the fact that the time series possesses a dominant cycle of that length. Alter-
natively, it may disguise the fact that there are two other underlying cycles -
one longer and one shorter than 5 years - that combine to give the appearance
of a prevailing 5-year cycle. Or there may be more than two underlying cycles.
Investigating the frequency domain not only allows us to identify the number
of cycles a series possesses but also to determine the contribution of each cycle
to explaining the total variance of the wage and whether at this frequency the
explained variance is significant.
However, the main economic interest behind the study of the frequency
domain lies in an extended framework. We cannot discern whether observed
cycles in the wage reflect underlying economic conditions unless we relate them
to other variables that are both reflective of the ups and downs of economic
activity and exhibit strong associations with the wage itself. Output, employ-
ment, unemployment, fixed capital formation as well as inventory and building
investments have been variously adopted for this latter role. These economic
indicators represent a range of cycles of differing phases and amplitudes. At
one end of the spectrum, inventory investment is typically found to follow a
three- to four-year cycle (Kitchin cycle). At the other, building investment
cycles are around 20 years (Kuznetz cycle). In the middle, output often cycles
in the five- to seven-year range (business cycle) while fixed capital formation
is more likely to be in the seven- to ten-year range (Juglar cycle). Moving
from univariate to multivariate analysis in the frequency domain necessarily
involves describing associations of the wage with one or more of these economic
cycles.
In this paper, we concentrate on average hourly real earnings.2 We de-
compose real earning into three components: the standard wage, the premium
mark-up, and the proportion of overtime workers. We apply spectral methods
to each component part. This provides us with richer detail of wages over
the cycle than previous studies that have tended to compare standard hourly
wages with average hourly earnings. We show that not only do cyclical effects
differ across earnings’ components but that even a single component may co-
vary with more than one representative measure of the economic cycle. We
2See the longitudinal microdata contribution of Devereux (2001). This argues the case
for analyzing wage behavior over the cycle in this broader earnings context.