The Evolution of Okun’s Law and of
Cyclical Productivity Fluctuations in
the United States and in the EU-15
Robert J. Gordon
Northwestern, NBER, CEPR, OFCE
Seminar Presentation at OFCE
Paris, September 14, 2011
Themes of Paper with Broader
Implications

Procyclical productivity shocks are not a
fundamental object in macroeconomics; they
are residual artifacts of lagging hours
– Productivity’s lead does not prove causation
– The productivity residual varies across time and
places (US vs EU) as a result of labor-market
institutions.
– Procyclical productivity fluctuations have nothing to
do with technology, and the phrase “technology
shocks” should be banished from
Further Broad Themes


EU vs. US differences in cyclicality of
productivity reflects American Exceptionalism
and European institutions
Among these are
– Rising inequality in US, shift in power from labor to
management
– Much less increase in EU inequality (outside of
UK), corporatist constraint on executive
compensation. Germany management and union
agreement on work sharing and kurzarbeit.
– Long history of EU corporatism and government
policy to share work by reducing hours
Changes in Cyclical Labor-market
Behavior

Point of Departure: Okun’s Law (1962)
– In response to a 1 percent change in the output gap,
procyclical responses of hours 2/3, of which employment
1/3, LFPR 1/6, hours/employment 1/6
– Procyclical productivity fluctuations make up remaining 1/3

US study: A new approach to detrending data
– Contrasts H-P and Kalman filters
– uses “outside information” from inflation to
determine the unemployment rate gap
– Feeds U gap into Kalman filter to eliminate cyclical
component of trend in output and aggregate hours.
Other New Findings:
Unconventional US Data and
Analysis of EU-15 Quarterly Data




For US only: a new approach to data
– US: Total Economy not NFPB Sector
– US: Conventional vs. Unconventional Measures
For EU, an attempt to create quarterly data for EU-15 aggregate that
duplicate those long available for US
– First quarterly data series back to 1963 on employment and output
with consistent aggregation
– Merged post-2000 quarterly hours with earlier annual hours data to
create a series back to 1970
Main finding: in the US, productivity no longer exhibits procyclical
fluctuations. But in the EU, productivity is still procyclical by about the
same amount as before.
A key finding for the US: hours gap ~= output gap in 2008-09. But as
predicted by Okun’s Law, hours gap < output gap in US in 1980-82 and in
EU-15 for 08-09.
Preview of Substantive Hypotheses
to Explain Changes
Joint explanation of US and EU behavior
based on “American Exceptionalism”
 US shift toward greater labor input response
is explained by the “Disposable Worker”
hypothesis

– Increased managerial power, new emphasis on
maximizing shareholder value, decreased power of
labor groups and employees
Preview of Substantive Hypotheses
(Continued)


Europe has not experienced a parallel shift in market
power between labor and management
Also, several important EU countries have developed
institutions and policies that explicitly or implicitly
restrict the responsiveness of labor to output changes,
e.g. “work sharing”
– These policies shift the impact of output changes from
employment level onto hours per employee and
consequently output per employee
– Work sharing in Europe in the form of shorter hours per
employee does not show up in our results
Simple Version and Conventional U. S.
Data Version for the Total Economy
Y H E L
Y
   N
H E L N
P
P
P
H
Y H E E L
Y  P P H
 N
H E E
L N
P
Introducing the Alternative
“Unconventional” Identity

Nalewaik’s 2010 Brookings Paper:
–
–
–
–

GDP and GDI are conceptually identical
But they differ (statistical discrepancy)
GDI is more procyclical
When GDP is revised, it tends to be revised toward what GDI already
shows
Hours
– All existing work uses hours based on payroll employment
– There is a little-known series on hours based on the household survey

In principle 2 numerators, 2 denominators = 4 possible
productivity measures, here we simplify by comparing only two
combinations, Conventional and Unconventional
Conventional Compared to
Unconventional Identity
P
Y
P
P
P
H
Y
H
E
E
L
 P P  H 
 N
H
E
E
L N
I
H
H
Y
H
E
L
Y  H  H 
 N
L N
H
E
I
First Part of Paper:
Detrending the
Full-Period US Data

Uses Kalman detrending, which allows use of an outside
feedback variable.
– Avoids excessive cyclicality of H-P trends
– For this outside information, turn to a technique for estimating the
time-varying natural rate of unemployment (TV-NAIRU)

When possible, we prefer to use Kalman over HP or Bandpass
filters, which use only a univariate series to detrend itself.

