adjustments designed to correct for unforeseen economic events at the previous
negotiation time point. This process may be expected to generate a mix of
pro- and countercyclical wage effects through time depending on the direction
of deviations from expected outcomes. Additionally, the wage may respond
positively to the business cycle. For instance, where compensation relates to
marginal product, human capital investment may produce procyclical wages
stemming from the fixity of the labor input. The wage may also associate
positively with the product cycle. Top quality workers earning relatively high
pay may be matched with new and innovative products with strong growth
potential. As these products are eventually superseded by new innovations,
wages may subsequently be associated with the hire of relatively poor quality
and less well-remunerated workers.
Yet, all three cyclical effects will serve to condition a long time series of
the real wage. This gives rise to a series of critical questions. Which, if any,
is the frequency band dominating the cyclical behavior of the wage? If a
given frequency dominates, what direction and strength of cyclicality does it
exhibit? Does an association between cycles in a certain frequency band and
the wage pattern represent a contemporaneous association or involve leads or
lags? Pursuing such lines of enquiry leads to a more general question. Is the
observed wage cyclicality in the frequency domain supportive of the general
view arising from aggregate time series analysis or does it serve to modify
that view? A seeming low correlation between the wage and a measure of the
cycle may simply reflect the fact that the underlying time series is composed
of a number of frequency bands between the two variables that are of different
amplitudes and timing. Separately, one or more bands may display strong
evidence of a systematic cyclical relationship. Taken together, countervailing
influences may serve to mask underlying patterns.
The analysis of the wage’s spectral representation allows us to tackle di-
rectly these issues since it can be decomposed into cyclical components defined
over multiple economic cycle frequencies. The starting point is univariate anal-
ysis. A stationary time series can be broken down into superimposed harmonic
waves of varying phases and amplitudes. To determine the length of the dom-
inant cycle it is necessary to search the spectrum between the endpoints of the
entire frequency interval and select the cycle that explains the greatest portion
of the total variance of the wage. An apparent 5-year cycle may simply reflect