The Medium Run

OLIVIER J.
BLANCHARD
Massachusetts Institute of Technology
The Medium Run
MACROECONOMICS iS largely divided into two subfields. One focuses

on the short run, on the study of business cycles. The other focuses on
the long run, on growth and its determinants. The assumption implicit
in this division is that the medium run is primarily a period of transition
from business cycle fluctuations to growth. This simplification is clearly
convenient, but it is misleading. Modern economies are characterized
by medium-run evolutions that are quite distinct from either business
cycle fluctuations or steady-state growth.
Two facts make the point. The first is well known: unemployment
rates have steadily increased in continental Europe over the past twenty-
five years, while remaining largely stable in "Anglo-Saxon" countries.
The other is less well known but equally important: capital shares have
steadily increased in continental Europe over the past fifteen years, and
in many cases currently stand at postwar highs; in contrast, capital
shares have remained largely stable in "Anglo-Saxon" countries. The
latter fact plays a central role in the story that follows and is documented
in figure 1. The upper panel shows the behavior of capital shares in the
business sector in Germany, France, Italy, and Spain. Note that after
lows in the late 1970s and early 1980s, shares in all four countries stand
I thank Pierre-Alain Muet for previous joint work on this topic, and Dale Jorgenson
and Eric Yip for parallel research. I am grateful to Christian Frois, Pierre Olivier
Gourinchas, John Simon, and Petya Koeva for excellent research assistance and to Jean-
Philippe Spector for help with OECD data. The paper has been much influenced by
discussions with Ricardo Caballero and by his work with Mohamad Hammour on this
topic. I also thank Daron Acemoglu, Robert Chirinko, Daniel Cohen, Peter Diamond,
Pierre Fortin, James Poterba, Jaume Ventura, William Nordhaus, Julio Rotemberg, and
Robert Solow for comments. I thank the National Science Foundation for financial
support.
89
90 Brookings Papers on Economic Activity, 2:1997
Figure 1. Capital Shares, Continentaland "Anglo-Saxon"Countries, 1970-96a

Capitalincome/GDPb
Continental
Spainf
0.30-
0.40 Canada
r " / \~~/ %.
/
0.35 v Un
UnitedKingdom
. I I I I
1975 1980 1985 1990 1995
Source: Author's calculations, based on data from the Organisationfor Economic Co-operation and Development's data
diskette OECD Business Sector Data Base, 1997/1.
a. Sample periodends in 1996 for Canada,Germany,and the United States;for all othercountries,in 1995.
b. All income is ascribed to either capital or labor, so that capital income includes profits, interest paid by firms, and profit
taxes.
Olivier J. Blanchard 91
at high levels today. The average share stood at 41 percent in 1995, up

from a low of 31 percent in 1981 and a value of 34 percent in 1970.
The lower panel shows the behavior of shares in the United States,
Canada, and the United Kingdom. Note that, in contrast to continental
Europe, capital shares in those countries have remained stable. In par-
ticular, in the United States the old stylized fact of a constant capital
share still largely holds true.
This paper offers the following integrated explanation of these two
facts. The countries of continental Europe were affected by large ad-
verse shifts in "labor supply" during the 1970s. Specifically, the wage
required by workers at a given rate of unemployment and a given level
of total factor productivity increased.' The causes of these shifts, al-
though not investigated here, have been the subject of much research
by others. There is wide consensus that these shifts came from the
failure of wages to adjust to the productivity slowdown and the adverse
supply shocks of the 1970s. In any case, their initial effect was to
decrease profit rates and capital shares. Over time, firms reacted by
moving away from labor, leading to a steady increase in unemployment,
a recovery, and even an increase in capital shares.
In most Continental countries, labor supply shifts have substantially
decreased, if not vanished. But since the early 1980s, their labor mar-
kets have been characterized by adverse shifts in "labor demand."
Specifically, the real wage offered by firms at a given ratio of labor to
capital and a given level of total factor productivity has decreased.2
There are two potential explanations for this decrease. The first is a
shift in the distribution of rents from workers to firms. The second is
technological bias: at given factor prices, firms have been adopting
1. Two semantic issues arise here. First, as my definitionof a labor supply shift
indicates, I do not take "labor supply" necessarilyto mean competitivelabor supply,
but rather (and more generally), the relation between the wage and unemployment
implied by wage-setting in the labor market. Other researchershave variously called
this the wage curve, the wage-settingrelation,or the pseudo-labor-supply
curve. Second,
in order not to have labor supply and labor demandshift along a steady-stategrowth
path, I look at the wage adjustedfor total factor productivity.Thus a slowdown in
productivitygrowththat is not fully reflectedin a parallelslowdown of real wages-at
a given rateof unemployment-shows up as a laborsupply shift.
2. Again, by "labor demand" I do not necessarilymeancompetitivelabordemand,
but rather,the relationbetween the real wage and employmentthat emerges from the
employmentand pricing decisions of firms. This is sometimes called the price-setting
relationor the pseudo-labor-demand curve.
92 BrookingsPapers on EconomicActivity, 2:1997
technologies that use less labor and more capital, thus decreasing the
marginal product of labor at a given ratio of labor to capital. It is
difficult to distinguish empirically between these two explanations on
the basis of aggregate evidence; to the extent that it speaks, this evi-
dence weakly favors the second. Whatever the cause, the effect of this
adverse shift in labor demand has been to increase unemployment fur-
ther, while at the same time increasing capital shares.
By contrast, the "Anglo-Saxon" countries appear to have been
largely shielded from both the adverse labor supply shifts of the 1970s
and the labor demand shifts of the 1980s and 1990s. This accounts for
the differences from the Continental countries in the evolution of un-
employment and of capital shares.
While it is true that the macroeconomic literature has not typically
focused on medium-run evolutions, this is not the first attempt to ex-
plore such issues. Michael Bruno and Jeffrey Sachs told the first half
of the story, showing how the failure of wages to adjust to lower
productivity growth and other adverse shocks could explain the rise in
unemployment in the 1970s.3 The present paper can be seen as an
update, emphasizing the role of labor demand shifts since the mid-
1980s. One of the purposes of the project led by Jacques Dreze and
Charles Bean in the 1980s was to identify the role of capital accumu-
lation in the rise of European unemployment, a theme closely related
to that discussed here.4 More recently, Edmund Phelps has argued that
the rise in European unemployment is best understood as a "structural
slump" distinct from business cycle fluctuations.5 Finally, the present
paper is closely related to the recent work of Ricardo Caballero and
Mohamad Hammour.6 Their analysis of the effects of specificity and
labor market institutions on capital accumulation and unemployment-
in particular, their explanation of "jobless growth" in France-can be
seen as providing some of the microeconomic foundations for the shifts
that I take instead as primitives.
The paper is organized as follows. With an eye on the evolution of
capital shares, the first section documents the evolution of capital,
employment, and wage and profit rates since 1970 in fourteen member
3. Brunoand Sachs (1985).

4. Dreze and Bean (1990).
5. Phelps (1994).
6. Caballeroand Hammour(1998).
OlivierJ. Blanchard 93
countries of the Organisation for Economic Co-operation and Devel-

opment (OECD). It starts from the basic proposition that under Harrod-
neutral progress, there should be a close relation between the ratio of
the profit rate to the wage rate (measured in efficiency units) and the
ratio of labor (measured in efficiency units) to capital. It then shows
that much of the increase in the ratio of the profit rate to the wage rate-
and by implication, much of the increase in capital shares-in the
Continental countries over the past fifteen years does not reflect corre-
sponding movements in relative factor quantities. I suggest three po-
tential explanations for this divergence. The first is that there are long
lags in the adjustment of factor proportions to factor prices, and one is
still seeing the dynamic effects of the earlier adverse labor supply shifts.
The other two explanations start from the premise that the relation
between factor prices and factor quantities has genuinely shifted. The
second explanation attributes the shift to changes in the distribution of
rents from workers to firms. The third attributes it to technological
change biased against labor.
To explore the logic and the role of these potential explanations, the
second section develops a simple model of employment and capital
accumulation. Firms are assumed to be monopolistically competitive.
There are costs to adjusting capital, as well as to adjusting the ratio of
labor to capital. Labor supply is upward sloping: the wage is a decreas-
ing function of the unemployment rate. The interest rate is given and
independent of capital demand. That model makes clear how an adverse
labor supply shift leads first to a decrease and then to an increase-
above its initial level if the elasticity of substitution between labor and
capital is greater than one-in the capital share, and to a steady increase
in the unemployment rate. It also shows how adverse shifts in labor
demand, whether caused by a shift in the distribution of rents or by
technological bias against labor, lead to increases in both unemploy-
ment and the capital share.
The third section explores how well the model can explain the evo-
lution of a particular country. Relying on my comparative advantage,
I focus on the evolution of France since 1970. I construct series for
shifts in labor supply, labor demand, and the user cost. Taking those
shifts as primitives, I simulate the model. The simulations show how
the shifts can explain the evolution of the ratio of labor to capital, the
capital share, and unemployment. It would be overambitious at this
stage of the research to try to explain the evolution of each individual

country. But I perform the same exercise mechanically for each country
and give a short assessment of the results.
Having shown the logic of the argument, I return to econometrics.
In the fourth section I look for evidence of lags in the response of the
ratio of labor to capital to real wages. I find evidence of long lags. But
even allowing for such lags, there is substantial evidence of adverse
shifts in labor demand in the Continental countries since the early
1980s.
In the fifth section I try to determine whether the shifts in labor
demand reflect biased technological progress or changes in the distri-
bution of rents. The empirical strategy is simple. Shifts in the distri-
bution of rents should not affect the production function, bias in tech-
nological change should be reflected in shifts in the production function.
However, estimating production functions is tricky, and the empirical
evidence speaks only weakly. Point estimates suggest technological
bias, but they are not tight.
I conclude with a discussion of open issues, including the sources of
the shifts in labor supply and demand, the sources of differences be-
tween the experiences of Continental and "Anglo-Saxon" countries,
and the relation (if any) of the shifts discussed here to shifts in relative
labor demand between skilled and unskilled workers.
Factor Prices and Factor Quantities
Movements in the capital share, such as those documented in fig-

ure 1, are not a puzzle in and of themselves. Changes in factor propor-
tions lead to changes in factor prices (and vice versa) and-unless the
elasticity of substitution between capital and labor is equal to one-to
changes in shares. Thus a natural first step is to see whether one can
account for the evolution of factor prices-and by implication, of factor
shares-by the evolution of factor quantities.
The following benchmark is useful. Suppose that output is a constant
returns to scale function of labor and capital. Suppose that technological
progress is Harrod-neutral, a natural benchmark, as this is the assump-
tion necessary for balanced growth. Then one can write output, y, as
y = y(zn, k),
where n is labor, k is capital, and z is the level of productivity. Rede-

fining labor in efficiency units (that is, adjusting for the level of pro-
ductivity) one can rewrite the production function as
y = y(fi, k),
where fi zn.
Under the further assumption that the marginal product of labor is
equal to the wage, the following relation holds:
where aris the profit rate (that is, profit divided by the capital stock in
volume) and wv- w/z is the wage rate per efficiency unit. Furthermore,
g' is greater than zero: an increase in the ratio of labor to capital both
increases the profit rate and decreases the wage, and by implication,
increases the ratio of the profit rate to the wage. If, for example, the
production function exhibits a constant elasticity of substitution (CES),
the relation specified in equation 1 is log linear, with coefficient equal
to 1/U, where u is the elasticity of substitution between capital and
labor.
A simple exercise following from this is to examine empirically the
relation between the left- and right-hand sides of equation 1. One can
then ask: what is the implied elasticity of substitution between capital
and labor? Can an elasticity different from one explain the evolution of
capital shares in Continental countries in the 1980s? Or was there a
shift in the relation between the factor price and the factor quantity
ratios during that period?
I perform this exercise for fourteen OECD countries. For most of
the analysis below, I divide these into two groups: the first includes
Australia, Austria, Belgium, Denmark, France, Germany, Ireland, It-
aly, the Netherlands, Spain, and Sweden; the second includes Canada,
the United Kingdom, and the United States. I refer to these, with some
license, as the Continental and the Anglo-Saxon countries, respec-
tively. Assignment to one or other of these two groups is based on the
evolution of capital shares. Typically, the period covered runs from the
late 1960s to 1995, depending on data availability.7
Throughout the paper, the main data source is the OECD data set for
the business sector in each country.8 Value added is net of indirect taxes
and is allocated either as labor income or as capital income.9 Labor
income includes the imputed income of self-employed individuals,
based on the average wage in the business sector; capital income in-
cludes the residual income of the self-employed. Employment is mea-
sured as the number of workers, without adjustment for hours worked
(so that given the decline in hours per worker, employment growth
typically is overstated and productivity growth understated). I make
one modification to the data. The OECD data are adjusted for the
number of unpaid family workers, who must be deducted from total
private employment because their output is not measured. When this
adjustment does not start at the beginning of the sample, I extend it to
earlier years, assuming a ratio of unpaid family workers to total em-
ployment equal to that in the first year for which it is available. '0
7. For Germany,OECD data referto West Germanyup to 1990 and to Germanyas

a whole after 1990. Althoughit turnsout not to makemuchdifference, the econometric
work below does not use the post-1990 data. Moreover,I have excluded a numberof
OECD countriesfrom this study, for variousreasons. Some-such as Luxembourg,or
Norway at the beginning of the sample-are small or have idiosyncraticeconomic
structures.Some-especially the more recent members, such as Greece, Turkey, and
Mexico-are at a differentlevel of development.Portugalshows a permanentdecrease
in the measured capital share by 15 to 20 percent of GDP around the time of its
revolution;while this fact is fascinating, I do not know how to interpretit. A similar
problemof interpretationarises with Japan, which shows a very large permanentde-
crease in the measuredprofitrate in the 1970s.
8. Organisationfor EconomicCo-operationandDevelopmentdatadisketteBusiness
Sector Data Base, 1997/1.
9. Note that capital income here is differentfrom corporateprofits;in particular,it
includes interest paymentsby firms, as well as profit taxes. The decrease in nominal
interest rates has increased measuredcorporateprofits since the early 1980s in many
countries;see Poterba(1998) for a discussionof the evolutionof corporateprofitsin the
United States. But this decreasehas had no direct impacton the measurethat I use.
10. In a parallelstudy, Dale Jorgenson,Eric Yip, and I use the dataconstructedby
Doughertyand Jorgenson(1996) for the G7 countries, for the period 1960-89. While
the methodologyused to constructmeasuresof capitaland laboris quite differentfrom
that underlyingthe OECD data, the main conclusions reachedbelow also hold for the
G7 data set. For France, a study by Cette and Mahfouz(1996) of the statisticalissues
associated with the measurementof shares yields an evolution of the capital shares in
the business sector, in the corporatesector, and in the nonfinancialcorporatesector very
similarto that based on the OECDdata for the business sector share.
Under the assumption that labor is paid its marginal product, one
can construct the series for Harrod-neutral technological progress, z,
by constructing the Solow residual for each year, dividing it by the
contemporaneous share of labor, and integrating it over time. Then one
can construct labor in efficiency units by multiplying employment by
the constructed z, and the wage in efficiency units by dividing the real
product wage (the wage divided by the deflator for business sector GDP)
by z.
Figure 2 plots the evolution of average factor price ratios and average
factor quantity ratios for the Continental and Anglo-Saxon countries.
The averages are constructed with 1980 relative GDPs, using purchas-
ing power parity exchange rates, and each ratio is normalized to equal
1.0 in each country in 1970. This figure makes two points. In the
Continental countries, the period up to the early 1980s was character-
ized both by a decrease in the profit rate relative to the wage rate (in
efficiency units) and by a decrease in labor (in efficiency units) relative
to capital. Since that time, however, the profit rate has improved rela-
tive to the wage, while the ratio of labor to capital has continued to
decrease, albeit more slowly. This is what lies behind the increase in
the capital share. By contrast, in the Anglo-Saxon countries, the evo-
lutions of the factor price and factor quantity ratios show little or no
trend, and the movements appear to reflect business cycle fluctuations
rather than medium-run evolutions.
Next, I take a more formal econometric approach, although still in
the spirit of data description. Let pratioi, ln(wi,/i-,) be the log of the
factor price ratio and qratioi, ln(fii,/ki,)be the log of the factor quantity
ratio, where i is the country and t is time. One can then run the following
panel regression:
(2) pratioi, = y(qratioi,) + xi + x, + Ei,,
where xi and x, are coefficients on country and time dummies, respec-
tively.
The results for both Continental and Anglo-Saxon countries are re-
ported in table 1 and figure 3. Table 1 gives the estimated coefficient y
and the significance level associated with the test that all the coefficients
on the time dummies are equal to zero. Figure 3 plots the time series
of estimated coefficients on the time dummies (normalized so that the
value in 1970 is equal to zero), along with two-standard-error bands.
98 Brookings Papers on EconomicActivity, 2:1997
Figure 2. Factor Price and Factor Quantity Ratios across Three Decades,
Continentaland Anglo-SaxonCountriesa
Continental
Index,1970= 1
1.00 Profit/wage 1.00 _ Labor/capital
0.90 0.90
0.80 0.80 -
0.70 0.70 -
Anglo-Saxon
Profit/wage Labor/capital
1.05-
1.20-
1.00
1.10
0.95-
0.90
1.00
l l l l l l
1970 1980 1990 1970 1980 1990