Last part of paper: study of US vs. EU redoes US results
in a restricted format to match data availability for EU.
Uses H-P filter as a stopgap variable for both US and EU.
TV-NAIRU: The Kalman Feedback
Variable

The TV-NAIRU provides the outside information for the Kalman
trends
– Calculated through an established procedure of regressing the
growth of inflation on lagged inflation, unemployment, and supply
shock variables.




The standard NAIRU is too volatile to plausibly represent trend
employment, so we take a 20-quarter centered moving average
The time period of 2008-11 distorts the entire NAIRU, because there
is a large output gap without a steady decline in core inflation
We cut off the NAIRU at its 2005:Q1 value of 5.0 percent
This is consistent with the subsequent decision to cut off changes in
estimated trends at 2007:Q4 in both the U. S. and Europe
The Unemployment Gap,
Fed Back to the Kalman
Trend
Figure 2. Actual Unemployment Rate, Time-Varying NAIRU, and Implied
Unemployment Gap,
1955:Q1 - 2011:Q2
12
Unemployment Rate
10
8
TV NAIRU
Percent
6
4
2
0
Unemployment Gap
-2
-4
1955
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
The Wild Differences in
Hours Trends
Figure 1a. Annualized Trend Growth Rate of Conventional Aggregate Hours,
Alternative Methods, 1955:Q1 - 2011:Q2
3
HP(1600)
2
Percent
HP(6400)
Kalman 07:4
1
Kalman
11:2
0
-1
-2
1955
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
Equally Sharp Differences
in Output Trends
Figure 1b. Annualized Trend Growth Rate of Conventional Output,
Alternative Methods, 1955:Q1 - 2011:Q2
6
HP(1600)
Percent
5
4
Kalman
07:4
3
2
HP(6400)
Kalman
11:2
1
0
1955
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
Alternative Productivity
Trends
Special Problem
Posed by 2008-11 Cycle




Hours and employment gaps for the US respond
roughly as much as output gap
The unemployment gap drives the trend adjustment
which treats the entire post-1954 interval as
homogeneous
Estimated through 2011:Q2, the Kalman procedure
thinks that the long term trend hours growth must be
implausibly low to generate the observed decline in
hours
We avoid making judgments on 2008-11 cycle by
constraining all growth trends as equal to 2007:Q4
growth rates throughout 2008-11
Conventional (C) vs.
Unconventional (U):
Medium-run Growth Trends





Major findings in Table 1
The mysterious upsurge in LP growth 2001-07 in C
data does not exist in U data
Big differences in AAGR of LP growth
Conventional 96-01: 2.11, 01-07: 2.10
Unconventional 96-01: 2.33, 01-07: 1.24
– This supports the view that the late 1990s US
productivity revival was a one-shot event, not a
permanent change in the trend
Kalman Trends: Conv vs.
Unconv Output & Hours
Unconventional Productivity:
New Story for 1994-2007
Basis of U. S. Analysis:
Kalman Trends for
Conventional Definitions
Continuing with U. S.-Only Data:
What We Learn from Cyclical
Deviations from Trend (“Gaps”)

The most interesting results
– Okun’s 2/3 hours vs. 1/3 productivity result worked
perfectly in late 1960s and early 1980s but at
almost no other time


