state and year fixed effects for 1980-91, the paper finds that with properly measured wages – administrative weekly UI earnings, hourly manufacturing earnings, or adjusted hourly CPS earnings – the coefficient on the lagged wage is consistently close to 1 (0.91 to 0.97), implying λ close to 0 (little autoregression), which replicates Blanchflower and Oswald’s (1994) finding of “little sign of autoregression in wage equations” (Sec. 4, Table 2, pp. 19-21). However, using the March CPS annual-earnings measure that Blanchflower and Oswald themselves used, the lagged-wage coefficient falls to 0.63, and using annual earnings specifically (rather than adjusted hourly) it falls further to 0.26 – close to Blanchflower and Oswald’s own estimate – which the authors attribute to systematic measurement error from mixing earnings changes with hours changes in an annual-earnings measure, concluding that “estimates from state regressions suggest that there is… a long-run relation between wages and unemployment, but that, in contrast to the results of Blanchflower and Oswald, the adjustment to this level is slow.”

The paper reports that OECD-Europe’s unemployment rate rose from roughly 3 percent in the early 1970s to around 11 percent by the mid-1990s (Figure 3, Sec. 5, p. 23), and argues that “since European inflation in the last five years has been roughly constant, there is… little doubt that most of the rise in actual unemployment corresponds to a large increase in the natural rate of unemployment” rather than to a persistent demand shortfall, since a persistent gap between actual and natural unemployment would be expected to show up as falling inflation.

The paper reviews and finds each leading explanation incomplete on its own: oil-price increases and the productivity slowdown “can explain much of what happened in the late 1970s” but cannot explain why the natural rate kept rising through the 1980s and 1990s, especially once oil prices returned toward their early-1970s level (p. 24); cross-sectional and time-series evidence on taxes shows no strong correlation between tax rates and unemployment levels or changes within Europe, and in a recent study by Jackman and others (1996), “taxes no longer appear in the equation used to explain time series and cross country movements in unemployment” (p. 24); and on labor market rigidities, the authors note that “although Spain and Portugal are classified by the OECD as having the most stringent legal restrictions on layoffs in Europe, Spain’s unemployment rate is equal to nearly four times that of Portugal,” a cross-country pattern hard to reconcile with a rigidities-only story (p. 25).

On skill-biased technological change, the paper notes the hypothesis – that technological shifts favoring skilled workers, combined with Europe’s higher minimum wages and wage compression, raised unskilled unemployment more in Europe than in the (more wage-flexible) United States – “works well when looking at two observations, the United States and Europe,” but states plainly “it is less clear that it can explain the cross-country differences within Europe,” and that research on the topic “is just starting” (pp. 25-26). On hysteresis – the idea that a history of high unemployment raises the natural rate itself, through channels such as the long-term unemployed losing bargaining relevance, sociological shifts in the acceptability of unemployment, or the political generosity of benefit programs responding to persistent unemployment – the authors state that “the empirical case for hysteresis is far from tight,” though they cite Ball (1996) finding that OECD countries with larger or longer 1980s disinflations, especially those combining this with generous unemployment benefits, also saw larger increases in the natural rate from 1980 to 1990 (pp. 26-27).

The authors conclude that “while substantial conceptual progress has been made in thinking about the natural rate of unemployment, empirical knowledge lags behind. Economists are a long way from having a good quantitative understanding of the determinants of the natural rate, either across time or across countries” (Sec. 6, p. 27). They note that “burgeoning empirical research on rich micro-data sets has improved our understanding… of some of the mechanisms highlighted by alternative models of the natural rate,” but conclude that “a divide has grown between macroeconomists and labor economists,” with too much macro-level theorizing divorced from microeconomic evidence and too much micro-econometric work lacking a broader theoretical framework, and “end with a plea for more joint efforts by macro and labor economists to better integrate theoretical and empirical work on wage determination and unemployment” (pp. 27-28).