Source:Author'scalculations(see figure 1).
a. The ratio of profitto wage correspondsto 7t/lv(the profit rate divided by the wage per efficiency unit). The ratio of labor to
capital refers to il/k (labor in efficiency units divided by the capital stock). Cross-countryaverages weight countries in proportion
to 1980 GDP, measuredat purchasingpower parityexchange rates. For the Continentalcountries, the sample period is 1970-93;
for the Anglo-Saxon countries, 1967-95.
Table 1. RegressingFactorPrices on Factor Quantities, 1961-95a

Coefficient on p value for
Panel logfactor quantity ratio time dummiesb
Continental countries: 0.96 0.00
(0.08)
Anglo-Saxon countriesd 1.12 0.03
(0.07)
Source: Author's regressions based on data from the Organisation for Economic Co-operation and Development's data
diskette OECD BuisiniessSector Data Base, 1997/1.
a. The dependent variable is the log factor price ratio (log of the ratio of the profit rate to the wage rate in efficiency units
for a given country and year). Independentvariables include the log factor quantity ratio (log of the ratio of labor in efficiency
units to the capital stock), country dummies, and annual time dummies. Samples are unbalanced panels. For some countries,
observations begin as late as 1972; for Germany, the final observation is 1990. Newey-West corrected standarderrors are
shown in parentheses.
b. Probability of obtaining these data given that time fixed effects are actually zero for all years.
c. Continental countries comprise Australia, Austria, Belgium, Denmark, France, Germany, Ireland, Italy, the Nether-
lands, Spain. and Sweden.
d. Anglo-Saxon countries comprise Canada, the United Kingdom, and the United States.
Together, table 1 and figure 3 confirm the visual impressions given by

figure 2. For both groups of countries, the estimated coefficient y is
close to one (and by implication, so is its inverse, the elasticity of
substitution in the CES case), implying little scope for movements in
the share in the absence of shifts in the relation. The time series of
estimated time dummies for the Continental countries shows a decrease
in the late 1970s and then a steady increase since the early 1980s: the
ratio of the profit rate to the wage rate is indeed higher than would be
predicted by the evolution in the ratio of labor to capital. The time
dummies for Anglo-Saxon countries show no trend.
These conclusions appear robust to a number of variations. For ex-
ample, allowing the parameter y to vary across countries yields a range
for the coefficient y; between 2. 10 (for Italy) and 0.29 (for the Neth-
erlands), but does not substantially affect the shape of the series of time
dummies. Running the regression in reverse-that is, regressing the
ratio of factor quantities on the ratio of factor prices-yields an esti-
mated elasticity of substitution of 0.75 (and thus an implied value of
1.33 for y), but does not affect the shape of the series of time dummies
substantially (up to a sign change). And dropping one country at a time
does not much change the estimated parameter; that is, no single coun-
try appears to be responsible for the results.
There are various reasons to expect the estimated parameter y to be
biased toward one. Business cycle fluctuations are associated with
Figure 3. Time Variationin Factor Price Ratios, Controllingfor Factor Quantity

Ratios, Continentaland Anglo-SaxonCountries,1963-96a
Log index, 1970 = 0
Continental
0.4
0.2
0.0
0 L* * <~~~~~~~~~00
-0.2
Anglo-Saxon
0.4
0.2 VZ<
0.0 7
L I I I I I I I
1965 1970 1975 1980 1985 1990 1995
Source:Author'sregressionsbased on data used in figure 1.

a. Figure plots, for each panel of countries,the time dummy coefficients from a regressionof the log ratioof profit to wage on
the log ratio of labor to capital, country dummies, and time dummies. Regression correspondsto equation 2 in the text and is
described in table 1, note a. For the Continentalcountries, the sample period ends in 1995; for the Anglo-Saxon countries, in
1996. Dashed lines are two-standard-error bands.
largely spurious fluctuations in measured total factor productivity, and

thus spurious fluctuations in both wages and labor, when these are
measured in efficiency units. In a boom, measured productivity in-
creases, decreasing the wage in efficiency units and increasing labor in
efficiency units; these similar proportional increases on both sides of
equation 2 are likely to bias the estimated parameter towards one. As a
rough solution to this problem, I have estimated equation 2 using values
of qratioi, lagged m,. . . m +9 times as instruments, for m = 3, 4, 5.
The instrumental variable results are largely similar to the ordinary least
squares results: y is equal to 1.03 for m = 3, to 1.12 for m = 4, and
to 1.26 form = 5."
I see four ways of interpreting these results and the apparent shift in
the relation between factor prices and factor quantities in the Continen-
tal countries:
MISSPECIFIED DYNAMICS. The first interpretation is that while there
has been a stable relation between factor prices and factor quantities, it
is not the static relation estimated in equation 2, but rather, a dynamic
relation in which firms respond to changes in factor prices only over
time. Factor proportions are largely embodied in existing capital: tech-
nology is putty clay. In other words, the relation in equation 2 is misspe-
cified, and the apparent shifts are artifacts of this misspecification.
The specific argument is the following: Firms have taken a long time
to respond to the adverse labor supply shifts of the 1970s by shifting to
technologies that use relatively less labor and relatively more capital,
to decrease their ratio of labor to capital. As they have done so, profit
rates have steadily recovered. Thus the continuing decrease in the ratio
of labor to capital in the face of an improving ratio of profits to wages
that is seen in Continental countries is simply the result of this long
drawn out dynamic adjustment.
SHIFTS IN THE DISTRIBUTION OF RENTS. If one assumes instead that
the shifts shown in figure 3 are genuine, logic dictates that they come
from one of two sources: shifts in the distribution of rents or biased
technological change. I start with the former.
One can think of shifts in the distribution of rents between workers
11. Another source of bias is measurement error in capital, which affects capital and
the profit rate-computed as profit divided by capital-in opposite directions. To the
extent that this measurement error is highly serially correlated (as is likely), the problem
will not be solved by using those instruments; indeed, it is not easy to solve.
102 BrookingsPapers on Economic Activity, 2:1997
and firms generically as changes in the relation between the marginal

product of labor and the wage. It is convenient here to define the
"markup" as the ratio of the marginal product to the real wage, and to
think in terms of changes in this markup. 12 Now suppose that at a given
factor quantity ratio, and thus a given marginal product of labor, the
real wage decreases-or equivalently, the markup increases. Clearly
this will lead to an increase in the factor price ratio (the wage goes
down, the profit rate goes up). Thus an increase in the markup can
account for the shifts in figure 3. But from where has such an increase
in the markup come? Again as a matter of logic, it may have come
either from changes in price-setting in the goods market or from changes
in wage-setting in the labor market.
Consider the goods market first. If firms take the wage as given, the
markup as I have defined it coincides with the markup of price over
marginal cost. '3 Thus the shifts in figure 3 could be interpreted as
showing that firms have steadily increased their markups in goods mar-
kets since the early 1980s. I find this explanation implausible. The
period since the early 1980s has been characterized by increased, not
decreased, competition-especially so in continental Europe, with the
reduction of barriers to trade within the European Union. Phelps has
argued that high real interest rates since the early 1980s have led firms
to care less about their customer base, and thus to go for higher markups
and higher profit margins in the short run. 14 But it is difficult to believe
that this effect, to the extent that it has been present, could have offset
the effects of steadily stronger competition in goods markets over the
past fifteen years.
It seems more plausible that if there has indeed been an increase in
the markup, it has come from the labor market. Which model of wage
determination best describes the labor market is still very much an open
question. But in a number of plausible descriptions of wage-setting, the
12. When there are costs of adjusting labor or factor proportions, as in the previous
interpretation, this ratio has to be redefined to include marginal costs of adjustment in
addition to the wage. The specifics are better left to the formal model below.
13. Some simple algebra may help here. Let pLbe the ratio of price (P) to marginal
cost (MC): P = pLMC.If firms take the wage as given, marginal cost is given by the
wage divided by the marginal product of labor (MP): MC = WIMP. Putting the two
relations together gives P = pLWIMP;or, defining the real wage as w = WIP, p. =
MPIw, which is my definition of the markup.
14. Phelps (1994).
wage need not be equal to the marginal product of labor.'5 In models

of "efficient bargaining," for example, the marginal product of labor
is set equal to the reservation wage. The wage itself is then set between
the average and the marginal products of labor, with the weights de-
pending on the relative bargaining powers of the union and the firm.
Under that interpretation, the increased markup reflects the fact that
unions have become less powerful in the Continental countries, and
that the wage has come closer to the marginal product of labor. To the
extent that unions were already weaker in the Anglo-Saxon countries,
this may explain the difference in the evolution of the two groups of
countries.
Another interpretation is based on a decline in labor hoarding, or
featherbedding practices, in the Continental countries. This story would
go as follows: In the early 1980s Continental firms were characterized
by chronic excess employment; equivalently, workers were employed
to the point where the product of the last worker was below his or her
wage. 16 As unions have become weaker, firms have been able to reduce
this excess employment, thus increasing the marginal product relative
to the wage. The markup has risen from a value below one (a marginal
product lower than the wage) to a value closer to one.
TECHNOLOGICAL BIAS. Alternatively, shifts in the relation between
factor prices and factor quantities may reflect biased technological
change; or perhaps more accurately, biased technological adoption,
given that technological knowledge is probably largely common among
OECD countries. If, since the early 1980s, Continental firms have
consistently introduced technologies that use less labor and more cap-
ital, the same ratio of labor to capital would correspond to a lower
marginal product of labor, and thus a lower wage, and, in turn, a higher
factor price ratio.
The question then becomes why the Anglo-Saxon and the Continental
countries might have adopted different technologies over time. A ten-
tative answer relies on induced bias in technological adoption, paral-
leling the argument developed in the misspecification interpretation
above. In the same way as labor supply shifts led firms, over time, to
move to technologies using less labor and more capital, they may have
15. Dickens (1995) provides a nice review.
16. For a discussion of why bargaining may produce featherbedding, see Johnson
(1990).
led firms to develop or adopt new technologies that were biased against
labor. Indeed, the distinction between movements along an isoquant
(choices among existing technologies) and shifts in the isoquant (de-
velopment and adoption of new technologies) is probably much sharper
in economists' models than in reality.
TRADE AND COMPOSITION EFFECTS. A fourth interpretation dismisses
the shifts in the relation between factor prices and factor quantities as
reflecting composition effects. A particularly strong statement along
these lines is the factor price equalization theorem. Simply put, to the
extent that trade leads to factor price equalization, one should not expect
any relation between the ratio of aggregate factor quantities and the
ratio of factor prices in a country. Thus what I have called shifts in that
relation may merely capture the effects of the world factor quantity
ratio on domestic factor price ratios-although the difference between
factor price evolutions in the Continental and in the Anglo-Saxon coun-
tries would appear to be prima facie evidence against the hypothesis of
factor price equalization.
One way to explore the relative importance of composition effects is
to look at more disaggregated evidence within each country. Under the
factor price equalization theorem, the relation between factor prices and
factor quantities should hold for each firm but disappear at the aggregate
level, because of the composition effects induced by trade. I have
started looking at sectoral evidence in France, at roughly the two-digit
(by Standard Industrial Classification) level of disaggregation, and find
that the increase in the share holds in nearly all tradable sectors. Unless
composition effects are only at work at a lower level of disaggregation,
this suggests that the factor price equalization theorem is not the main
source of the evolutions described above.
Shifts, Capital Shares, and Unemployment
The purpose of this section is to build a simple model that both

formalizes the arguments presented above and prepares the way for
a quantitative interpretation of medium-run evolutions in the next
section. 17
17. Caballero and Hammour (1998) build a richer model, with explicit putty clay
The Model
The basic structure of the model is a set of demand and supply
equations for labor and for capital. The two demand relations are de-
rived from costs of changing capital and costs of changing factor pro-
portions. Capital accumulation depends on current and expected future
marginal profits; adjustment in factor proportions, on the relation be-
tween current and expected marginal revenue products and wages. On
the supply side, the wage is an increasing function of employment,
while the interest rate is assumed independent of capital accumulation.
The specific assumptions are as follows.
The economy is composed of monopolistically competitive firms.
The reason for introducing monopolistic competition is to be able to
trace the effects of markup changes, taking these as a stand-in for shifts
in the distribution of rents in the economy.
Each firm uses one unit of capital, which it combines with variable
amounts of labor to produce output. The production function of a firm
is given by"8
(3) y = f(n, 1).
The capital stock is thus equal to the number of firms in the economy,
and changes in the capital stock correspond to the entry and exit deci-
sions of firms. A continuing firm makes only one decision at any point
in time: how much labor to employ. Note that n is both employment in
a given firm and the ratio of labor to capital for the economy as a whole.
This separation between capital accumulation decisions and factor pro-
portion decisions is inessential; but keeping these decisions sharply
distinct is helpful for the discussion of adjustment of capital and labor
below.
As noted, each firm is monopolistically competitive in the goods
market. The demand for its good is given, in inverse form, by
technology and explicit bargaining in the labor market. This allows them to relate
macroeconomic outcomes to institutional changes in the structure of bargaining, un-
employment benefit rules, and so on. In this paper, I take a number of shortcuts that
keep the model simpler but, admittedly, poorer.
18. Harrod-neutral technological progress can be introduced straightforwardly; all
that is needed is to measure labor and wages in efficiency units. For notational simplicity,
I leave it out at this stage but reintroduce it when I look at actual economies.
P= 0y ? y K 1,
ly\
where p is the price charged by the firm relative to the price level, y is
average output, and y is the inverse of the elasticity of demand. It
follows from this constant elasticity specification that the markup of
price over marginal cost charged by a firm will be equal to pL
1/(1 - y).
Each firm faces costs of adjusting its ratio of labor to capital; equiv-
alently, its employment level. Rather than explicitly allowing for a
putty clay structure of technology, I assume that each firm faces costs
of adjusting factor proportions. Specifically, I assume the cost of ad-
justinig n to be given by (c/2)(dn/dt)2, where c is a parameter.
Each firm faces a constant probability of death 8, a real interest rate
r, and a real wage (in terms of the price level) w. Under these assump-
tions, at any point in time, the firm chooses employment so as to
maximize its value, given (for time 0) by
v e ( ) LT-(2)(d)2 dt,
where
7rT p y- w n.
The first order conditions and the symmetry condition that all firms
must charge the same price, so that p = 1, are then given by
dn I
dt c
dq =
(r + 8)q -T,,
dt
= (ii)f,(n, 1) -w.
Firms adjust the ratio of labor to capital in response to the present

