8/13/2019 A Time-Series Analysis of the Real Wages-Employment Relationship http://slidepdf.com/reader/full/a-time-series-analysis-of-the-real-wages-employment-relationship 1/12 A Time-Series Analysis of the Real Wages-Employment Relationship Author(s): Salih N. Neftçi Source: Journal of Political Economy, Vol. 86, No. 2, Part 1 (Apr., 1978), pp. 281-291 Published by: The University of Chicago Press Stable URL: http://www.jstor.org/stable/1830065 . Accessed: 27/11/2013 04:13 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. . The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to Journal of Political Economy. http://www.jstor.org This content downloaded from 128.243.253.108 on Wed, 27 Nov 2013 04:13:24 AM All use subject to JSTOR Terms and Conditions
12
Embed
A Time-Series Analysis of the Real Wages-Employment Relationship
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
8/13/2019 A Time-Series Analysis of the Real Wages-Employment Relationship
A Time-Series Analysis of the Real Wages-Employment RelationshipAuthor(s): Salih N. NeftçiSource: Journal of Political Economy, Vol. 86, No. 2, Part 1 (Apr., 1978), pp. 281-291Published by: The University of Chicago Press
Stable URL: http://www.jstor.org/stable/1830065 .
Accessed: 27/11/2013 04:13
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
A Time-Series Analysis of theReal Wages-Employment Relationship
Salih N. NeftyGeorgeWashingtonUniversity
In this paper, the theory of covariance-stationary stochastic processes isused in order to investigate the sign and thesignificanceofthe relationshipbetween employment and real wages. It is shown that when appropriatedistributed lags are estimated the data suggest that employment andreal wages are negatively correlated. The response appears to be non-contemporaneous and statistically significant.
The observed correlation between real wages and employment has
puzzled macroeconomists ever since Keynes (1936). Several economists,
including Kuh (1966), Bodkin (1969), and Modigliani (1977), have noted
that the contemporaneous correlation between real wages and employ-
ment is usually not statistically significant, and even when it is, is often
positive. For example, in a typical regression of real wages on a trend,
constant, and unemployment, Bodkin (1969) finds the coefficient of the
unemployment variable insignificant most of the time and negative when
it is significant.' Such findings seem to hold for most real wage, unemploy-
ment, and employment series in the U.S. and Canadian economies. Thus,
Bodkin (1969) concludes that . . . the majority of the analyses performed
with the U.S. data support the view that real wages are positively related
I have benefited from discussions with Thomas Sargent and Christopher Sims. Mycolleagues, James Barth and Robin Sickles, made several improvements in the presenta-tion. All remaining errors are my own.
' In the GeneralTheory,Keynes asserts: .. . in general an increase in employment canonly occur through the accompaniment of a decline in real wages. Thus, I am not dis-puting this vital fact which the classical economists have (rightly) asserted as indefeasible(1936, p. 17). More recently, Modigliani (1977) makes the same point in his AmericanEconomic Association presidential address: Similar tests of my own, using quarterlydata, provide striking confirmation that for the last two decades . . . the association oftrend adjusted real compensations of the private non-farm sector with ... employment ...is prevailingly positive and very significant (p. 7).
[Journal of Poli/ical lconoooy, 1978, vol. 86, no. 2, pt. 1]? 1978 by The University of Chicago. All rights reserved. 0022-3808/78/8602-0006$01.02
28I
This content downloaded from 128.243.253.108 on Wed, 27 Nov 2013 04:13:24 AMAll use subject to JSTOR Terms and Conditions
employment without placing any restrictions on the level or movement
of real wages (p. 87).
The presence of lags seems to be the main characteristic of this causal
chain of events. If so, such lagged responses can be generated by adjust-ment costs to labor input and by price inflexibilities. But the important
point is that the presence of distributed lags implies that one cannot com-
pletely characterize the relationship between two variables by using simple
regression. In the presence of distributed lags, simple regression will not,
in general, detect lagged responses and thus could give the erroneous
impression that real wages and employment are positively correlated.
In this paper it will be shown that this is indeed the case. An application
of the appropriate time-series methodology reveals that real wages andemployment are negatively related and that the puzzling positive corre-
lation reported by Bodkin (1969) is a result of ignoring the dynamics of
the underlying problem.
In the next section we describe the statistical model used in this paper
to analyze the relationship between real wages and employment. We
then report the empirical results. The estimation methodology and the
data are described in the Appendix.
Model
In order to examine the relationship between real wages and employment
we will use an explicit stochastic representation known to exist for every
multivariate linearly regular and jointly covariance-stationary time
series. Thus, let W(t) and L(t) denote the real wage and employment,
respectively. Also, assume that these series are linearly regular and jointly
covariance-stationary.3 Then W(t) and L(t) admit the following represen-tation (Sims 1972):
14(t) = E b1(s)L(t -s) + E b2(s) S(t-s) (1)S - 0S
00 00
L(t) = E a1(s)W(t -s) + E a2(s)W2(t - s) (2)S= -X S= 0
The first term on the right-hand side of equation (1) represents theprojection of W(t) on the L2 space4 generated by [L(t): -o < t < oo].
