Transcript Discussion of Nordhaus on Alchemy and the New Economy

Discussion of Kahn-Rich on
Tracking the New Economy
Robert J. Gordon
Northwestern University and NBER
Federal Reserve of San Francisco,
November 7, 2003
Productivity Growth is the Hot Topic
in Macroeconomics






Look at the numbers, 6.8 in 2003:Q2, 8.1 in 2003:Q3
How much of this is an unusual cyclical event? How
much of this is an acceleration of trend?
Advertisement: Read another paper, “Exploding
Productivity Growth: Historical Context, Possible Causes,
Future Implications” BPEA 2003:2, forthcoming
Brookings disallows NBER Yellow-covered Papers
Will be on my web site by Monday morning. Search
Google (not that Scottish University).
97 double-spaced pages
Two New Ideas in Kahn-Rich Paper


Idea #1. Productivity growth shifts
between regimes. “High” and “Low”
productivity growth.
Idea #2. We can do better in estimating
the trend of productivity growth by using
outside information going beyond
productivity growth itself


Info A: Real compensation per hour
Info B: Real consumption per hour
Does their Approach Signal
Increase in late 90’s Trend Earlier
than Other Methods?

Their claim:



“One could not decisively conclude that there
was a return to a higher growth regime on
the basis of productivity alone.”
“Only the corroborating evidence from other
cointegrated series can swing the balance
strongly in favor of a regime switch.”
Does Their Series Detect Post-1995
Acceleration Faster than Alternatives?
Motivations for the Kahn-Rich
Reevaluation of Growth Trends

One can be skeptical about both ideas

“We treat that trend as a stochastic process whose
mean growth rate has two `regimes’”.




Why two?
Why not high medium low?
Why not four regimes? Ten regimes?
Use two additional series to provide additional
information

Skeptic: Why do these two series provide
independent information? Why do the authors not
take us through the arithmetic of labor’s income
share?
Additional justifications for their
approach

Criterion for use of two additional variables,
consumption and real wage:

#1 “We show that aggregate productivity data alone
do not provide as clear . . . a signal of changes in
trend growth as does the joint signal from the series
we examine.”


#2 We do not have to choose break dates


What aspect of time-series dynamics provide an additional
contribution from those two series?
Nor do H-P filter nor Kalman time-varying coefficient
#3 How long regimes last (contingent on only two
regimes?)
Arithmetic of Labor’s Share



Does their real compensation variable provide
additional information?
S = (WH/PY)
Change in log labor’s income share



Δs = Δw – Δp – (Δy – Δh)
Usual cyclical behavior, labor’s share rises in
recessions, shrinks in recoveries (like now).
In using compensation per hour as a proxy for
the productivity trend, they are making a
statement about the cyclical behavior of labor’s
share. But they have no model.
Consumption/Hour



How is Consumption/Hour related to
productivity?
C/H = (C/Y) * (Y/H)
So now we need a model of the share of
consumption in GDP. Lots of models –
Keynesian, RBC, but not in this paper.
Alternative Productivity Trends to
Compare to Kahn-Rich

#1 Hodrick-Prescott (H-P) filter





Parameter value: meaning
Everyone uses 1600. This is the square of 5/(1/8)
=40. When detrending GDP, a 5 percent GDP gap
causes the trend to decelerate by 1/8 percent per
quarter.
Great Depression: 25 percent GDP gap, trend
decelerates at 5/8 percent per quarter, 2.5 percent
per year, 10 percent after four years.
Starting at 3 percent per year in 1929, trend by 1933
is growing at 3 – 25 percent or -22 percent per year.
H-P parameter 6400 is much better than 1600
#2 Kalman Filter with Time-Varying
Coefficients




The Kalman filter explains the change in productivity
growth (Δpt) by a time-varying constant and any set of
other explanatory variables (βXt):
(5) Δy - Δh(t) = α(t) + βX(t) + w(t)
The next step is to specify a time-series process for the
time-varying productivity trend, and the most
straightforward is a random walk:
(6) α(t) = α(t-1) + v(t)
Implementation of Kalman Filter
Productivity Trend with TimeVarying Coefficient





Replace βX(t) by change in output deviation
from trend (or change in unemployment rate).
Could also use supply-shock variables, e.g.,
change in oil prices, that cause temporary
changes in productivity
Kalman filter can use more info than H-P
Compare K-filter with H-P filter
Without output variable they are identical, same
smoothness coefficient
Kalman with Output vs. H-P
Figure 2. H-P and Kalman Trends for Productivity Growth, 1955-2003
Kalman with Q
3.5
3
Percent
2.5
2
1.5
1
0.5
HP 6400
Kahn-Rich Compared
Figure 2. H-P and Kalman Trends for Productivity Growth, 1955-2003
Kalman with Q
3.5
3
Percent
2.5
2
1.5
1
HP 6400
Kahn-Rich
Summary: Problems with their
Productivity Growth Trend



Problem #1. Why is it so Jagged?
Problem #2. Why must there be only two
regimes?
Problem #3. Why does the behavior of the real
wage tell us something about the productivity
trend?


What if super-normal productivity growth goes into
profits? How long does it take for real wage growth
to catch up?
Problem #4. Why does consumption tell us
anything about the productivity trend. What is
the theory of the behavior of the C/Y ratio?
Thanks to the Authors



For bringing the Productivity Growth Trend to
the Attention of this Audience one day after the
announcement of 8.1
For helping to focus attention on the strengths
and weaknesses of Hodrick-Prescott and Kalman
And for reminding us that there is a lot of
macroeconomics devoted to explaining the
change in labor’s income share and the C/Y ratio