value of marginal profit, denoted by q. Marginal profit is equal to the
marginal revenue product of labor (which is itself equal to the marginal
product multiplied by the inverse of the markup) minus the wage. In
the long run, the marginal product of labor must be equal to the real
wage times the miarkup(equivalently, the marginal revenue product of
labor must be equal to the real wage). Denoting steady-state values by
a star,
(4) f,,(n*, 1) = L we.
Note that from the point of view of determining employment, a higher

markup acts like a tax on the marginal product of labor (this is a tax
collected by the firm, however, so that its effect on profit is quite
different). A higher markup induces firms to choose a lower level of
employment, and thus leads to a lower ratio of labor to capital.
As mentioned, the evolution of the stock of capital comes from the
entry and exit of firms. To capture the slow adjustment of capital, I
assume costs of adjustment for capital: the relative price of capital is
an increasing function of the net rate of entry (equivalently, the net
change in the capital stock). Specifically,
dK
(5) Pk 1 + hd
where h is a parameter and K denotes the capital stock. Free entry

implies that the following condition must hold:
V Pk,
where v is the value of the firm, defined above. If firms could freely
choose their initial factor proportions, the model would yield a distri-
bution of factor proportions across firms, with the proportion depending
on time of entry. To avoid such heterogeneity, I assume that new firms
enter with the same ratio of labor to capital as existing firms. This keeps
the model tractable; but it also eliminates the entry and exit of firms as
a candidate channel for change in aggregate factor proportions over
time. 19
The value of a new firm must be equal to the price of the machine
needed to produce its goods. From the definition of v above, v is char-
acterized by
19. One of the contributions of Caballero and Hammour's (1998) model is that it
keeps track of the distribution of firms and its implications, for example, for wage
bargaining.
dv = (r + 8)v - [ - (c) ()2]
Entry takes place (equivalently, the capital stock increases) when the
value of an existing firm is greater than one. In steady state, dvldt =
dnldt = dK/dt = 0, so that the previous equations imply that
(6) w = * (r + 8) = (r + 8).
Profit per unit of capital is equal to the user cost.

This ends the description of the dynamic demands for capital and
labor. The aggregate demand for labor is given by N = nK, the ratio
of labor to capital in each firm times the number of firms.
I specify the supply of factors as follows. I assume the real wage to
be a constant elasticity function of the ratio of employment to the labor
force, NIN,
(7) w =0 (0 )
where ,Bis the elasticity of the wage with respect to employment and 0
is a multiplicative constant. And I assume r to be exogenous. This is a
strong assumption. In combination with equation 6, it implies that the
long-run supply curve of capital is infinitely elastic and the profit rate
always returns to the same value: r + 6.
Functional Forms and Parameters

I choose functional forms and parameters as follows. The unit time
period is one year. The production function is CES, of the form
_ _ cr
(8) y = A (1 - a)n + aJ_-

(Capital is implicitly present in the production function, but since each
firm uses one unit of capital, it is equal to one.)
The coefficient multiplying capital (a) is 0.3 and the multiplicative
constant (A) is 0.5. In the long run, the response of the capital share to
an increase in wages depends on the elasticity a. The evidence below
points to a value for a- close to 1.0. Therefore I use 1.0 as the benchmark
value, but to show how wage increases can potentially lead to an in-
crease in the capital share, I also examine the case where a is equal to
2.0.
The probability of death for firms (8)-equivalently, the depreciation
rate-is equal to 0.1 and the real interest rate (r) is equal to 0.05. The
initial value of y is 0, corresponding to the case of perfect competition;
this implies a value for the markup ([t) of 1.0.
The value for c is equal to 4.0. In a world in which production was
strictly putty clay, only the newly installed capital stock-roughly
10 percent of the total each year-would embody the new desired factor
proportions. This would imply a mean lag of adjustment of 4.5 years.
In the present case, my chosen value of c implies that firms each year
close roughly 17 percent of the gap between desired and actual factor
proportions. This, in turn, implies a mean lag of 4.8 years.
The value for h is equal to 10.0. This implies an elasticity of invest-
ment with respect to the relative price of capital (pk) of 1.0. Empirical
evidence on the relation of investment to Tobin's q yields lower elas-
ticities, and thus higher implied values for h. But as discussed in that
literature, these estimates of h are likely to be biased upward. The
instrumental variable approach used by Jason Cummins, Kevin Hassett,
and Glenn Hubbard yields an elasticity of about 0.7.20
I normalize the labor force to be equal to 1.0. I choose 0 equal to
0.35, which implies zero unemployment in the initial steady state. The
elasticity of the wage with respect to employment ,Bis equal to 1.0. For
an average unemployment rate of 10 percent, this corresponds to an
elasticity of roughly 0. 1, which is close to the estimates of David
Blanchflower and Andrew Oswald for a number of countries. My work
with Lawrence Katz suggests that the elasticity is, in fact, lower in the
short run and higher in the long run; I ignore those dynamics here.22
These parameters and their implications for steady-state values of
output and other variables are presented in table 2.
Shifts in Labor Supply, Changes in the Distribution of Rents, and

Technological Bias
With the discussion of the previous section in mind, I consider the
effects of three different types of shift.
20. Cummins, Hassett, and Hubbard (1994).
21. Blanchflower and Oswald (1994).
22. Blanchard and Katz (1997).
Table 2. Parameters and Steady-State Values

Parameters
(r Elasticity of substitution 1.0 or 2.0
a Coefficient on capital in production function 0.3
A Multiplicative constant in production function 0.5
8 Depreciation rate 0.1
pLGross markup 1.0
r Interest rate 0.05
c Cost of adjusting ratio of labor to capital 4.0
h Cost of adjusting capital stock 10.0
c Elasticity of wage with respect to unemployment 1.0
0 Wage at zero unemployment 0.35
Steady-state values
Output 0.50
Employment 1.00
Capital 1.00
Wage rate 0.35
Profit rate 0.15
Capital share 0.30
Unemployment 0.00
ADVERSE SHIFTS IN LABOR SUPPLY. Figure 4 shows the effects of an

unexpected, permanent, adverse shift in labor supply: a 10 percent
increase in 0 in equation 7. Put less formally, the figure shows the
effects of an adverse wage push. Throughout, the model is solved under
the assumption of rational expectations. Figure 4 plots the evolution of
the profit rate, the wage rate, the ratio of the profit rate to the wage
rate, the ratio of labor to capital, the capital share, and the unemploy-
ment rate, for two values of cT: 1.0 and 2.0.
At the initial ratio of labor to capital, the increase in wages leads to
a corresponding decrease in the profit rate, as well as a decrease in the
capital share. This triggers two dynamic responses. First, firms shift
away from labor over time, leading to a decrease in the ratio of labor
to capital. This, in turn, leads to a partial recovery of the profit rate
and, if the elasticity of substitution is greater than one, to a more than
full recovery of the capital share. Second, lower profit leads, over time,
to a net exit of firms and a decrease in the capital stock.
Both lower capital and a lower ratio of labor to capital lead to lower
employment, higher unemployment, and a decrease in the wage. The
effect of the initial wage push on the wage is steadily offset by the
effect of higher unemployment. In the long run, the profit rate must
Figure4. DynamicEffects of an Adverse Labor Supply Shifta
Profit rate Wage rate

0.15
/- - ~~~0.37-
0.14
0.13
6 1
-:=1 0.35
0.122
0.40 / 1.00
0.98-
0.36
-
0.32 ~~~~~~~0.96
Capitalshare Unemploymentrate
0.30 /--0.08- -
0.28 0.06
0.26 0.04
0.24 0.02
4 8 12 16 20 24 4 8 12 16 20 24
Time(quarters) Time(quarters)
Source:Author'scalculations,as describedin text.
a. Graphsshow simulateddynamic effects of a 10 percent increase in the parameter0; the increase occurs in quarter5. Solid
line results when elasticity of substitutionis set to one; dashed line, when elasticity is set to two. All other parameterstake the
values given in table 2.
return to its initial value. This implies that the wage must also return
to its initial value. Given the unit elasticity of the wage with respect to
employment, both employment and capital decrease by 10 percent. The
capital share returns to its initial value, and the unemployment rate
increases to 10 percent.
In summary, adverse labor supply shifts can generate both an in-
crease in unemployment over time and the kind of movement in the
capital share that has been observed in the Continental countries-
namely, an initial decrease followed, in the medium run, by a more
than full recovery. But to the extent that unemployment puts pressure
on the wage to return to its initial value, the medium-run increase in
the capital share is quite small. In the simulation corresponding to a =
2.0, the capital share never rises much above its initial level.
INCREASE IN THE MARKUP. Figure 5 shows the effects of an unex-
pected, permanent, increase in the markup: a 10 percentage point in-
crease in w.23 Recall that the markup acts like a tax on labor. Thus in
response to an increase in the markup, firms decrease employment over
time. This leads to an increase in unemployment, and higher unem-
ployment, in turn, leads to a decrease in the wage. As a result of both
of these effects, the capital share increases. Thus in the medium run,
the markup shift leads to both an increase in the capital share and an
increase in unemployment.
The decrease in the wage, in turn, leads to an increase in the profit
rate. Thus a second mechanism comes into play: the entry of firms in
response to the higher profit rate. Unemployment, after its initial in-
crease, starts to fall; the rise in the number of firms dominates the
decrease in the ratio of labor to capital in each individual firm.
In the long run, the implication of free entry and a given interest rate
is that the profit rate must return to its original value. Thus to a first
approximation, the wage also must return to its initial value, as must
unemployment; the effect on employment of a lower ratio of labor to
capital in each firm is offset by a larger number of firms, a larger capital
stock.24 The capital share, however, remains permanently higher, as
the ratio of labor to capital is lower at any given wage rate.
23. Rotembergand Woodford(1991, 1998) draw attentionto the role that changes
in markupsplay in business cycle fluctuations.This paperdrawsattentionalso to their
potentialmedium-runimplications.
24. "To a firstapproximation"meansthat the resultholds for small changes in the
Figure 5. DynamicEffects of an Increasein the Markupa
Profitrate Wagerate
0.165- -- 0?35?h0
0.160 - Jr 0.345
0.155 - 0.340 -
0.150 0.335 -
=2x
Profit/wage,, - Labor/capital
1.0
0.47-
0.49i -< . ..
0.45-
0.43 0.
Capitalshare ,-
0.38 -
0.36 -
o
0.02
0.32-
0.00
0.30 Unemploymentrate
4 8 12 16 20 24 4 8 12 16 20 24
a. Graphsshow simulated dynamic effects of a 10 percent increase in the markup(the parameterp); the increase occurs in
quarter5. Solid line results when elasticity of substitutionis set to one; dashed line, when elasticity is set to two. All other
parameterstake the values given in table 2.
While the markup in the model comes from monopoly power in the
goods market, any change in the distribution of rents that leads to an
increase in iL will have similar effects. Consider, for example, an in-
crease in V. coming from a decrease in featherbedding practices. The
dynamics will be identical to those shown in figure 5.25 At a given
wage, firms want to reduce employment, and they do so over time. This
leads to an increase in unemployment, but also a higher profit rate and
a higher capital share. Over time, the increase in the profit rate leads
to entry of firms and capital accumulation, and consequently unem-
ployment decreases until it has returned roughly to its initial level. By
this time, the profit rate has returned to its normal level, but the capital
share remains high.
In summary, in the short and medium runs increases in the markup
lead to an increase in unemployment and an increase in the capital
share. Unemployment does eventually return to its initial value but, as
the simulation shows, this takes a very long time. Thus upward markup
shifts-or, more generally, shifts in the distribution of rents from work-
ers to firms-appear potentially able to explain the evolutions in the
Continental countries since the early 1980s.
BIASED TECHNOLOGICAL CHANGE. I formalize the effects of biased
technological change as an increase in the coefficient on capital (a) in
the production function given by equation 8. In the Cobb-Douglas case,
this takes the simple form of an increase in the exponent for capital and
a corresponding decrease in the exponent for labor. This formalization
captures the idea that new technologies save on labor, and has the
advantage that the bias is well defined even in the Cobb-Douglas case.26
Figure 6 shows the effects of a 10 percentage point increase in the
markup,startingfroman initialvalue of 1.0 (thatis, thereareno distortionsat the start).