Similarly, the first term on the right-hand side of equation (2) represents
the projection of L(t) on the L2 space generated by [W(t): - o < t < coCI.The random variables el (t) and p2 (t) are jointly covariance-stationary and
3 A time series is linearly regular (nondeterministic) if the best linear forecast of itsinfinitely removed future consists only of knowledge of its mean (Rozanov 1967).
4 For a definition of L 2 spaces used in the present context, see Ash and Gardner (1975).
This content downloaded from 128.243.253.108 on Wed, 27 Nov 2013 04:13:24 AMAll use subject to JSTOR Terms and Conditions
More importantly, ?1(t) and ?2(t) are orthogonal to the corresponding
independent variables. That is to say, E[e8(s)L(t - s)] = 0 and
E[82(s) W(t - s)] = 0 for all s.
Equations (1) and (2) constitute a nonstructural but still a well-
defined system which can be used to estimate the relationship between real
wages and employment. This last point is important, since the previous
empirical work in this area has been in terms of nonstructural regressionmodels (e.g., Bodkin 1969).5 To me such uses of nonstructural models
seem to be fruitful, because the purpose of these studies is to determine the
empirical evidence which structural models of the labor market should be
able to explain. However, these nonstructural models should at least be
justifiable on statistical grounds so that the properties of the estimated
parameters and of the test statistics can be objectively evaluated. The
representation given by equations (1) and (2) is one such model, since one
can always obtain consistent and asymptotically efficient estimates of theparameters [ai(s), bi(s): - o < s < so, i = 1, 2] by using generalized
least squares, or as in this paper, a version of the Hannan efficient pro-
cedure (see Hannan 1965, 1970).
Empirical Results
In this section we first estimate the two-sided distributed lags represented
by equations (1) and (2) and then test the significance of the future lags
in order to see which one (if any) of these expressions can be collapsed to
a one-sided distributed lag. These one-sided distributed lags are then
estimated for several employment and unemployment series in the U.S.
economy.
In order to estimate the parameters in equations (1) and (2), one natu-
rally has to truncate the length of the lag distributions. Accordingly, these
expressions were estimated with 12 future and 24 past lags. This particular
truncation was expected not to affect the results in any significant way, for
in a two-sided relation such as equation (1) or (2) there exists a finitetruncation point so that the omitted tails of the lag distributions contribute
negligible explanatory power (Sims 1974).
The distributed lags in equations (1) and (2) were estimated using the
time series on the manufacturing employment, the number of employees
5 Bodkin (1969) never states explicitly whether the reported regressions are estimates
of structural relationships. Yet, one gets the impression that he is estimating demand
schedules.
This content downloaded from 128.243.253.108 on Wed, 27 Nov 2013 04:13:24 AMAll use subject to JSTOR Terms and Conditions
Test statistic for the hypothesis that{a1 Is) = 0, 0< s< - 12} is F* = 1.25
R2 = .48
D.W. = 1.96
Plot of{ b1 (s) }:**
5
Test statistic for the hypothesis that { b1 (s) = 0, 0 < < - 12} is F* = 3.26
R2 = .45
D.W. = 1.93
*Degrees of freedom: Numerator 12. denominator 128 [F12, 128(.05) 1.8].**Dotted horizontal lines represent typical standard errors.
FIG. 1.-Estimates of the lag distributions in eqq. (1) and (2)
at the payroll of manufacturing establishments, the nonagricultural em-ployment, and the layoff rate. The estimates for these variables are basi-
cally similar. Here we report the case of manufacturing employment, which
we consider to be the most representative result.
The results reported in figure 1 show that the future coefficients of the
real wages in equation (2) are insignificant. The F-statistic for the hypo-
thesis that equation (2) is one sided is equal to 1.2, which is insignificant
at the 25 percent level. This indicates that a one-sided distributed lag
relation of employment on real wages can be estimated consistently andwithout loss of any significant explanatory power. The appropriate F-
statistic for equation (1) is 3.2, which suggests that relation (1) is indeed
two sided.6
6 Some caution is warranted in interpreting this test. Although the corresponding t-and F-statistics are insignificant, the future coefficients of aI(s) seem to have some pattern;they are not randomly distributed around the horizontal axis. For the skeptic who has
reservations about the so-called exogeneity test, equation (3) was also estimated withW(t) on the right-hand side. None of the conclusions changed.