The 2008-09 cycle has been as big for hours
than for output, while 1980-82 saw a much
larger response of output than hours
Correlation of productivity gap with output gap
changes timing and disappears after mid-1980s
US: Gaps for C & U Average:
Output, Hours, Productivity
US: Gaps for Three Components
of Aggregate Hours
Regression Analysis for
US-Only 1954-2011
• All variables expressed as QUARTERLY FIRST DIFFERENCES OF
DEVIATION FROM TREND, i.e. Δ log gap in X
• Changes in gaps for output identity components explained by
–
–
–
–
–
Changes in output gap (current, four lags, and four leads)
Lagged dependent variable (lags 1-4)
Error correction term
Interactive dummy on output gap (for full period)
End-of-expansion dummies
(7) xt 
4
 x
i 1
i
t i

4
  y'
j  4
j
8
t j
 xt1   ( y 't 1  xt1 )   k Dk  t
k 1
End of Expansion Variable

End of Expansion Effect
– Productivity slows late in expansion, hours grow too
fast (“overhiring”)
– Constrained to be completely offset by faster
productivity growth early in recovery (“Early Recovery
Productivity Bubble”

Implementation
– Not 0,1 dummies. They enter in the form 1/M, -1/N
– M is the number of quarters in late expansion, N in
early recovery
– These sum to zero
Error Correction Term

Error Correction Term
– Linked to the concept of cointegration
 Informal Definition: A linear combination of two series is
stationary. See Engle and Granger (1987)
– A regression using only differenced data is misspecified, and one
using only levels will ignore important constraints.
– Solution: Add an error correction variable to the regression
consisting of the lagged log ratio of the gdp gap to the dependent
variable gap, allowing for separate coefficients on the numerator
and denominator

For further reference on the concept of error correction in
a bivariate model, see Hendry, Pagan, and Sargan (1984)
Regression Results for USOnly, 1954-86 vs. 1986-2011



Hours gap lags output by roughly one quarter
Productivity leads output by roughly two and half quarters
End-of-expansion dummies (8 recessions)
– To simplify tables, constrained to be equal within subsample
– Significant in Hours and LP equations pre 1986 and in full-period
regressions

Split sample: 1954-86 vs 1986-2011
– Big change in long-run responses
– They don’t pass Chow tests, which are too demanding
– Interactive dummy shows a statistically significant change in the sum
of the output coefficients

To simplify paper, regressions are presented only for
conventional concept of hours & LP
– Unconventional data are noisier due to household hours
US: Long-Run Responses,
Before and After 1986
Implications of Regression
Analysis


Okun’s Law is overturned, Hours now
respond by >1 to output deviations, not <1
Productivity no longer responds
procyclically to output fluctuations
– No more Okun’s Law
– No more SRIRL
– No more RBC
– No more procyclical “technology shocks,” i.e.,
productivity fluctuations as exogenous inputs in
DSGE and other modern macro theories
The “Early Recovery
Productivity Bubble”



On average since 1970 LP has grown 1.4 percent
AAGR faster than trend in first four quarters of
recovery
0.00 percent faster in following eight quarters
2002-03 was unusual because fast growth continued
in the subsequent 8 quarters
– (but not in the unconventional data)

EOE effect explains about 2/3 of first four quarters
Actual and Fitted, Early and
Late Equations for Hours
Actual and Fitted, Early and Late
Equations for Productivity
Now We Turn to
Comparison of US and EU




The motivating puzzle, shown on the next slide,
is that conditional on the decline in output, U
rate increased more in U. S. than in EU
To avoid keeping track of 15 countries, we
study only the EU-15 aggregate
Some of data are for Euro Area, not EU-15
Reason horizontal axis is change in output, not
output gap (OECD gaps are implausible)
Relationship of Unemployment
and Output in US and EU