For largerchanges, or if one starts from a value of ,u greaterthan one, the wage rate
ends up a bit lower and the unemploymentratea bit higher.
25. The welfare implicationswill be different, however. An increasein the markup
coming from increasedmonopolypower(thatis, an increasein ,uabove one) represents
an increase in distortions.The eliminationof featherbedding(that is, an increase in p.
towardone from a lower value) representsa decreasein distortions.
26. A while back, Houthakker(1956) showed how one could think of the Cobb-
Douglas productionfunctionas the resultof aggregationof Leontief functions, with the
coefficientsof the underlyingfunctionsjointly Pareto-distributed.
The samejustification
applies in the presentcontext. Firms introducenew technologies that increasethe pro-
portionof relativelycapital-intensivemethodsof production,and thus the value of the
coefficient on capital in the aggregateproductionfunctionis increased.
Figure 6. Dynamic Effects of Biased TechnologicalChangea
Profitrate ," -- Wagerate
0.170 F 1 0.35
0.16 0.34 -
0.15 03
0.55 ---- 0.90
0.50 ~~~~ -
~~~~~0.80
0.45 0.70
Capitalshare
0.45 0.06
0.40 - - 0.04 - -
0.35 - 0.02 -
0.30 0.00
Unemploymentrate
4 8 12 16 20 24 4 8 12 16 20 24

a. Graphsshow simulateddynamic effects of an increase in the coefficient on capital in the productionfunction:a is increased
from 0.3 to 0.4 in quarter5. Solid line results when elasticity of substitutionis set to one; dashed line, when elasticity is set to
two. All otherparameterstake the values given in table 2.
coefficient a, from 0.3 to 0.4.27 It is striking how much figure 6 looks

like figure 5-the effects of technological bias are nearly identical to
those of an increase in the markup. At the initial ratio of labor to capital,
the change leads to an increase in the profit rate and an adverse shift in
labor demand. This, in turn, triggers the same dynamics as an increase
in the markup, resulting in increases in both unemployment and the
capital share.
Thus this last simulation offers two lessons. First, technological bias
provides another potential explanation for the evolutions in the Conti-
nental countries since the early 1980s. And second, it may be difficult
to distinguish these effects from those of markup shifts, unless one
looks directly at the production function, which is affected by techno-
logical bias but not by changes in the distribution of rents.
Looking at Evolutions in Specific Countries
The next step is to see whether and how the model can explain
observed evolutions in specific countries. It would be overambitious
and space-consuming to analyze each country individually in this paper.
Therefore I present the results for one country-France-and then sum-
marize the results for the others.
Constructing the Shifts

I first construct empirical series for labor supply shifts, labor demand
shifts, and shifts in user cost (which are present in the sample, although
I have not focused on them so far) with which to simulate the model.
SHIFTS IN LABOR SUPPLY. I start from equation 7. Allowing for
Harrod-neutral technological progress, taking logarithms, and approx-
imating the difference between the log of the labor force and the log of
actual employment by minus the unemployment rate, that equation
becomes
27. In changing relative marginalproducts, one must be careful not to introduce

productivity-leveleffects. In general, if the ratio of labor to capital is different from
1.0, the change in coefficientswill changeproductivity,unless offset by a change in the
multiplicativeconstant. This issue does not arise here, as in the initial steady state the
ratioof laborto capital is equal to 1.0.
(9) ln w = ln 0 - u,
where u is the unemployment rate and w-is, again, the wage in efficiency
units. I first construct the series for the wage in efficiency units.28 I
then construct the series for (log) labor supply shifts as ln 0 -lnii +
3u. Finally, I normalize this series to equal zero in 1970.
By constructing labor supply shifts in this way, I do not mean to
imply that the "true" labor supply relation has the form of equation 9.
As mentioned above, the true labor supply or wage relation has richer
dynamics-from overlapping wage-setting to hysteresis-and includes
many other variables. Thus the labor supply shifts that I construct are
combinations of these dynamic effects and movements in these other
variables. The best way to think about equation 9 and these labor supply
shifts is as giving the distance of the wage from that which, in the
absence of other shifts, would allow the economy to return to its 1970
unemployment rate.
Figure 7 shows labor supply shifts in France for three values of 3:
1.0 (the value that I use above and again in the simulations below),
0.5, and 1.5. All three series show a large increase in the early 1970s,
with the wage increasing much faster than measured total factor pro-
ductivity; a peak at around 15 percent (for the intermediate case) in the
early 1980s; and a subsequent decline. In 1996, the value of labor
supply shifts stands between 1 and 10 percent, depending on the value
of P. Put another way, French wages in efficiency units are lower today
than they were in 1970; but they would be too high if unemployment
decreased, subjecting them to upward pressure. How high is "too high"
depends on the assumed value of ,3, the effect of unemployment on the
wage.
Relating these labor supply shifts to specific changes in the economic
environment and in labor market institutions must wait for another
paper. But based on the large amount of research on European unem-
ployment, I do not think that there is any great mystery about what lies
28. RobertHall has shown that if the markupis differentfromone, the computation
of total factorproductivitygrowthmust be modifiedto take accountof the effect of the
markupon the shares; see, for example, Hall (1990). The results reportedhere are
derivedunderthe assumptionthatthe averagemarkupduringthe periodis equalto 1.0.
I have carriedout the same exercise underthe assumptionthat the average markupis
equal to 1.2, with very similarresults.
/
(tolt/~~~ 2- '9
00
- -~~~~~~~~~~~~~~~~E7
1 I
I- '-
- /~~~~~~~~~~~~~~~~~~~~~~C
/,momQo O 00
; o2/
iii
oY, o o o
behind them.29 The slowdown in productivity growth starting in the

mid-1970s and the increase in oil prices were not instantaneously re-
flected in lower- real wage growth, as a result of which the wage in-
creased relative to total factor productivity during the 1970s. In re-
sponse to higher unemployment, labor market institutions were changed
so as to protect employed workers against the risk of unemployment
and improve the welfare of the unemployed. Those changes lessened
the effect of unemployment on wages. Since then, oil prices have re-
turned to their 1970s levels and workers have presumably adjusted their
expectations to the lower rate of productivity growth. And indeed, this
is reflected in the decline in labor supply shifts since the early 1980s.
As to why the shifts have not been fully undone, hysteresis is probably
relevant. Labor market institutions are not the same as those in 1970.
In France the treatment of unemployment, for example, is now consid-
erably more generous. Also, long-term unemployment has marginalized
many workers, reducing the weight of the long-term unemployed in
wage determination.
SHIFTS IN LABOR DEMAND. I take labor demand shifts to be shifts in
the markup. As I make clear below, if one were to interpret these as
the result of technological bias the constructed series would be identi-
cal, and the simulation results would be largely similar (as shown
above).
I define the markup above as the ratio of the marginal product to the
wage, ignoring the costs of adjusting the ratio of labor to capital.
Assuming that the production function is CES with elasticity of substi-
tution cr, this relation takes the form
(10) Inwi + In F = constant + ln(1 - a) - ( )ln(n).
The right-hand side gives the logarithm of the marginal product of

labor. The left-hand side gives the logarithm of the real wage times the
markup. Given a value for cr, and under the maintained assumption that
the parameter a is constant, one can construct a series for the log markup
(up to a constant term) using equation 10.30
29. See, for example, Nickell (1997) for a recentassessment.

30. Underthe additionalassumptionu = 1.0, the changein the log markupis equal
to minusthe change in the log of the shareof labor-a convenientfact for back-of-the-
This equation is only correct, however, in the absence of costs of

adjusting factor proportions. If it is costly for firms to adjust those
proportions, an increase in the wage will be associated with little con-
temporaneous change in nlk, and thus little change in fily. This, in turn,
will lead to a decrease in the measured markup. I show below (in the
upper panel of figure 8, with the series corresponding to A = 0.0) that
this is indeed what happens in the case of France: absent a correction
for costs of adjustment, markup shifts turn out to be highly negatively
correlated with the labor supply shifts constructed above.
To take account of the dependence of factor proportions on the his-
tory of wages as well as on the current wage, I replace the current wage
(w) by a weighted average of current and lagged wages (17pa). This is a
simpler but rougher approach than trying to account for both past and
expected future values of the wage, as implied by the model of the
previous section.3' Specifically, I construct the log markup (up to a
constant) as
where
ln 0i, = A ln + (1 - X) ln i,
and A is a parameter reflecting the speed of adjustment of factor pro-

portions. In line with the parameters chosen for the model above and
the empirical evidence presented below, I choose a value of cr equal to
1.0 and a value of A equal to 0.8, implying a mean lag of four years in
the adjustment of factor proportions to the wage. Figure 8 plots labor
demand shift series for these values of cr and A, and for two sets of
alternative values of these parameters.
The three series presented in the upper panel of figure 8 correspond
to different values for X: 0.0 (corresponding to no costs of adjustment),
0.8 (the benchmark), and 0.9; in each case, cris set equal to 1.0. Under
the assumption of zero costs of adjustment, the markup series shows
envelope computations. For example, a decrease in the labor share from 0.7 to 0.6
implies an increasein the markupof roughly 15 percent.
31. For a detailed discussion of this and other issues in the constructionof the
markup,see Rotembergand Woodford(1998).
Figure 8. Labor Demand Shifts under AlternativeElasticityand AdjustmentCost

Assumptions,France, 1968-95a
Log markupindex, 1970= Oa
AlternativeadjustmentCoStSb
0.10_
0.50 -
0.00
-0.50
Source: Author's calculations (see figure 1).

a. Figureplots the log of the markup(In 1), normalizedto equal zero in 1970. Using equation 11 in the text, In 1 is constructed
as -In vA- (1/l) In (tiy),v where is the elasticity of substitution,
cv,,n is effective labor,and y is output.The series In in tum, is
constructed as x In wv, , + (I - X) In wv,, where wv,is the wage in efficiency units and x is a parameter capturing the speed of
adjustment of factor proportions.
b. In each case, = ua1.
c. In each case, x = 0.8.
large negative values beginning in the mid- 1970s and then turns positive
in the mid- 1980s. But the other two graphs show that the initial decrease
is largely spurious, coming from the slow adjustment of firms away
from labor in the face of the wage push. For X equal to 0.8, the markup
shows a small decline in the late 1970s and a large increase since the
late 1980s. For A equal to 0.9, the increase only starts in 1990. Note
that in all three cases, the value of the markup shift is large and positive
at the end of the period; the value of A affects only the timing of the
increase.
The two graphs in the lower panel of figure 8 correspond to different
values of cr: 1.0 (the benchmark) and 2.0. In both cases, A is set equal
to 0.8. The higher value of cr yields a more pronounced decline in the
1980s and a smaller value of the shift at the end of the period.
The series for the markup shifts is derived under the assumption that
the production function is time invariant (up to Harrod-neutral techno-
logical progress); in particular, ln(1 - a) is constant in equation 10.
But these shifts could equally have been called technological bias shifts,
corresponding to changes in the coefficient a, with ji remaining con-
stant. Equation 10 makes clear that if one looks only at labor demand,
changes in ln ji and changes in - ln(1 - a) are observationally equiv-
alent. Thus one could equally interpret figure 8 as showing technolog-
ical bias in favor of capital-an increase in a-since the mid-1980s. I
discuss below whether and how one can use other evidence to distin-
guish between the two explanations.
SHIFTS IN USER COST. To construct the time series for shifts in the
cost of capital (r + 8), I use the depreciation ("scrapping") rate from
OECD data for 8. I construct r in three different ways. Figure 9 presents
the resulting user cost series for France.
In the series that I use for simulations below, I construct r as equal
to the long nominal interest rate minus the average rate of inflation over
the previous five years. This is labeled "bonds, method 1" in figure 9.
The second series uses the nominal interest rate minus the inflation rate
over the previous year, and is labeled "bonds, method 2." The third
series, "bonds and equity," constructs the required rate of return as a
weighted average of the real interest rate on bonds-which, in turn, is
constructed as the nominal interest rate minus a five-year average of
inflation-and of the required rate of return on equity-which itself is
constructed as the sum of the ratio of dividends to prices plus a five-
ol~~~~~~~~~~~~~~~~.
Eu
Cu
- -~~~~~~~~~~~~~~~~4
~~~~~~~~~~~~~~~~~~~~~~~-1
00 ~~~0&
001
Cu~~~~~~~~~~~~~~~~~~~~~~~~~
Cu ~~~~~~
G.0~~~~~~~~~~~
o~~~~~~~~~~~~~~~~~~~~~~~~~~~-
0 2
Cu~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
~ ~~~~~~~~~~~~~~~~,0
;z~~~~~~~~~~~~~~~~~~~~~~~*
00~
year average of past output growth (an admittedly rough proxy for the
expected rate of growth of dividends). The weights are 0.7 on bond
finance and 0.3 on equity finance.
All three series show a low user cost in the 1970s, a peak associated
with disinflation in the early 1980s, and another peak in the early 1990s,
associated with German reunification and the "Franc fort" policy. The
user cost using one year of lagged inflation rather than a five-year
average is higher during most of the 1980s and a little higher at the end
of the series. The user cost assuming bond and equity finance shows
little trend and finishes lower than the other two series. This is because
the steady decrease in the ratio of dividends to prices and in the growth
rate over the past fifteen years imply a steady decrease in the estimated
required rate of return on equity over that period.32
Simulating the Model

Figure 10 plots, for France, the evolution over 1970-96 of the profit
rate, the wage rate per efficiency unit, the ratio of the profit rate to the
wage rate (per efficiency unit), the ratio of labor (again in efficiency
units) to capital, the capital share, and the unemployment rate. These
are the facts to be explained. For ease of comparison, I normalize each
variable to have a 1970 value equal to that in the steady state of the
model. Thus in 1970, the profit rate is normalized to 0.15, the wage
rate to 0.35, the ratio of profit rate to wage rate to 0.428, the ratio of
labor to capital to 1.0, the capital share to 0.30, and the unemployment
rate to 0.0. The basic evolutions are by now familiar; in particular, the
increases in the capital share and in unemployment.
The simulations described below assume zero values for the shifts
before 1970 and actual values thereafter. They are run under rational
expectations, with expectations of future shifts equal to their current
values. The corresponding figures show simulation results under two
alternative assumptions about the value of a: 1.0 and 2.0.
Figure 11 presents the results of a first simulation, allowing for shifts
in labor supply and the cost of capital but ignoring shifts in labor
32. I have so far ignoredthe trenddecline in the relative price of capital, which is
absentfromthe modelabove but is empiricallyrelevantfor a numberof OECDcountries.
In France, however, this factor is not very important.The relative price of investment
goods has decreasedby only 6 percentsince 1970.
Figure 10. HistoricalPaths of Variables,France, 1970-95

Indexa
Profitrate Wagerate
0.15 03
0.14-
0.36
0.13-
0.12 0.3
0.45 - 0.95
0.40 0.90
0.35 -0.85 -
0.30 0.80
0.36 -0.08 -
0.06-
0.32-
0.04-
-
0.28 - 0.02
1975 1980 1985 1990 1975 1980 1985 1990

a. Each variableis normalizedso that its 1970 value correspondsto the steady-statevalue assumed in the model; see table 2 for
parameterand steady-statevalues.
Figure 11. SimulatedEffects of HistoricalLabor Supply and User Cost Shifts,

France, 1970-96a
Profitrate Wagerate
0.18 - 0.37
0.16 -0.36
0.14 0.35
0.12 --=10.34'v
Profit/wage lo ratal
0.50
0.40 09
0.85-
0.30
0.35 0.10
0.30 /0.05 '
0.25 0.00
1975 1980 1985 1990 1995 1975 1980 1985 1990 1995
a. Graphsshow the paths that variablesfollow in response to the constructedlabor supply and user cost shifts that occurredin
France over the sample period. The constructionof the time series of these shifts is described in the text and in the notes to
figures 7 and 9; the user cost series is constructedusing long bond rates deflatedby the inflationrate over the previous five years.
Solid line results when elasticity of substitution(a) is set to one; dashed line, when elasticity is set to two. All other parameters
take the values given in table 2.
demand (markup or technological bias shifts). It is clear the adverse

labor supply shocks can explain the increases in unemployment and the
capital share up to the mid-1980s. But they are unable to explain the
further increases in unemployment and the capital share since the mid-
1980s. Instead, as labor supply shifts decrease in magnitude and factor
proportions adjust, the ratio of labor to capital turns around, leading to
reductions in unemployment and the capital share.
Figure 12 depicts the effects of all three shifts combined. It shows
how the adverse labor demand shifts since the early 1980s help to
account for the subsequent evolution of both unemployment and the
capital share. In contrast with the previous simulation, the ratio of labor
to capital declines throughout the period (as it does in the data), despite
the fact that due to the decrease in labor supply shifts and the downward
pressure from unemployment, the wage (in efficiency units) is below
its initial value. Unemployment remains high, and so does the capital
share.33
A Glance at Other Countries