This content downloaded from 128.243.253.108 on Wed, 27 Nov 2013 04:13:24 AMAll use subject to JSTOR Terms and Conditions
These results suggest that the following equation can be used to
determine the relationship between employment and real wages:
k r
L(t) = c(s)W(t -s) + A, d(u)c(t- u), (3)s=O u0=
where the variables W(t) and e(t) have the same propertiesas in equations
(1) and (2).Since the disturbances in this expressionare orthogonal to the indepen-
dent variables, equation (3) can be consistently estimated by ordinaryleast squares (OLS), even in the presence of serial correlation. Statedanother way, expression (3) is a projection and under our assumptions,
projections can be consistently estimated by OLS (Ljung 1976). However,our purpose is not only to determine the sign of
k
E C(S)s=O
but also to see whether the relationship is significant. As is well known(Theil 1971), the appropriate test for significance requires correction forserial correlation. Accordingly, an adjustment for serial correlation was
made, as explained in the Appendix.The estimates of equation (3) using several employment and unemploy-
ment series are very similar. Here we report the results for manufacturingemployment and the unemployment rate.
To explore the adequacy of the contemporaneous relations used byBodkin (1969), equation (3) was estimated for different values of k:k = 24, 18, 12, 6, and finally for k = 0. This latter case is equivalent toBodkin's equation. By varying the length of the distributed lags, we
demonstrate how the use of the simple regression misrepresents the cor-relation between real wages and employment and therefore gives theincorrect impression that real wages and employment are positivelyrelated.
The estimates of c(s) for different values of k are shown in tables 1 and
2.7 These results are strikingly different from the ones reported in Bodkin
(1969). An analysis of tables 1 and 2 shows that the long-run relation-
ship between real wages and employment is not only significant but nega-
tive. The sumof the coefficients in the distributed lag is negative foremployment and positive for unemployment. The only significant coeffi-
cient which has the wrong sign is the contemporaneous one. This
coefficient is positive in the case of employment and negative in the case
of unemployment. As may be seen from these tables, it is this coefficient
that eventually dominates the correlation between the two variables when
7 The results with k = 24 are very similar to the ones with k = 18. Thus, they areomitted.
This content downloaded from 128.243.253.108 on Wed, 27 Nov 2013 04:13:24 AMAll use subject to JSTOR Terms and Conditions
where X1(t) and X2(t) are assumed to be jointly covariance-stationary and
linearly regular. Similarly, n(t) is assumed to be a covariance-stationary, linearly
regular, and serially uncorrelated process. It is orthogonal to the X2(t -S),
-k1 < s < k2.
To estimate {a(s)} the data were first prefiltered. To do this logarithms of raw
series were taken. Then each series was filtered through the filter (1 - .9L)
(where L denotes the lag operator).
Next, in order to purge the series of their nonstationary components, each serieswas regressed on a constant, linear trend and a set of seasonal dummy variables.
The residuals from this regression were then seasonally adjusted and used as thenondeterministic part of the series under consideration. In order to remove
seasonality the residuals were Fourier transformed and set to zero for a band of-r-/24 around the seasonal frequencies. 0 The resulting series were then inverseFourier transformed.
9 Modiglieni (1977), on the other hand, suggests that the evidence can be accotinted
for by the oligopolistic pricing model according to which price is determined by long-runminimum average cost up to a mark-up reflecting entry-preventing considerations (p. 7).
'o Following Sargent (1976) the same regressions were run using seasonally unadjusteddata which were then purged of only the deterministic seasonals (i.e., without erasing aband around the exact seasonal frequencies). Doing this, however, yielded essentiallythe same results.
This content downloaded from 128.243.253.108 on Wed, 27 Nov 2013 04:13:24 AMAll use subject to JSTOR Terms and Conditions
Baily, M. N. Wages and Employment under Uncertain Demand. Rev. Econ.Studies 41 (January 1976): 37-50.
Barro, Robert J., and Grossman, Herschel I. A General Disequilibrium Modelof Income and Employment. A.E.R. 61 (March 1971): 82-93.
. Money, Employmentand Inflation. Cambridge: Cambridge Univ. Press, 1976.Bodkin, Ronald G. Real Wages and Cyclical Variations in Employment: A
Reexamination of the Evidence. Canadian J. Econ. 2 (August 1969): 353-74.BusinessStatistics. Suppl. to Surveyof CurrentBusiness. 18th biennial ed. Washington:
Dept. Commerce, 1971.Hannan, Edward J. The Estimation of Relationships Involving Distributed
Lags. Econometrica33 (January 1965): 206-24.
. Multiple Time Series. New York: Wiley, 1970.
Keynes, John NI. The General Theory of Employment, Interest and Money. London:
Macmillan, 1936.
Kuh, E. Unemployment, Production Functions and Effective Demand. J.P.E.
74, no. 3 (June 1966): 238-49.
Ljung, L. Consistency of the Least-Squares Identification Method. IEEE Trans-actions on Automatic Control (October 1976): 779-81.
Lucas, Robert E., Jr. Capacity, Overtime and Empirical Production Functions.A.E.R. 60, no. 2 (May 1970): 23-27.
- . Understanding Business Cycles. In Stabilisation of the Domestic and theInternationalEconomy,edited by Karl Brunner and Allan H. Meltzer. Amsterdam:
North-Holland, 1977. Supplement to J. Monetary Econ. 5 (1977): 7-29.Modigliani, Franco. The Monetarist Controversy or, Should We Foresake