The first graph shows the relationship
between the cumulative growth rates of outpu
and unemployment in the recession.
The US is an outlier: the increase in
unemployment is higher than the decrease in
output would predict
Next Figure: The U.S. 2010 level of
unemployment is not unusually high; 2007
unemployment was unusually low
US vs EU: Cumulative Change of
Output and Unemployment 07-10
US vs EU: Level of Unemployment
vs. Growth of Output
Stripped Down Identity for Comparing US
and European Data
Y H E
Y º× × × ×N
H E N
• No suitable quarterly data (yet) for the EU on
Employment Rate or LFPR
– Y/H: Output per Hour: labor productivity
– H/E and E/N: Components of aggregate hours, the
labor input
– H/E problem, a hybrid between payroll survey and
household survey
Comparing the US and the
EU, Graphs and Regressions
Uses simplified output identity: only two components of
aggregate hours: H/E and E/N
Other differences from full US regression
– No EOE variable (not available for Europe)
– Shorter time period, 1971:Q2 – 2011:Q1
 Early = 1971-1991, Late = 1991-2011
– No outside cyclical variable, so we can’t use
Kalman detrending, instead we use HodrickPrescott filter with a parameter of 6400, running
the trends to 2007:Q4 and extending the trend
growth rates to 2011:Q1
US vs EU: Actual Four-Quarter
Growth of Output
Observations on the Actual
Growth Rates in US and EU
 Change in output growth from 2008-2010 is
nearly identical for EU and US
 Between 1986 and 2006 volatility of output
growth is about the same
 Pre-1986 the US has a much more volatile
business cycle with back-to-back recessions
in 1980 and 1981-82
Observations on the Trend
Growth Rates in US and EU



European productivity trend growth starts high
(4.5 percent), then steadily declines
Difference between the productivity trends is
especially high after 1994, when US LP begins to
increase steeply
The growth disparity after 1994 has its
counterpart in levels: PPP-adjusted productivity
in the EU relative to the US reached 92 percent
in 1995 then slipped back to about 83 percent
US vs EU: Trend Four-Quarter
Growth of Output, Hours
US vs EU: Trend Four-Quarter
Growth of Labor Productivity
US vs EU: Trend Four-Quarter
Growth of H/E, E/N
Now Come Graphs of
Actual vs. Trend Changes



The next two charts show the division
of the hours response between Hours
per employee (H/E) and employment
per capita (E/N).
Notice the lack of a strong difference
between US and EU regarding E/H
A much more visible difference in
response of E/N
EU vs. US Actual vs. Trend
Hours per Employee
EU vs. US Actual vs. Trend
Employment per Capita
Observations on Gaps in US and
EU for the Main Variables




The gap for a variable is the percent log ratio
between actual and trend
A gap is a level variable and thus cumulates
changes
If a variable grows slower than trend quarter
after quarter, the gap keeps becoming more
negative
Can see that the depth of the 2008-2009
recession was virtually identical in US vs. EU
Observations on Gaps in US and
EU (cont.)




Output gaps restate the result that the depth of the
recession of 08-09 was about equal in US and EU
Hours gap is much more negative for the US, with a
trough of -8.7 percent versus -4.4 percent for EU
Leads to a higher productivity gap for the US during
the recession, even briefly going positive
US had an earlier and shorter lived drop in
productivity gap during the recession, even relative to
higher trend productivity growth
Observations on Gaps in US and
EU (cont.)



Discrepancy in US and EU hours gap in the
08-09 recession is almost entirely due to
employment per capita rather than hours per
employee
US E/N gap slipped to almost -8 percent and is
yet to show much recovery, while EU gap
reached only around -5 percent
The lack of difference between the EU and US
on H/E is the most surprising result shown in
the data
US vs EU: Gaps for Output, Labor
US vs EU: Gaps for Productivity
(Output per Hour)
US vs EU: Gaps for Hours per
Employee
US vs EU: Gaps for
Employment per Capita
Overall Differences in US and
Europe Graphs

In EU, hours tend to respond less than in
the U.S. to output changes in the late half of
the data (1991-2011)

Difficult to analyze differences before 1991,
because both output and hours were more
volatile in the US than in the EU during the
first 15 years of the data (1971-1986)
Regression Analysis Europe vs.
US, 1971-2011


Dependent variables: Hours (H), Productivity (Y/H),
Hours per Employee (H/E), and Employment per
Capita (E/N)
Independent variables:
– 4 lags of dependent variable
– Current value, 4 leads, and 4 lags of output
– Error correction term
EU vs US Regression Results




US: early-period response of hours to output is 0.63
EU: early-period response of hours is 0.70. Mean lag
in hours response of 0.21 quarters vs. 0.29 quarter lag
in US
Key Difference: In late interval, the long-run responses
shift in opposite directions, increasing to 1.28 for the
US and decreasing to 0.61 for the EU.
Responses of productivity also shift in the opposite
direction (0.37 to 0.45 for EU, 0.37 to -0.26 for the US)
EU vs US Regression Results
(cont.)