For the other thirteen countries in my sample, I have carried out the
same exercise mechanically, without any attempt to fine tune, even
when the simulations results are poor. Fine tuning would take me be-
yond the scope of this paper, and so I leave it to later work. Table 3
summarizes the results.34 For each country, the table reports the change
in constructed labor supply and labor demand shifts for the subperiods
1970-81 and 1981-95. And for each subperiod and country, it also
gives both the change in the unemployment rate implied by the simu-
lation and the actual change. To interpret the numbers for labor supply
33. The fit between the actualevolutions in figure 10 and the simulatedevolutions
in figure 12 is clearly very good. This, however, is largelyby construction.Recall that
the series for the laborsupply and demandshifts are constructedso as to makethe labor
supplyandthe labordemandrelationsfit exactly. If the model hadno internaldynamics,
the overall fit would be perfect. To the extent that the model determinesthe dynamics
of capital accumulationand factorproportionsendogenously,the fit can still turnout to
be poor. Table3 below, which reportsthe simulationresultsfor each of the otherthirteen
countries in the sample, shows that predictedunemploymentcan differ substantially
from actualunemployment.
34. I was unableto solve the model for Spain and the United Kingdom,apparently
because of large negative real interestrates in the mid-1970s. The simulationsreported
for these countriesassume a constantuser cost.
Figure 12. SimulatedEffects of HistoricalLabor Supply and Demand and User Cost
Shifts, France, 1970-96a
Profitrate Wagerate
0.16 03
0.36-
0.14
0.35
0.12 - 103
cT2
0.45 0.96
0.40 0.92 7/
0.35 /0.88
0.35 -- 0.10
0.30 0.05 -
l~ ~~~
l I
0.25 -0.00
1975 1980 1985 1990 1995 1975 1980 1985 1990 1995
a. Graphsshow the paths that variablesfollow in response to the constructedlabor supply, labor demand,and user cost shifts
that occurredin Franceover the sample period. The constructionof the time series of these shifts is described in the text and in
the notes to figures 7, 8, and 9. Solid line results when elasticity of substitution(a) is set to one; dashed line, when elasticity is set
to two. All otherparameterstake the values given in table 2.
Table 3. Labor Supply and Demand Shifts and Implied and Actual Unemployment
Rate Changes, 1970-95a
Percentage points
1970-81 1981-95
Change in Change in
Labor Labor unemployment

Labor Labor unemployment
supply demand rate supply demand rate
Country shift shift Implied Actual shift shift Implied Actual
Continental
Australia 12 -4 5 4 -5 3 8 3
Austria 23 0 13 1 0 6 5 4
Denmark 22 0 15 8 -13 8 0 0
France 16 -1 7 6 -10 11 3 4
Germanyb 19 -6 6 5 -10 3 -2 -1
Netherlands 7 -3 7 6 -12 8 -1 1
Spainc 39 -5 20 13 -3 11 10 8
Sweden 21 -3 9 2 -3 5 5 5
Belgium 30 -9 17 9 -1 4 9 3
Ireland 3 -3 -3 5 -18 8 -1 2
Italy 6 -2 -4 3 -4 8 7 5
Anglo-Saxon
United States -1 -3 -3 2 -4 0 -1 -2
Canada 4 11 2 2 20 -4 22 1
United Kingdomc 11 1 4 7 -10 -6 -4 0
Source: Author's calculations, based on data used in table 1.
a. Labor supply shifts are changes in In 0 times 100. Labor demand shifts are changes in the log of the markup (In p.)
times 100. Both series are constructed as described in text and in notes to figures 7 and 8. Numbers shown in this table use
= I and A = 0.8. Implied changes in the unemployment rate are taken from dynamic simulations using the constructed
shifts in labor supply and demand and in user cost that occurred over the sample period, and the parameterand steady-state
values given in table 2.
b. Sample period ends in 1990.
c. Assuming constant real interest and depreciation rates.
and demand shifts, it is useful to recall that over a period of five to ten
years, the effect of 1 percent adverse shift in labor supply is to increase
the unemployment rate by about 0.8 percentage points. Over the same
period, a 1 percent adverse shift in labor demand increases the unem-
ployment rate by about 0.4 percentage point. Table 3 suggests a number
of conclusions.
The model does a decent job of explaining unemployment evolutions
across countries and across subperiods. The cross-country correlation
between predicted and actual changes in unemployment is 0.60 for the
first subperiod and 0.36 (excluding Canada, 0.65) for the second
subperiod.
For a large number of countries, however, the model overpredicts

the increase in the unemployment rate from 1970 to 1981. The worst
case is Austria, with a predicted increase of 13 percentage points versus
an actual increase of only 1 percentage point. Similarly, the predicted
increase for Spain is substantially larger than the already large actual
increase. The origin of these overpredictions points to an important
shortcoming of the model. The model assumes that each economy was
on its steady-state growth path in 1970. Starting from such a balanced
path, any increase in the wage above that implied by total factor pro-
ductivity growth must lead to higher unemployment. But a country that
is below its steady-state growth path experiences an increase in the ratio
of capital to labor, which allows wages to grow faster than total factor
productivity without adverse effects on unemployment. This appears to
have been true of Austria, where the profit rate was unusually high in
the early 1970s and has declined steadily as the ratio of capital to output
has increased. Likewise for Spain, although in this case the increase in
wages has been much larger than was warranted either by convergence
to the steady-state path or by factor productivity growth.
Some other large discrepancies point to potential data issues as much
as to deficiencies of the model. The proximate cause of the very large
predicted increase in unemployment in Canada since 1981, for example,
is the large constructed adverse shift in labor supply since that date.
This shift reflects wage growth in excess of low measured total factor
productivity. The low measured factor productivity growth, in turn, can
be traced to very high measured capital growth in the 1980s, which
leads to a small Solow residual. And the low measured capital growth
comes from a very large decrease in the measured price of investment
goods, leading to high measured capital accumulation in volume, given
observed capital accumulation in dollars. According to the data, the
relative price of capital goods has decreased more than twice as fast in
Canada as in the United States. This does not seem plausible.35
35. The OECD is not to blame, as its data on the price of investmentgoods are
consistent with those publishedby Canada.Accordingto StatisticsCanada,the price of
producer'sdurableequipmentrelative to the GDP deflatorhas decreasedfrom 100 in
1970 to 33.8 in 1995 (comparedwith 62.9 in the UnitedStates, accordingto the Bureau
of EconomicAnalysis). Partof the problemappearsto be the OECD's use of an implicit
price deflator for Canada, but a chain index for the United States. Canadahas now
introduceda chain index and its evolution is much closer to that of the United States. I
could not, however, use this new index in my simulations,as it only goes back to 1981.
Leaving aside such problems and puzzles, table 3 suggests that the
increase in unemployment in the Continental countries is typically ex-
plained by adverse labor supply shifts in the first subperiod and adverse
labor demand shifts in the second. Just as has been seen for France,
these later adverse labor demand shocks also explain the increase in
capital shares. It is interesting to note that the model suggests that labor
demand shifts have played a more limited role in Germany than in
France: most of the decrease and later recovery of capital shares in
Germany is accounted for simply by the dynamic response of factor
proportions to factor prices.
Lags in Labor Demand
In this section, I return to econometrics and look at dynamics of

labor demand. The important role played by these dynamics in the
analysis so far raises two related questions. First, is there evidence for
the lags assumed in the model and the construction of markup shifts?
Second, when such lags are allowed and estimated, how much evidence
remains of shifts in labor demand since the mid-1980s? There is a large
literature on the dynamics of labor demand in general, and of labor
demand in the context of European unemployment, in particular.36My
goal here is simply to provide a description of the data and attempt to
separate dynamics and time effects.
For each of the two groups of countries-Continental and Anglo-
Saxon-I run the following panel regression:
(12) In (f) = (U(L)In vi, + xi + x, + Ei,,
where i denotes a country; t denotes time; fi is labor in efficiency units;

To see the potentialeffects of shifting to a betterindex, I have redonethe simulations
for Canadausing the U.S. relative price of investmentinsteadof the Canadianone to
computethe evolution of the capital stock in volume, total factor productivity,and so
on. The labor supply shift for 1970-81 decreases from 4 (in table 3) to -6, and for
1981-95, from 20 to 10; the increase in the unemploymentrate over 1981-95 is
12 percent-still too high, but roughlyhalf the numberin the table.
36. On labor demand in general, see, for example, Hammermesh(1993); in the
context of Europeanunemployment,see, for example, Layard,Nickell, and Jackman
(1991).
Table 4. Regressing Factor Quantity Ratios on Current and Lagged Wage Rates,
1961-95a
Coefficient on wages
Sum p valuesb
Panel and Time t -3 to t -I to Lagged Time
estimator dummies Current t-I t-2 t -9 t -9 wages dummies
Continental
OLS No -0.14 -0.51 -0.31 -1.31 -2.27 0.00 ...
OLS Yes -0.74 -0.35 -0.30 -1.09 -2.47 0.00 0.00
IVC No -0.51 -0.44 -0.44 -1.40 -2.79 0.00 ...
IVC Yes -0.87 -0.37 -0.32 -1.13 -2.69 0.00 0.00
Anglo-Saxon
OLS No -1.19 -0.49 -0.16 -0.10 -1.94 0.00 ...
OLS Yes -1.09 -0.34 -0.32 -0.20 -1.95 0.00 0.09
IVC No -1.30 -0.47 -0.16 -0.03 -1.96 0.00
IVC Yes -1.37 -0.33 -0.60 0.79 -1.51 0.00 0.05
Source: Author's regressions (see table I).
a. The dependent variable is the log factor quantity ratio, In (filk)j,, where ht is labor in efficiency units and k is the capital
stock in country i and year t. Independent variables include current and lagged values of the log wage per efficiency unit,
In wi,,in addition to country and time fixed effects. Equation estimated is ( 12) in the text. Samples are as described in notes
to table I.
b. Probability of obtaining these data given that all lagged wage or time dummy coefficients are zero. Covariance matrix
is Newey-West corrected.
c. Estimated using constructed labor supply shifts (see figure 7), lagged zero to nine years, as instruments for wages in
efficiency units.
k is capital; wviis the wage per efficiency unit of labor; xi and xt are
country and time fixed effects, respectively; and +(L) is a lag polyno-
mial. As is well known, this log-log specification does not hold exactly,
except in the Cobb-Douglas case-this is why the estimation above
relied instead on a log-log relation between factor price and factor
quantity ratios, which holds for the CES case-and must be thought of
as a log approximation. Under that interpretation, the sum of estimated
coefficients on current and lagged wages, 4(1), is approximately equal
to the ratio of the elasticity of substitution to the share of capital, u/a.
This fact is useful for interpreting the results below.
The results of this estimation are reported in table 4 and figure 13.
Table 4 presents four sets of results for each group. Two are obtained by
ordinary least squares, with and without time dummies. The other two are
obtained using instrumental variables, with and without time dummies.
For the present purposes, the labor supply shifts constructed earlier are
natural instruments.37The regression results yield two conclusions.
37. Another issue arises from the fact that business cycles generate a correlation
Figure 13. Time Variationin Factor Quantity Ratios, Controllingfor Wage Rates,
Continentaland Anglo-SaxonCountries,1963-95a
Log index, 1970= 0
Continental
0.1 _ Withoutlagged wages
0.0
{).1
-0.1I > wages
h ~~~~~~~~~~~lagged
-0.2
-0.3
Anglo-Saxon
0.1
Withlagged wages
0.0
-0.1I
-0.2
Withoutlagged wages
-0.3
I I ~~~I ~ ~I ~ ~I ~ I
1965 1970 1975 1980 1985 1990
Source:Author'sregressions,based on data used in figure 1.

a. Figure plots, for each panel of countries, the time dummy coefficients that result when the log ratio of labor to capital is
regressedon the current log wage in efficiency units, country dummies, and time dummies. Where indicated, regression also
includes lagged values of the log effective wage. Regressioncorrespondsto equation 12 in the text and is describedin the notes to
table4; the figurepresentsresultsfrom the instrumentalvariablesestimation.
First, there is evidence of long lags in labor demand. While the best
lag length varies across specifications, the results suggest that up to
nine years are needed to capture the dynamics of adjustment. If one
takes the average capital share to equal 0.35, the sum of coefficients
on current and lagged wages implies an elasticity of substitution a little
under 1.0, which is unable to yield a significant medium-run increase
in the capital share in response to adverse labor supply shifts.
Second, even allowing for lags, time dummies remain highly signif-
icant for the Continental countries, but are only marginally so for the
Anglo-Saxon countries. More information is given in figure 13, which
plots the series of time fixed effects estimated by instrumental variables,
for each group. For comparison, it also plots the series of time effects
from a regression that allows only for current wages-and country
dummies. For the Anglo-Saxon countries, the shifts are small, whether
or not one allows for lags in labor demand. For the Continental coun-
tries, allowing for lags reduces the size of shifts in the 1970s but does
not otherwise change the general shape of the shifts: there is still a large
unexplained adverse shift in labor demand increasing from the early
1980s to the present.
The Source of Shifts in Labor Demand
A key unanswered question is whether the source of the labor demand

shifts lies in technological bias away from labor or changes in the
distribution of rents. A similar question has been discussed in the con-
text of the shift in demand for skilled relative to unskilled labor. The
prevalent view is that this shift reflects technological bias away from
unskilled workers. But some argue that it may reflect institutional
change; for example, the weakening of unions leading to an increase in
between the ratio of labor (in efficiency units) to capital and the wage (in efficiency
units), due either to deviations of the marginalproductfrom the wage or to mismea-
surementof total factorproductivity.A demand-drivenboom is typicallyassociatedwith
high measuredtotalfactorproductivitygrowth.Highmeasuredproductivitygrowthleads
to a large increase in labor measuredin efficiency units (fi) and a large decrease in the
wage per efficiency unit (w). In orderto alleviate some of the business cycle effects, I
have tried using only values of the labor supply shifts lagged by three or more years.
The resultingpatternof estimatedcoefficients is typically nonsensical.
wage differentials across workers.38 The approach in this debate has

been to look at cross-country or cross-sectoral evidence and try to link
the shifts to factors that are more likely associated with one or the other
explanation. For the question at hand, there would appear to be a more
direct approach. Under the hypothesis of technological bias, one should
see changes in the production function; under the hypothesis of changes
in markups, one should not. This suggests a simple test.
Consider the following example. Suppose that the production function
is Cobb-Douglas, with exponent a on capital, and that except for possible
changes in a over time, technological progress is multiplicative:
(13) y = z (4, n)('alc ka,
where 4, is a constant that depends on the units in which labor is

measured. One typically ignores units in production functions, as a
change in units merely changes the constant term. But in the present
context, where one wants to allow for changes in a, the units do make
a difference: one does not want the effect on y of a given change in a
to depend on whether labor is measured in thousands or in millions of
workers.
Suppose that the markup is equal to pLand assume away costs of
adjustment, so that the wage times the markup is equal to the marginal
product of labor. It is straightforward to derive that the capital share
will be equal to ox 1 - (1I/V)(1 - a). Assume that, initially, pLis
equal to one, so that ot is equal to a.
Suppose that one observes an increase in the share of capital, (x. This
could be due to one of two factors (or a combination of them). One
hypothesis (HO) is that it comes from technological bias; that is, an
increase in a, reflected one for one in an increase in o(. The other
hypothesis (HI) is that it comes from an increase in the markup, pL,
increasing ot given an unchanged value of a, denoteda.
Under HO, the production function is given by
y = z N, n) k-.
Or, taking logs and rewriting,
n() (1 - ct) ln() + (1 - co) ln 4, + ln z.
38. See, for example, Fortinand Lemieux (1997).

Similarly, under HI the production function is given by

y = z (4 n)( I- ) kTi.
Or, taking logs and rewriting,
n() (1 aI) ln() + (I - I) ln q, + ln z.
In essence, under HO the change in the share should be reflected in a

change in the effect of the ratio of labor to capital on the ratio of output
to capital; under HI it should not.
One can nest these two hypotheses by writing
n() = 4(1 - ot) ln( ) + (1- 4))(1 - In)in(k)
+ [4(1 - oX) + (1 - +)(1 -a-]ln t4 + ln z.

