5 Conclusions
Frequency domain techniques establish that U.S. wage earnings are markedly
pro-cyclical. This is true for all wage components. Moreover, while the earn-
ings components are not strictly in phase with the dominant cycles, we do not
detect leads or lags that extend beyond one year’s duration. Beyond these gen-
eral conclusions, studying the frequency domain permits more detailed insights
into the cyclical behavior of earnings.
In the first place, both univariate and multivariate findings indicate that
each component measure of the wage may display significant со-variations
with more than one cycle measure in different frequency ranges. It turns out
that there is typically a dominant range in terms of explaining total wage
variation within each component. But findings of two significant ranges are
quite common and, occasionally, we find three significant ranges. For example,
it is not misleading to claim that a cycle with a 5-7 year length associates most
strongly with the cyclical movements of the real standard hourly wage. It is
misleading, however, to treat it as the only significant range association.
Secondly, the fact that earnings components respond to a range of fre-
quency ranges suggests that, in the multivariate analysis of wage cyclicality,
use should be made of more than one economic indicator of the economic cycle.
For longer cycles, Fxed capital formation offers a useful cyclical proxy to in-
vestigate. This associates relatively strongly with producer wages. For shorter
cycles, inventory investment is found to relate strongly to the hours-dominated
measure of the wage premium. In between lie the more familiar output and
employment measures and these associate particularly strongly with consumer
wages and with the proportions of overtime workers. Reliance on one repre-
sentative indicator certainly provides only a partial insight into cyclical forces
acting on wage earnings.
23