US: the response of both components of hours
increases. H/E from 0.32 to 0.40 and E/N from 0.39 to
1.05.
EU: both responses decrease, H/E from 0.39 to 0.16
and E/N from 0.72 to 0.70
Most of the US increase in H response is due to E/N,
but most of the EU decrease in H response is due to
H/E.
Puzzle: the sums of the responses of H components
are larger than the direct responses of H
US Long Run Responses
EU Long Run Responses
Implications of Regression
Results



In the US there was a distinct shift toward
unitary response of labor input to output
changes, and zero response of productivity as
in the full U.S. regression results.
In Europe there was an opposite shift toward
increased responsiveness of productivity and
decreased responsiveness of the labor input.
We need an explanation for these opposing
trends
Dynamic Simulations




Alternate coefficients from regressions of the early
(1971-1991) and late (1991-2011) periods.
Stripped down regression equation using only the
leads and lags of the GDP gap.
Figure 15 shows the actual levels of H/N, H/E, and
E/N against those predicted by the early and late
period simulations
The late period coefficients predict a much greater
drop in E/N for the US than the early coefficients do
Simulated vs. Actual H/N
Simulated vs. Actual H/E
Simulated vs. Actual E/N
Unified Explanatory Hypothesis:
“American Exceptionalism”



Joint explanation of changes in American and
European behavior
American shifts toward greater labor response
explained by “disposable worker” hypothesis
Europe’s opposite shift explained by the absence of
the conditions of the “disposable worker” idea and
by differing institutions and policies that promote
work sharing.
Substantive Explanation of U.S.
Increased Flexibility of Labor Input




Disposable worker hypothesis
Based on increased managerial
power, diminished worker power
Separate causes at top and bottom
Same set of causes that has been
developed previously to explain rising
U.S. inequality
The U. S. Disposable Worker




Explains both rise in cyclical responsiveness and of
income inequality
Ingredients in increased management power: exec
pay based on stock options, sensitivity to 2000-02 and
2007-09 stock market debacles
Stock options help explain huge increase in share of
top 1 percent 1982-2000 and fluctuating share since
then. Stock option income treated as labor income
Increased emphasis by management on maximizing
shareholder value
Decomposing the Top Decile US Income
Share into 3 Groups, 1913-2008
Not just Strong Management,
Weak Workers

Contributions of weak labor bargaining power
the same list as the sources of increased
income inequality in the bottom 90 percent
– Reduced share of work force was unionized
– Unions became weaker
– Lower real minimum wage
– Globalization: low-skilled immigrants and low-cost
imports
Application of this Hypothesis to
2000-04


2001-03, large employment response and long period of
employment decline (19 months after NBER trough month, Nov
2001)
– Output recovery was so weak that output gap got worse, not
better
– Savage corporate cost cutting (intertwined nexus of
executive compensation, stock market, profit collapse)
– Why did productivity rise so fast? Delayed spillover of ICT
inventions of the late 1990s
 Recall that productivity growth slowed down in the
unconventional data during 2001-04
The savage cost-cutting hypothesis has been validated by
industry cross-section results of Oliner-Sichel-Stiroh (2007)
2008-09 Responses: Similarities and
Differences to 2001-03




Similar: collapse of stock market and corp. profits (bigger than
2000-02)
– Same incentive for savage cost cutting
Different: It was much much bigger
– Output gap widened by 5x as much
– Apocalypse Now: Fear in late 2008 and early 2009 of
another Great Depression
For every deck chair thrown off the Titanic in 2001-02, five
deck chairs were tossed over in 2008-09
Management didn’t just cut labor costs, but also capital
– GDPI declined at annual rate of -32 percent 2008:Q4-2009:Q2
Explanations for EU Behavior