Or, rearranging,
(14) ln(k) - (1 I) n()
=( - o) ln( ) + [1 - -(1 (I - 4)a] ln 4 + ln z.
A value of 4 equal to zero implies that the change in the share is due
to markup changes. A value of 4 equal to one implies that the change
in the share is due to biased technological progress.
Estimating equation 14 requires the specification of z, the technolog-
ical level.39 I assume that ln z;, (for country i at time t) is equal to a
country-specific quadratic trend, f,(t), plus a stationary component, Ei,.
If one assumes that -a, the underlying constant value of a under HI, is
equal to the mean value of the share in the sample, one can construct
the time series for the dependent variable, X1 [ln(ylk) - (1 --a)
ln(nlk)], and the first right-hand-side variable, X2 -- [(a-- o)
ln(nlk)], and run the following panel regression:
39. In this case, it is not possibleto expressvariablesin efficiencyunitsandeliminate

z from the regression, as I have done above, because the weights needed to constructz
dependon which of HOand HI holds.
Table5. Regressionsto Distinguish between TechnologicalBias and Markup

Changes, Fourteen-CountryPanel, 1961-95a
Estimatedtechnological p value:
Estimator bias parameterb equalityacross countriesc
OLS 0.98 0.03
(0.17)
IVId 1.47 0.00
(0.35)
1V2e 0.69 0.73
(0.39)
Source: Author's regressions (see table 1).
a. Equations estimated correspond to ( 15) in the text. Sample includes both Continental and Anglo-Saxon countries; for
details, see notes to table 1.
b. The coefficient 4 from equation 15. A value of zero implies that changes in capital shares are due to markupchanges;
of one implies that they are due to biased technological change.
c. Probability of obtaining these data if the parameter4 is the same across countries.
d. Estimated using current and lagged values of constructed labor supply shifts (see figure 7) as instruments.
e. Estimated using current and lagged values of user cost (see figure 9) as instruments.
(15) XI,i = + X2i, + xi + p(xj,,+ f(t) + Eit.
Table 5 presents the results of this estimation. All fourteen countries

are pooled together, as there is no obvious reason to treat the two groups
separately in this context.40 The first line gives the results of ordinary
least squares estimation, reporting the estimated value of 4)and also
the test of the constraint that when allowed to differ, 4)iis the same
across countries. The second and third lines report the results of instru-
mental variable estimation; the second uses current and lagged labor
supply shifts, whereas the third uses current and lagged shifts in user
cost. To the extent that the user cost is dominated by movements in the
world interest rate, it is a good candidate instrument.
Point estimates of 4)are significantly closer to one than to zero. This
argues for technological bias (which corresponds to 4)= 1) rather than
markups. Allowing for country-specific values of 4), however, yields a
wide range of estimated values, from - 2.5 to 4.0. These have large
standard deviations: equality of all 4)'s is rejected only at the 3 percent
level in the ordinary least squares case, and is not rejected at all in the
second instrumental variables case.
The great variation in the coefficient 4),across countries-even if
40. The results are very similar if only the Continentalcountriesare used. The test
has little power for the Anglo-Saxon countries, because their shares do not vary very
much.
the variation is not always significant-makes me uneasy about the

results in table 5. One reason for the variation may be that the assump-
tions underlying this approach (a Cobb-Douglas production function
and no costs of adjustment, so that the share is the right measure of the
effect of capital on output under HO) are overly strong. I suspect,
however, that aggregate data may not be able to speak precisely. In
effect, the question is whether, given total factor productivity, an in-
crease in the ratio of labor to capital has a larger effect on the ratio of
output to capital in those countries where the capital share has in-
creased. But little is known about the exact process of factor productiv-
ity, nor are there large changes in the ratio of labor to capital that would
yield sharp estimates. Thus further progress will probably have to rely
on the cross-sectoral evidence.
Conclusions
In the 1970s, most of the Continental countries were affected by

large adverse shifts in labor supply. The initial effect of these shifts
was to decrease profit rates and capital shares. Over time, firms reacted
by moving away from labor, which led to a steady increase in unem-
ployment and the recovery of capital shares. In most of these countries
labor supply shifts have now abated, and in some they have vanished
altogether. But since the early 1980s, they have been replaced by ad-
verse shifts in labor demand. These shifts explain why unemployment
has remained high, and also explain the further rise in capital shares.
By contrast, the Anglo-Saxon countries have largely been shielded
from both the adverse labor supply shifts of the 1970s and the later
labor demand shifts. This accounts for their different evolutions-both
in terms of unemployment and in terms of capital shares-from the
Continental countries.
These findings yield a mixed message for the future course of un-
employment in the Continental countries. On the one hand, labor mod-
eration-defined as the disappearance of the adverse labor supply
shifts-may no longer be sufficient to ensure a quick return to low
unemployment. On the other hand, one should expect that the adverse
effects of labor demand shifts on unemployment will disappear over
time. The decline in the ratio of labor to capital should eventually be
offset by the entry of new firms and an increase in capital. Thus,

assuming no further adverse labor demand shifts (which is more likely
to hold if such shifts come from changes in the distribution of rents
than if they come from technological bias), my results imply a slow
decline in unemployment in the future.
Even if the story presented in this paper is broadly correct, many
questions need to be examined before forecasts or policy advice can be
given with much confidence. I discuss two of these here.4' First, can
one relate shifts in labor supply and demand to underlying observable
variables? Identifying the source of the adverse shifts in labor supply
may be the easiest part, as it builds on a large body of research. There
is wide consensus that these shifts come both from a combination of
economic events (such as increases in oil prices, the slowdown in total
factor productivity growth) and induced changes in labor market insti-
tutions (such as more generous treatment of unemployment, increases
in employment protection, minimum wage legislation). One of the puz-
zles faced by previous research was how to reconcile the fact that
unemployment has remained high while oil price hikes have been more
than reversed, workers by now must have adapted to slower underlying
productivity growth, unions appear to have become weaker, and gov-
ernments have started tightening social insurance programs. My find-
ings that labor supply shifts have indeed largely decreased, and that the
persistence of high unemployment is a result of shifts in labor demand
since the mid- 1980s, offer a resolution to the puzzle. Indeed, the paper
makes a more general methodological point. Much of the econometric
research on the increase in unemployment has involved estimating,
across countries and time, a reduced-form equation for the unemploy-
ment rate as a function of a number of observable variables.42 I suggest
41. Those who have read an early version of the presentpaper(Blanchard, 1996)
may note the absence of a theme developed there: the relation between inflation and
markups.My initial work showed a strongtime-seriesrelationbetween the increaseof
the measuredmarkupand the decrease in inflation. Now that I have adjustedthe con-
structionof markupsto take into account lags of adjustmentof factor proportions,the
relationremainsbut is weaker. And when time effects are allowed in panel regressions
of factorprices on factorquantities,domestic inflationfor each countryis only margin-
ally significant. For these reasons, I have left the explorationof a potential relation
between markupsand inflationto furtherwork.
42. See, for example, Jackman,Layard,and Nickell (1996).
instead a potentially more productive approach: to distinguish supply

and demand shifts and then try to explain each of them separately.
Identifying the source of the more recent labor demand shifts may
be more difficult. The attempt to distinguish between shifts in the dis-
tribution of rents and biased technological adoption in this paper is
inconclusive. As a next step, I have started looking at cross-sectoral
evidence for France. I find that labor demand shifts are present in most
sectors, but I have not yet made progress in relating relative shifts to
potential underlying factors, such as the initial level of rents, the esti-
mated initial degree of labor hoarding, the initial structure of bargain-
ing, and so on.
There may also be something to be learned from the cross-country
evidence. Based on the numbers in table 3, the cross-country correlation
between the labor supply shifts over 1970-81 and the labor demand
shifts since 1981 is 0.40: countries that had larger adverse labor supply
shifts in the 1970s typically have also suffered larger adverse demand
shifts since. It is tempting to see this relation as causal. Firms in coun-
tries where labor supply shifts were stronger may have decided to adopt
technologies that use less labor and more capital. The lags in introducing
these new technologies may be even longer than those associated with
43
changing factor proportions within the set of existing technologies.
This correlation between labor supply shifts and labor demand shifts
also suggests a tentative explanation for the differences between the
experiences of the Anglo-Saxon and the Continental countries. In the
Anglo-Saxon countries, the slowdown in productivity growth was
smaller, and the induced adjustment of labor market institutions was
more limited. Building on the argument of the previous paragraph,
smaller adverse labor supply shifts from 1970 to 1981 may then explain
why adverse shifts in labor demand have also been more limited since
1981.
The second question that must be answered is: what is the relation
of the shifts between labor and capital documented in this paper to the
shifts between skilled and unskilled labor documented in recent research
in labor economics? It is an intriguing fact that relative demand shifts
between skilled and unskilled workers appear to have been particularly
43. A numberof recentpaperstell of endogenousbias in technologyadoptionalong

broadlyrelatedlines; see, for example, Acemoglu (1997) and Zeira (forthcoming).
strong in the Anglo-Saxon countries, and relative demand shifts be-

tween labor and capital appear to have been particularly strong in the
Continental countries. One wonders whether there may be an integrated
explanation; whether the Anglo-Saxon countries have seen a shift from
unskilled workers to skilled workers, whereas the Continental countries
have seen a shift from unskilled workers to capital.
A preliminary look at numbers, however, does not seem to indicate
that the increase in the capital share in the Continental countries since
the early 1980s has come primarily at the expense of unskilled labor.
Data for France constructed by Jean Pierre Laffargue and Anne Saint-
Martin, for example, imply that the decrease in the labor share from 68
percent in 1982 to 58 percent in 1990 has come not only from a reduc-
tion from 10 percent to 7 percent in the share of unskilled workers
(defined as blue collar workers plus unskilled employees), but also from
a reduction of the share of skilled workers, from 58 percent to 51
percent.44 In general, the data from France and other countries appear
to suggest two largely unrelated evolutions: a general and steady shift
away from unskilled labor and, in continental Europe, a shift away
from labor as a whole since the early 1980s. A next step will be to
extend the exercise of this paper to three factors of production: skilled
workers, unskilled workers, and capital.
44. Laffargueand Saint-Martin(1997). I thank Jean Pierre Laffarguefor making

those data availableto me.
Comments
and Discussion
William D. Nordhaus: In this paper, Olivier Blanchard struggles with

one of the major puzzles of current macroeconomics. Why do the econ-
omies of Europe and North America look so different? The paper con-
tains many interesting insights and a few remaining puzzles. I organize
my comments around three topics: the facts, the model, and the expla-
nations.
Begin with the facts, which concern the returns to capital and labor.
Blanchard examines the capital share and the relative returns to labor
and capital in a number of OECD countries. He concentrates on the
business sector. The major findings are the following: First, the behav-
ior of factor shares looks quite different in the "Anglo-Saxon" coun-
tries of the United States, Canada, and the United Kingdom from in the
continental European countries of Germany, France, Italy, and Spain
(Blanchard's list of Continental countries is longer, but nothing is lost
by focusing on these four). Whereas capital shares have been quite
stable in the Anglo-Saxon countries, in the Continental countries, they
fell sharply in the 1970s and then have risen even more sharply since
the early 1980s. Second, as is well known, unemployment trends differ
in the two groups of countries. Unemployment is roughly trendless in
the United States, while it has been increasing steadily in continental
Europe.
I see no reason to quarrel with the basic facts, although the business
sector is an unfortunate target for this kind of analysis. It contains too
much that muddies the statistical picture if one really is interested in
the return to capital. For example, in the United States the business
142
sector contains the services to owner-occupied housing, earnings of

proprietorships, government enterprises, and farming. For 1995, in the
corporate sector the share of labor in national income (or, national
output at factor cost) was 80 percent. In the proprietor's sector, that
share was 32 percent, while in "other private business"-which is
primarily the imputed profit that you and I get renting our houses to
ourselves-the share of labor was 3.4 percent. These animals are so
different that I would keep them in different cages for the statistical
analysis. And one sometimes forgets that the profits of financial insti-
tutions include the seigniorage earned by the Federal Reserve.
As long as one is beating up on the data, it is useful to remember
that the denominator of these return numbers is the net stock of repro-
ducible capital at replacement cost. And the numerator includes not
only the implicit return on owned capital but also any return to owned
land, subsoil assets like oil deposits, intangibles, and goodwill, as well
as any monopoly power or excessive managerial compensation; fur-
thermore, it omits the substantial capital losses on capital that have
been experienced in recent years. It is interesting to note that the post-
tax return is close to the weighted average cost of capital for this period.
Just for fun, I have compared the data on U.S. rates of return and
shares for nonfinancial corporations taken from the National Income
and Product Accounts (NIPA) with Blanchard's U.S. data from the
OECD. The property share and rate of return data are significantly
different. Blanchard's share and rate of return series show statistically
significant positive time trends, whereas the NIPA data show statisti-
cally significant negative time trends. Over the 1960-95 period, Blan-
chard's estimated rate of return rises from 11.2 percent to 17.7 percent,
but the NIPA return falls from 8.6 to 8.2 percent; and Blanchard's
estimated share of property income rises from 27.3 percent to 33.3
percent, but the NIPA share falls from 19.6 to 17. 1 percent. The NIPA
data are clearly a cleaner concept for these purposes, but I am puzzled
by the difference. Most of that difference comes from the fact that
Blanchard includes depreciation, which seems ill advised in an analysis
of trends in factor shares and rewards. Moreover, it does not make
much sense to apply stories about labor rigidities, biased technological
change, labor unions, labor hoarding, and constraints on hiring and
firing to owner-occupied housing or sole proprietorships run by neuro-
surgeons. Blanchard's story might well survive translation to the prop-
erty income of nonfinancial corporations in different countries, but until

that is shown, my enthusiasm will remain somewhat dampened.
Because profits are such an intoxicating topic, consider two other
points. First, what has happened to the post-tax corporate rate of return
in recent years? Complete data are available only through 1995, but it
might come as a surprise that the rate of return is basically trendless.
In fact, the 1995 post-tax return of 6.0 percent was only marginally
above the 1960-95 average of 5.7 percent. There is no brave new world
of capitalism in the rate of return data-but no falling rate of profit
either. What has happened to the average tax rate on corporate capital,
however, is striking. After the corporate tax rate had averaged about
35 percent from 1965 through 1980, the supply-side revolutionaries
lowered it to 20 percent in 1982, and it had only risen to 26 percent as
of 1995. If Arthur Laffer had been paid his marginal private product as
chairman of the tax-cutting committee of corporate America, he would
have earned $350 billion in the 1980s alone.
Blanchard accounts for these strange developments in capital shares
and profit rates by developing an interesting new calibrated model of
the medium run and then using it, along with some simple economet-
rics, to compare his theory with developments in the two groups of
countries. His model has four features. First, there is a conventional
production structure, in which output is produced by a Cobb-Douglas
(or sometimes, a CES) production function in capital and labor. Second,
there is a reduced-form labor supply relationship, in which the employ-
ment rate depends on the real wage. Third, there is a price markup over
labor costs, which in the model reflects multiple little monopolies, so
that the markup is determined by the price elasticity of demand. Fourth,
there is a novel set of assumptions regarding firms and adjustment
dynamics for firms, capital, and labor.
In the long run, the model behaves like standard supply and demand
functions for labor. (Just to protect everyone's reputation here, I
should-although I do not-use quotation marks each time I write labor
supply and labor demand because, as Blanchard emphasizes, these are
not the usual competitive functions, but reflect more complicated mar-
ket structures, institutional details, and so forth.) The new twist is that
labor supply determines the employment rate, or employment relative
to a benchmark level of employment, which is a moderately elastic
function of the real wage. Given the parameters of the model, the
demand for labor is an infinitely elastic function of the real wage rate;
the labor demand curve shifts down with an increase in the rate of
interest, the rate of capital taxation, the depreciation rate, and the Cobb-
Douglas share of capital. The steady-state properties are that adverse
or upward shifts in labor supply have no effect on real wages but do
increase the "unemployment rate." Adverse or downward shifts in
labor demand are fully reflected in a lower real wage and also produce
a steady-state change in the unemployment rate.
Blanchard uses the model to simulate the impact of different shocks
to the system. He suggests that Continental countries suffered from a
series of adverse labor supply shifts in the 1970s. These would clearly
lead to increases in the unemployment rate. (Be warned that the un-
employment rate is a mixture of voluntary unemployment, efficiency
wage and wait unemployment, and classical unemployment, but no real
cyclical unemployment of the Keynesian variety.) Blanchard's model
also shows that these shocks lead to a decrease in capital shares, al-
though I am not convinced that this result is robust to changing the
modeling dynamics.
The new piece of the story concerns the last decade. Since the early
1980s, according to Blanchard, continental Europe has suffered a num-
ber of adverse labor demand shocks. There is some question in his mind
as to whether these represented biased technological change or shifts in
the markup, although his econometrics leads him to the conclusion that
they are likely to have been biased technological change, which in-
creased the Cobb-Douglas coefficient on capital. As for the Anglo-
Saxon world, Blanchard writes: "By contrast, the 'Anglo-Saxon' coun-
tries appear to have been largely shielded from both the adverse labor
supply shifts of the 1970s and the labor demand shifts of the 1980s and
1990s. This accounts for the differences from the Continental countries
in the evolution of unemployment and of capital shares."
I conclude by reflecting on which of the explanations for these phe-
nomena seem plausible and consistent with the cross-section of human
experience. To do so, I round up both the usual suspects and those
chosen by Blanchard. The augmented list of suspects comprises the
business cycle, the cost of capital, markups and market power, inter-
national trade, biased technological change, and labor market rigidities.
On the issue of the business cycle, profits and profit rates are highly
cyclical. This is probably a good part of the recent profits surge in the
United States, but given the great depression in continental Europe, it

can hardly explain the rising capital share and rate of profit there. I
would have liked some attention to cyclical issues, but given the paper's
focus on the medium term, this is a pardonable offense.
The paper does not discuss the cost of capital or increases in capital
taxation as possible reasons for the higher profit rate or capital share.
It would, in fact, be hard to make this case given the stock market
boom, the unprecedented rise in Tobin's q, and the declining tax rate
of corporate capital around the world. So this suspect is let off the
hook.
Blanchard has some kind words to say about the possibility that
higher capital shares in the Continental countries resulted from higher
markups. I find this part of the discussion confusing. Blanchard's
markup is, by definition, the ratio of price to marginal cost. (By the
way, this is different from the usual markup-that is, the relationship
between price and some measure of standardized average cost-as it is
generally used by industrial firms or recognized by those responsible
for the marking up.) Under the usual simplified conditions, the Blan-
chard markup is the usual function of the price elasticity of demand. If
one is to take his model literally (which he does not), the implication
is that the increase in the capital share is due to an increase in monopoly
power or a decrease in the demand elasticity in Europe. This idea would
be immediately thrown out of court, and rightly so. If anything, com-
petition has grown in continental Europe as a result of increasing open-
ness to international trade. If anything, the movement to a single market
should drive down Blanchard markups in Europe relative to those in
the United States, which unified its market two centuries ago. There
may be different interpretations of the markup, but the obvious one
would seem to be unrealistic on prima facie grounds.
In terms of the impact of increased openness on different countries,
there has been much debate about the effect on the structure of wages,
and, on balance, the evidence seems to suggest that unskilled labor (a
close substitute for inputs that are abundant in developing countries)
has had its wages depressed. Little attention has been paid to the ben-
eficiaries of the increased imports. The usual view is that the capital
and skilled labor of the OECD are relatively scarce in the world as a
whole, and that these factors should benefit from increased openness.
I would emphasize, instead, that the really scarce factor in the world
economy is able and ruthless managerial talent in multinational corpo-

rations. If so, the opening of high-income markets over the past two
decades has had the effect of increasing the return and share of capital
income in large corporations. Moreover, to the extent that Europe was
closed relative to the United States, profits will have been more bene-
fited in that region than in the United States.
Blanchard puts forward the novel argument that the rise in capital's
share in continental Europe is due to biased technological change. In
his framework, biased technological change is simply an increase in
the Cobb-Douglas coefficient on capital in the production function. This
argument strikes me as completely implausible. To the extent that this
is technological change rather than substitution, one would expect it to
affect almost all comparable countries, including those classified as
Anglo-Saxon. What are these German labor-saving inventions that have
not crossed the English Channel? Has a French chef discovered a labor-
saving recipe for pommes frites that has not yet been translated by
McDonald's? Martin Baily and his associates at the McKinsey Global
Institute have uncovered evidence that continental Europe has actually
lagged behind the United States and Japan in introducing the latest
technologies over the past decade or so. Lagging behind hardly seems
a likely route for a region-specific burst of technological change.
A more promising line of argument about biased technological
change would look to the information and computer revolution. The
use of computers could easily be seen as capital saving (lowering the
ratio of capital to output) through the increased efficiency in manage-
ment of production. The miraculous ways in which computers allow
better management of airline, railroad, and trucking fleet, of oil refi-
neries, of overnight delivery services, and of inventories are well
known. There is, however, only weak evidence of a decline in the ratio
of chain-weighted capital to output for the United States.
The final issue concerns institutional labor market policies, which
seem closest to what Blanchard calls labor supply shocks. In this cat-
egory are union density and coverage, benefit levels and the duration
of unemployment insurance and welfare policies, the strength of em-
ployment protection and limitations on separations, and the structure
and level of labor and other taxes. I believe that it is here, rather than
along linguistic or geographical lines, that one might find the source of
the puzzling movements in labor and capital shares. Blanchard's Anglo-
Saxon countries are actually a subset of those with relatively free labor
markets; his Continental countries are high on the list of those with
major labor market rigidities. For example, consider the following
OECD ratings. Ranked on unemployment protection-using an inverse
French grading scale, where 1 affords the least employment protection
(the United States) and 20 affords the most (Italy)-the three Anglo-
Saxon countries have an average score 3.7, while France, Germany,
Italy, and Spain average 16.5. On the labor standards rating (a scale
from 0 to 7), the Anglo-Saxon countries average 0.7, while these four
Continental countries average 6.5. In terms of unemployment insur-
ance, these four Continental countries have a generosity rate (replace-
ment rate times maximum duration) of 180 percent-years, while the
Anglo-Saxon countries average 79 percent-years, a difference of almost
2 1/2 times. ' It would therefore be more plausible to classify the Anglo-
Saxon countries as the "FreeLabs" (those with largely free labor mar-
kets) and the Continental countries as the "RegLabs" (those with heav-
ily regulated labor markets).
This distinction would resolve half of the distributional puzzle, that
concerning the labor market, but the profit puzzle remains unsolved.
Unless the elasticity of substitution of capital for labor is greater than
one, there is no simple and robust story about why the RegLabs would
experience increasing profit rates or capital shares while these were
relatively stable in the FreeLabs. There may be lots of unemployment
in Europe, but there is still much work to be done by future Brookings
Panels on this fascinating and important issue.
Edmund S. Phelps: The main advance of this paper is that it introduces

wage and markup behavior explicitly into the analysis of the evolution
of unemployment together with that of factor shares. To date, most
economists have been doing reduced-form analyses of the unemploy-
ment rate with only an implicit consideration of wage rates. Since
looking at wages and factor shares provides a useful check on the
consensus explanation of the rise of European unemployment, this is a
very worthwhile exercise. The real importance of the paper, however,
lies at the level of theory wars. Blanchard, who used to stick to a
Keynesian approach fortified by hysteresis in dealing with secular
1. These numbers are drawn from Nickell (1997).

change in unemployment, seems now to have switched, as have several

others, to the intertemporal equilibrium endogenous natural rate ap-
proach, in which neither money nor hysteresis is required.
I want, first, to explain that Blanchard's results are in line with the
consensus that has emerged on the rise of unemployment in Europe.
Even the equational model is based on off-the-shelf theoretical com-
ponents, such as incentive wage models underpinning the wage curve.
Unfortunately these foundations are sometimes obscured or even obfus-
cated-for example, when the wage curve is called "labor supply"
(Blanchard's quotation marks). Second, I argue that the paper seriously
misleads when it portrays the experience of the "Anglo-Saxon" coun-
tries as radically different from that of the core Continental countries.
The existence of a consensus on European unemployment is dem-
onstrated by the symposium forthcoming in the May 1998 issue of the
Economic Journal, and can be outlined as follows. Early in the con-
vergence of views, most investigators came to believe that the large
expansion of the welfare state in the 1960s and 1970s had shifted up
the wage curve above where it would otherwise have been by the mid-
1970s, which, taken alone, must have raised the natural rate of unem-
ployment. Since welfare entitlements-the various social insurance
and social assistance programs-were generally not contingent on the
beneficiary being employed, they operated to devalue earning; the re-
sulting deterioration of employee performance (that is, more quitting,
shirking, absenteeism, and so forth at a given unemployment rate)
directly raised employee costs, as employers raised wages to shore up
employee incentives, or accepted lower performance, or both. Further,
the increase in payroll taxes that financed this expansion of welfare also
pushed up the wage curve in terms of hourly labor costs-wage and
nonwage costs. The reason is that the after-tax wage rates required for
optimum incentives will not drop sufficiently to accommodate the tax
as long as the private wealth and social entitlements of workers have
not fallen in proportion to the decline in the after-tax wage. Only in the
long run, as wealth decumulates in response to the decline in the after-
tax wage, might there be a full accommodation of the downward shift
in the after-tax demand wage. Indeed, it is possible that one is seeing
the fruits of such an accommodation in the 1990s.
The two energy price shocks of the 1970s could also have pushed up
the wage curve, raising labor's share of value added at given levels of
employment in nations where domestic workers' real private wealth and

real entitlements did not fall in proportion to the drop in their marginal
productivity, either because these workers owned shares in energy firms
or because the government owned or taxed energy.
In the latest stage of the consensus, it has become accepted that some
portion of the rise of unemployment in continental Europe in the 1970s
was the result of the sharp slowdown of productivity growth from the
extraordinary heights of the economic miracle to the sluggish rates that
have prevailed since 1974. After that fateful year, accumulated wealth,
which had been barely able (in some countries, unable) to keep pace
with productivity, began to rise to a new level relative to productivity.
And that gradual accumulation of wealth relative to the wage rates that
employers could afford put mounting upward pressure on the wage
curve, and thus drove up the natural rate. Furthermore, the prospect of
slower growth in employee productivity lowers the "marginal effi-
ciency" of hiring, thus reducing demand for labor
The "labor supply" shifts-that is, computed shifts of the wage
curve-that Blanchard calculates occurred in the 1970s (see table 3 and
figure 5) appear to corroborate the tenets of the consensus described
above. In most cases, the computed adverse shift of the wage curve
was quite large. Where it was small-the United States, Canada, Ire-
land, and Italy-the productivity slowdown was slight and the welfare
state was relatively narrow, focused on old people and young mothers.
The finding that much of the shift was later given back may be due to
a decumulation of wealth, following the decline of after-tax wages
brought by tax increases, and, in some countries, to the eventual scaling
back of entitlements.
The consensus on unemployment adds two shocks to labor demand
in the 1980s. Without these shocks, as Blanchard notes, the theory
would imply a full, or at least substantial, recovery after the 1970s.
One shock, the huge rise in the early 1980s in the average expected
long-term world real interest rate, which remains somewhat elevated to
this day, immediately reduced labor demand through two channels.
First, the higher cost of capital dampens the willingness of employers
to invest in either new employees (so they hire fewer of them) or new
customers (so they stop keeping their markup low, and thus the product
wage that they are willing to pay is reduced). Consequently, labor's
share is driven down as firms earn a better return on their existing
employees and customers, and the natural rate is driven up. One can
call this markup increase, such as Blanchard talks about, but I am
talking about markup increases that are induced by the elevation of real
interest rates. Second, there is obviously a physical capital channel
from real interest rates to labor demand, which may operate only very
gradually.
The other suspect shock is hard to get a handle on, although the
paper makes some strides toward doing so. This shock is the new or
intensified labor-saving bias in technical change that set in around the
end of the 1970s. A slowdown of the marginal productivity of labor
coupled with a speed-up of the marginal productivity of capital (relative
to the old bias of technical progress) that leaves growth in total factor
productivity unchanged will slow labor demand at any given capital
stock, but at first the wage curve will not slow in tandem. Thus the
natural rate will be gradually pushed up.
In Blanchard's computations, the labor demand shifts from 1981 are
indeed adverse in almost all of the OECD countries studied, and they
are fairly large. (Interestingly, for reasons that I do not understand,
there tend to be some positive labor demand shifts in the 1970s; so, the
net change is even larger.) I am very pleased to see that the consensus
view, into which I have put a lot of effort, comes out quite well here.
Nevertheless, in places Blanchard seems to want to cast doubt on some
of these ideas.
In regard to the real interest rate hypothesis, in figure 9 Blanchard
uses a time series for France that depicts the real rate as having by 1986
returned all the way to its low levels of the late 1970s. If those data
were correct, the rise of real rates would be short-lived and one could
hardly ascribe any durable part of the rise of unemployment from 1980
onward to real interest rates. I think this peculiar series ("bonds,
method 1") is the result of Blanchard weighing distant inflation rates,
including the very high rates around 1980-82, as heavily as more recent
ones when calculating the real interest rate from the nominal interest
rate. (Lest there should be any doubt about the apparent importance of
the real rate, if one juxtaposes the graph of the evolution of the unem-
ployment rate in the United Kingdom and that of the world real interest
rate from the 1960s to recent times, it is difficult not to be impressed
by the fit.)
At another level, Blanchard pays close attention to the phenomenon
of adjustment costs smoothing out expansions and contractions in firms'

workforces. He leaves the impression that this mechanism is capable
of providing a new basis for the wage curve. The idea seems to be that
whenever new workers are hired, managers and their employees have
to learn how best to divide their labor and the bits and pieces of capital
among the workers. So, there is some marginal adjustment cost when
new workers are hired and, very plausibly, the marginal adjustment
cost per new hire is greater the faster the rate of hiring. There is also a
marginal adjustment cost of downsizing, which also makes sense, as
the employees that survive the cuts have to learn how best to use the
bits and pieces released by the employees who have been let go. But
this story does not generate incentive wages above the market-clearing
level, nor, as a corollary, involuntary unemployment. (For example, in
this model no expensive learning or training is involved in replacing a
worker who has quit; the replacement worker will immediately be up
and running, slotted in at no cost among the bits and pieces already
optimally set up. So, the employer does not need to pay incentive wages
to dampen quitting.) Thus the wage curve is left unfounded. In fact,
even the terms "wage curve" and "incentive wage" are missing.
My comments so far have been quibbles. But there are more serious
problems in the paper: its results, and thus the thrust of its message,
are seriously misleading in certain respects. One of the problems stems
from the entirely aggregate analysis of unemployment. The paper re-
ports that in the Anglo-Saxon countries, the unemployment rate by 1996
had returned to its level in the early 1970s, or fairly close to that. It
would seem that there has been full or substantial recovery from the
"great slump" of the 1980s. (And in those countries, capital's share
in the business sector is reported to have subsided in the 1990s to the
levels of the early 1970s.) But one knows that within categories of
educational attainment-high school dropouts, high school graduates
only, some college, and college graduates-unemployment rates gen-
erally have risen, except in the top group. In the United States, for
example, the unemployment rate for adult men in the bottom group
rose from about 5 percent in the early 1970s to the neighborhood of 11
percent in 1996. How can it be that the average unemployment rate
recovered while almost all within-group rates increased? Part of the
answer, my research suggests, is that the low paid-especially men-
have been leaving the labor force in huge numbers. Another part of the
answer is substantial movement up the educational ladder. Fewer and

fewer workers have left education before receiving a high school di-
ploma, and fewer and fewer spent no time in college.
In principle, one way to allow for this phenomenon in the analysis
would be to work with a fixed-weight index of unemployment rates
constructed from within-group wage rates (and likewise, a fixed-weight
index of wage rates). For both the United States and the United King-
dom, such an unemployment index will show a far larger rise over the
past twenty-five years than does the measured average unemployment
rate: an additional increase of about 2 percentage points in the United
States and about 1.5 percentage points in the United Kingdom, accord-
ing to research with Gylfi Zoega. I And the index of within-group hourly
compensation will show a far smaller rise in the United Kingdom and
a far greater fall in the United States than does average hourly compen-
sation. But to write a paper parallel to this using these indexes would
require an index of within-group output per manhour, and such data do
not exist.
Another way of proceeding would be to look at the evolution of
unemployment and wage rates in a particular educational group. But
likewise, one could not expect Blanchard to determine whether the
unemployment of dropouts, for example, has soared because they have
priced themselves out of the market or because they have suffered from
reduced demand as a result of higher markups or increased labor-saving
bias in technological change, since there do not exist the output and
capital series for high school dropout workers that his type of calcula-
tion would require.
There is one final problem with the paper. The recovery and subse-
quent swelling of capital's share in the 1980s among the Continental
countries are presented in the paper as strong evidence for the thesis
that markup increases-perhaps prompted by the elevation of world
real interest rates, or a shift toward more labor-saving technology, or
some combination of both-are major forces behind the rise of the
natural rate of unemployment in Europe since 1980. But the failure of
capital's share to show a cumulative rise in the Anglo-Saxon countries
encourages the inference that if the rise in world real interest rates did
not push up capital's share in the United States and the United King-
1. Phelps and Zoega (1996).