Three broad differences between the
US and Europe offer a point of
departure for developing explanations:
– 1) Different evolution of inequality
– 2) Longstanding European regulations
that protect employment
– 3) Explicit European institutions
encouraging work-sharing and reducing
hours, both in the long run and during a
downturn
Differences in Inequality


The U.S. exhibited a move toward maximizing shareholder
value and cost-cutting. This move has the same causes as the
increasing income inequality in the U.S. as compared to
Europe.
Factors leading to lower European inequality and lower
responsiveness of labor to output:
– Smaller role of short-term profit maximization in management
– Greater power of unions
– Corporatist tradition: unions join with management in making
decisions that ultimately effect labor responsiveness
Differences in Inequality (cont.)





Income share of top 0.1 percent in the US quadrupled from 2 to 8
percent between 1975 and 2000.
Top share in France has remained remarkably stable, increase in
U.K. has been relatively moderate compared to U.S.
Gini Coefficients in 2007: EU Average = 0.31, US = 0.45
Cultural customs and institutions (e.g. traditional role of labor of
German corporate boards) play a large role in determining inequality.
US unions have very little influence over management, leading to
decisions that can cut jobs and make labor much more responsive to
output swings
Top 0.1% Income Shares in the U.S.,
France, and the U.K.,1913-2007
Employment Protection
Legislation



Pre-1980, EU had consistently lower unemployment than US
EU Governments enacted policies that reduced employment per
capita to deal with the hardships of unemployment in 2008 crisis.
Employment Protection Legislation (EPL) – An attempt by EU
governments to directly regulate layoffs
– Outright bans as well as mandated severance packages. This helps to
explain the small shift toward less elasticity in the response of labor to
output swings in Europe.
– Timing question: EPL reached its peak in the early 1990s

Backlash against EPL: After 1995 several EU countries
introduced a flexible second tier of employment
Work Sharing

Legislation and policies by EU countries since 1985 aimed
at cutting work hours instead of firing employees
– Sweden: reduction in hours is aimed at providing parental leave to
parents of both genders
– Netherlands: shift to part-time work to accommodate the cultural
norm that mothers should not work full time
– Germany: hours reductions have been achieved through
corporatist negotiations between employers and unions
– France: switched to a compulsory 35-hour work week
Work Sharing (cont.)


Work sharing in Europe represents a link
to the responsiveness of labor input
– shows that European countries view hours as
an adjustment mechanism to respond to
output changes, while US cost-cutting most
often takes the form of layoffs
{WARNING} – The regressions for H/E in US
vs. EU don’t support this hypothesis
– Behavior in countries outside NL and GE may
dominate EU results
– Big employment responses in Spain and Ireland
Conclusions for Macro


US Changes after 1986
– Okun’s Law is Dead
– Procyclical productivity innovations are dead
– RBC model and “technology shocks” are no longer
relevant as core determinants of business cycles
Europe
– Comparisons are tentative under the absence of a
quarterly labor force series
– Analysis shows that changes in the responsiveness
of labor and productivity have occurred in U. S. but
not in Europe
Conclusions (cont.)

Differences between Europe and the U.S. can be traced to
institutional differences – Institutions Matter.

As a implication of this importance of institutions, the entire
concept of procyclical “technology shocks”, a tenet of modern
macroeconomic theory, becomes obsolete.
– Why should there be procyclical productivity shocks in Europe but not in
the US?

Productivity fluctuations are not an autonomous
representations of random technological change, but an artifact
of the fact that employment and hours respond with different
elasticities and lags to output changes
Further Research
Much remains to be accomplished in this
line of investigation.
 Need a data series on European
employment rate and labor-force
participation
 Need to recognize differences among EU
countries

– Explicit, formalized work sharing programs like
kurzarbeit in Germany are not duplicated in the
other EU nations
– Could form EU-15 sub-aggregates of “north” and
“south” Europe. Are the PIGS or PIIGS different?