dom, it did not push up the unemployment rate in those countries-or

in continental Europe either. Similarly, if global technological change
did not develop a new labor-saving bias that changed capital's share in
the United States and the United Kingdom, then it did not push up
unemployment there-or in continental Europe. Since I believe that the
world real interest rate is a very consequential factor in the United
States, and I also incline toward the view that a new or increased bias
in technological change is also a factor throughout the West, I need
now to deal with these bothersome facts.
First, a rise in capital's share is no smoking gun and it would not
necessarily be visible-or visible forever-after an increase in the
world real interest rate. In Blanchard's model, for example, where
profits are a return only on physical capital, a permanent rise in capital
share would result from the rise of the real rate of interest only if the
substitution elasticity was less than one. And it could very well be that
most Continental countries have a much lower elasticity of substitution
than does the United States. The paper's estimates of the substitution
elasticity are all over the place, ranging from 3.45 in the Netherlands
to 0.48 in Italy. Furthermore, it is clear that after world real interest
rates rose in 1981, the profit share stopped falling and strongly reversed
direction, even if it did not ultimately post a very large net cumulative
rise since 1970.
Second, capital's share is not driven by one factor alone. It is subject
to many influences, certainly to those of the real exchange rate as well
as the world real interest rate. Currencies were weak in Europe relative
to the dollar in the 1980s and they are somewhat weak again in the late
1990s. A strong dollar operates to reduce the markups set by U.S. firms
that produce tradables. So, the fact that capital's share is not booming
in the United States as it is in continental Europe is not cause to reject
the hypothesis that the elevation of real interest rates has contributed to
the unemployment problem.
Third, I suspect that the huge shift in the educational composition of
the labor force plays a role here too. Suppose that the thinning out of
workers on the lower rungs of the educational ladder has a beneficial
effect on quit rates, absenteeism, shirking, and so forth. Such an effect,
in turn, must operate to narrow the margins between price and unit cost
that are needed to cover turnover training outlays in replacing workers
who quit and the costs of monitoring and supervising workers who are
otherwise likely to lie down on the job or worse. Thus on this account,
the educational broadening in the American and British labor forces,
taken alone, must have raised labor's share and lowered the natural
rate. This development was not so strong in the Continental countries;
in Italy, for example, the differentials in unemployment rates between
educational groups are not so pronounced, or even go the wrong way.
It may also be true that desk jobs do not need as much physical capital
per worker as do production line jobs.
Fourth, the greater number of American workers with some educa-
tional credentials certainly increases the wage bill in the economy, since
the employer will pay these an educated worker's wage on the prospect
of recovering the cost when their wider knowledge and greater versa-
tility are needed. But the eventual payoff from their greater "promise,"
which justifies the wage premium, may be a long time in coming.
Although wages go up, productivity may increase only much later, after
the introduction of a new technology, a reorganization of the workforce,
or entry into a new market-that is, when the workers' knowledge and
versatility are really needed. On this account, then, labor's share may
grow for some time, as the fraction of the labor force that is paid on its
promise keeps rising. The fact that capital's share does not rise in the
United States and the United Kingdom therefore is not evidence that
the influence of world real interest rates on capitals share is absent or
weak.
To conclude, I do find this paper quite stimulating. In the end,
though, it leaves me wiser but in the same place.
General discussion: Robert Gordon was doubtful of Phelps's view that

the reduction of the U.S. unemployment rate in recent years largely
reflects a change in the relative importance of educational groups, and
that the aggregate rate masks a substantial increase in unemployment,
controlling for education. He asserted that it would require an implau-
sibly large compositional change to explain the fall in the unemploy-
ment rate, and argued that if Phelps were correct, productivity should
have risen by a corresponding amount, which it has not. Gordon re-
called a Brookings conference organized by Charles Schultze a decade
ago that identified the minimum wage, unemployment insurance, and
other institutional rigidities as culprits of Europe's high levels of un-
employment, and he saw these factors explaining the difference be-
tween the Continental and the Anglo-Saxon experiences. Both sets of

countries have been exposed to globalization and shifts of technology,
but the labor market structure of the Continental countries has delivered
lower inequality and higher unemployment. He also suggested that
labor market differences help to explain why there are more low-wage
workers in the United States, and hence why service sector productivity
in the United States is relatively low. With a plentiful supply of low-
cost workers, firms have little incentive to substitute capital for labor.
Gordon cited anecdotal evidence from Europe-the absence of grocery
baggers in stores, busboys in restaurants, and attendants in parking
lots-that accords with this explanation. Nordhaus observed that the
huge difference between continental Europe and the United States in
rent-seeking institutions in the labor market is another important reason
for differences in labor market reaction to shocks.
N. Gregory Mankiw believed that the paper should have paid more
attention to the possible role of skill-biased technical change in explain-
ing labor market performance. Although such change may have been
common across countries and may have caused a decline in the demand
for unskilled workers in all of them, the effect on employment and
wages should reflect the structural differences emphasized by Gordon.
In the United States, unskilled wages have gone down, whereas the
welfare states of Europe have provided a higher floor to unskilled
wages. As a result, European countries experience great unemployment
and the United States experiences increased inequality. Mankiw noted
that this explanation relies on only a single shift and one institutional
difference, instead of the multiple shifts discussed in Blanchard's
paper.
Robert Hall observed that the different experiences of Europe and
the United States highlight the importance of different theoretical
models. Search models, which explain how workers who have lost jobs
become reemployed, are essential for understanding the U.S. economy,
whereas models explaining the determination of the efficiency wage are
essential for understanding the Continental experience.
References
Acemoglu, Daron. 1997. "Why Do New Technologies Complement Skills'?
Directed Technical Change and Wage Inequality." Unpublished paper. Mas-
sachusetts Institute of Technology (August).
Blanchard, Olivier J. 1996. "Notes on Profit Shares in the OECD." Unpub-
lished paper. Massachusetts Institute of Technology (December).
Blanchard, Olivier J., and Lawrence F. Katz. 1997. "What We Know and Do
Not Know About the Natural Rate of Unemployment." Journal of Economic
Perspectives 11(1): 51-72.
Blanchflower, David G., and Andrew J. Oswald. 1994. The Wage Curve. MIT
Press.
Bruno, Michael, and Jeffrey D. Sachs. 1985. The Economics of Worldwide
Stagfiation. Oxford: Basil Blackwell.
Caballero, Ricardo J., and Mohamad L. Hammour. 1998. "Incomplete Con-
tracts, Factor Proportions, and Unemployment." Carnegie-Rochester Con-
ference Series on Public Policy 48 (forthcoming).
Cette, Gilbert, and Selma Mahfouz. 1996. "Le partage primaire du revenu: un
constat descriptif sur longue periode." Economie et Statistique (6-7):
165-89.
Cummins, Jason G., Kevin A. Hassett, and R. Glenn Hubbard. 1994. "A
Reconsideration of Investment Behavior Using Tax Reforms as Natural Ex-
periments." BPEA, 2:1994, 1-74.
Dickens, William T. 1995. "Do Labor Rents Justify Strategic Trade and In-
dustrial Policy?" Working Paper 5137. Cambridge, Mass.: National Bureau
of Economic Research (May).
Dougherty, Chrys, and Dale W. Jorgenson. 1996. "International Comparisons
of the Sources of Economic Growth." American Economic Review, Papers
and Proceedings 86(2): 25-29.
Dreze, Jacques H., and Charles R. Bean, editors. 1990. Europe's Unemploy-
ment Problem. MIT Press.
Fortin, Nicole M., and Thomas Lemieux. 1997. "Institutional Changes and
Rising Wage Inequality: Is There a Linkage?" Journal of Economic Per-
spectives 11(2): 75-96.
Hall, Robert E. 1990. "Invariance Properties of Solow's Productivity Resid-
ual." In Growth, Productivity, Unemployment, edited by Peter Diamond.
MIT Press.
Hammermesh, Daniel S. 1993. Labor Demand. Princeton University Press.
Houthakker, H. S. 1956. "The Pareto Distribution and the Cobb-Douglas
Production Function in Activity Analysis." Review of Economic Studies
23(1): 27-31.
Jackman, Richard, Richard Layard, and Stephen Nickell. 1996. "Combatting
Unemployment: Is Flexibility Enough?" Paper prepared for the conference
"Interactions between Structural Reform, Macroeconomic Policies and Eco-
nomic Performance." Organisation for Economic Co-operation and Devel-
opment, Paris, January 18-19.
Johnson, George. 1990. "Work Rules, Featherbedding, and Pareto-optimal
Union-Management Bargaining." Journal of Labor Economics 8(1, part 2):
S237-59.
Laffargue, Jean Pierre, and Anne Saint-Martin. 1997. "Une simulation retro-
spective d'un modele WS-PS pour la France de 1974 'a 1993." Unpublished
paper. Paris: Cepremap and Eureqa (August).
Layard, Richard, Stephen Nickell, and Richard Jackman. 1991. Unemploy-
ment: Macroeconomic Performance and the Labour Market. Oxford Uni-
versity Press.
Nickell, Stephen. 1997. "Unemployment and Labor Market Rigidities: Europe
versus North America." Journal of Economic Perspectives 11(3): 55-74.
Phelps, Edmund S. 1994. Structural Slumps: The Modern Equilibrium Theory
of Unemployment, Interest, and Assets. Harvard University Press.
Phelps, Edmund S., and Gylfi Zoega. 1996. "The Incidence of Increased
Unemployment in the Group of Seven, 1970-1994." Centro Studi Paper
115. Rome: Centro Studi Confindustria (September).
Poterba, James M. 1998. "Recent Developments in Corporate Profitability:
Patterns and Explanations." Carnegie-Rochester Conference Series on Pub-
lic Policy 48 (forthcoming).
Rotemberg, Julio J., and Michael Woodford. 1991. "Markups and the Busi-
ness Cycle." In NBER Macroeconomics Annual 1991, edited by Olivier J.
Blanchard and Stanley Fischer. MIT Press.
. 1998. "The Cyclical Behavior of Prices and Costs." In Handbook of
Macroeconomics, edited by John Taylor and Michael Woodford. Amster-
dam: North-Holland (forthcoming).
Zeira, Joseph. Forthcoming. "Workers, Machines, and Economic Growth."
Quarterly Journal of Economics.

The Medium Run

Uploaded by

Copyright:

Available Formats

The Medium Run

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Medium Run

Uploaded by

Copyright:

Available Formats

OLIVIER J.

The Medium Run

MACROECONOMICS iS largely divided into two subfields. One focuses

Figure 1. Capital Shares, Continentaland "Anglo-Saxon"Countries, 1970-96a

1975 1980 1985 1990 1995

at high levels today. The average share stood at 41 percent in 1995, up

3. Brunoand Sachs (1985).

countries of the Organisation for Economic Co-operation and Devel-

stage of the research to try to explain the evolution of each individual

Factor Prices and Factor Quantities

Movements in the capital share, such as those documented in fig-

where n is labor, k is capital, and z is the level of productivity. Rede-

7. For Germany,OECD data referto West Germanyup to 1990 and to Germanyas

1.00 Profit/wage 1.00 _ Labor/capital

1970 1980 1990 1970 1980 1990

Table 1. RegressingFactorPrices on Factor Quantities, 1961-95a

Together, table 1 and figure 3 confirm the visual impressions given by

Figure 3. Time Variationin Factor Price Ratios, Controllingfor Factor Quantity

Log index, 1970 = 0

1965 1970 1975 1980 1985 1990 1995

Source:Author'sregressionsbased on data used in figure 1.

largely spurious fluctuations in measured total factor productivity, and

and firms generically as changes in the relation between the marginal

wage need not be equal to the marginal product of labor.'5 In models

Shifts, Capital Shares, and Unemployment

The purpose of this section is to build a simple model that both

(3) y = f(n, 1).

Firms adjust the ratio of labor to capital in response to the present

(4) f,,(n*, 1) = L we.

Note that from the point of view of determining employment, a higher

where h is a parameter and K denotes the capital stock. Free entry

dv = (r + 8)v - [ - (c) ()2]

Profit per unit of capital is equal to the user cost.

Functional Forms and Parameters

(8) y = A (1 - a)n + aJ_-

Shifts in Labor Supply, Changes in the Distribution of Rents, and

Table 2. Parameters and Steady-State Values

ADVERSE SHIFTS IN LABOR SUPPLY. Figure 4 shows the effects of an

Figure4. DynamicEffects of an Adverse Labor Supply Shifta

Profit rate Wage rate

Figure 5. DynamicEffects of an Increasein the Markupa

markup,startingfroman initialvalue of 1.0 (thatis, thereareno distortionsat the start).

Figure 6. Dynamic Effects of Biased TechnologicalChangea

Profitrate ," -- Wagerate

0.55 ---- 0.90

Source:Author'scalculations,as describedin text.

coefficient a, from 0.3 to 0.4.27 It is striking how much figure 6 looks

Looking at Evolutions in Specific Countries

Constructing the Shifts

27. In changing relative marginalproducts, one must be careful not to introduce

behind them.29 The slowdown in productivity growth starting in the

(10) Inwi + In F = constant + ln(1 - a) - ( )ln(n).

The right-hand side gives the logarithm of the marginal product of

29. See, for example, Nickell (1997) for a recentassessment.

This equation is only correct, however, in the absence of costs of

and A is a parameter reflecting the speed of adjustment of factor pro-

Figure 8. Labor Demand Shifts under AlternativeElasticityand AdjustmentCost

Log markupindex, 1970= Oa

Source: Author's calculations (see figure 1).