Download References - Brad DeLong`s Website

Productivity Growth in the 2000s1
J. Bradford DeLong
U.C. Berkeley and NBER
January 2002
1
I would like to thank Lawrence Summers, Chad Jones, David Romer, Andrei Shleifer,
Dan Sichel, Dale Jorgenson, Hal Varian, and many others for helpful comments and
discussions.
2
I. Introduction
In the early 1970s for reasons that remain largely but not completely a
mystery2 U.S. productivity growth collapsed, producing a two decade-long
age of diminished expectations.3 The period of slow growth continued not
just through the oil shock- and inflation-ridden 1970s, but through the
1980s and early 1990s as well in spite of some confident forecasts that the
apparently high marginal product of capital in the mid-1980s heralded a
return to an era of rapid productivity growth.4 In the mid-1990s for
reasons that appear reasonably well understood5 U.S. productivity growth
accelerated, producing a half decade-long period of exuberance—a “new
economy” —that came to an end with the NASDAQ crash of 2000 and the
recession of 2001-2002. Neither of these sudden sharp changes in
productivity growth was forecast beforehand. This makes my task here—
to forecast the likely pace of productivity growth in the U.S. over the next
decade—doubly hazardous. Not only does my assignment here violate the
cardinal rule of preserving one’s forecasting reputation (that one should
give a number or a date but never both), but the sudden shifts of recent
decades have made all economists’ attempts to forecast medium-run
productivity growth fools’ exercises.
2
See
See Paul Krugman (????), The Age of Diminished Expectations (Washington:
Washington Post: ????).
4
See Blanchard and Summers (),
5
See Steve Oliner and Dan Sichel (), “”
3
3
Outline
I. Introduction
--Purpose of this paper is to lay out what the very simple yet standard
models
of economic growth and business cycles tell us about the likely
course
of productivity growth over the next decade
--Present a menu of models: a productivity-of-computer-capital model,
a
demand-for-high-tech-goods model, a fluctuating-NAIRU-andhighpressure-economy model.
--Also necessary to grapple with measurement issues...
--And very necessary to look at what micro studies have to tell us about
the parameters
of the macro models: because the implications of the macro models
depend critically
on production function parameters, demand elasticities, and other
things rarely
sighted in the real world, it is very important to draw the map
between real world
phenomena and macro conclusions...
--Conclusions: to be added later
II. "Measured" vs. "True" Productivity Growth: Omissions and New
Goods
--Boskin Commission and other studies
III. "Measured" vs. "True" Productivity Growth: The Dot-Com Bubble
--Overinvestment theories of booms
--If the economy is highly productive at making investment goods,
overinvestment
will produce rapid productivity and output growth
4
--But to the extent that we value these investments put into place at
their "real"
value, the boom is significantly smaller. (By how much?)
IV. Computer Capital and Economic Growth
--A model with one final good
--The final good produced by labor, regular capital, computer capital,
and residual
--Take investment in these two types of capital as exogenous
--Moore's Law
--Conclusion: any belief that productivity growth will not be rapid in
the 2000s
rests on a belief that the exponent on computer capital is about to
fall off
a cliff.
V. Demand for High-Tech Goods and Economic Growth
--A model with two final goods
--Technological change produces shifts in the relative prices of these
goods
--Demand elasticity is key to the productivity growth trend: is the
elasticity
of demand for the goods whose prices are falling greater than one
or
less than one?
--Assessing the demand elasticity for high-tech goods with rapidly
falling prices.
--Conclusion: any belief that productivity growth will not be rapid in
the 2000s
rests on a belief that the high-tech sectors have close to exhausted
their
set of potential uses.
VI. Can We Find a Model in Which Productivity Growth in the 2000s
Will Be Slow?
5
--Okun's law and the falling NAIRU: how much of rapid productivity
growth
can be attributed to simply a high-pressure economy?
--The dot-com bubble--causing overinvestment, and artificially
generating a
"false" high value for the elasticity of demand for high-tech goods.
--Learning about productivity growth: maybe the NAIRU will rise
rapidly.
--Conclusion: if everything goes wrong at once--if workers' wage
growth
aspirations rise quickly to return the NAIRU to its pre-1995 level,
if the high-pressure economy unwinds according to Okun's law,
and
if "true" elasticity of demand for high-tech goods is relatively lowthen we should expect productivity growth to be at a rate of X
VII. Conclusion
6
I. Introduction
--Purpose of this paper is to lay out what the very simple yet standard
models
of economic growth and business cycles tell us about the likely
course
of productivity growth over the next decade
--Present a menu of models: a productivity-of-computer-capital model,
a
demand-for-high-tech-goods model, a fluctuating-NAIRU-andhighpressure-economy model.
--Also necessary to grapple with measurement issues...
--And very necessary to look at what micro studies have to tell us about
the parameters
of the macro models: because the implications of the macro models
depend critically
on production function parameters, demand elasticities, and other
things rarely
sighted in the real world, it is very important to draw the map
between real world
phenomena and macro conclusions...
--Conclusions: to be added later
The essence of the "new economy" is quickly stated. Compare our use of
information technology today with our predecessors' use of information
technology half a century ago. The decade of the 1950s saw electronic
computers largely replace mechanical and electromechanical calculators
and sorters as the world's automated calculating devices. By the end of the
1950s there were roughly 2000 installed computers in the world: machines
7
like Remington Rand UNIVACs, IBM 702s, or DEC PDP-1s. The
processing power of these machines averaged perhaps 10,000 machine
instructions per second.
Today, talking rough orders of magnitude only, there are perhaps 300
million active computers in the world with processing power averaging
several hundred million instructions per second. Two thousand computers
times ten thousand instructions per second is twenty million. three
hundred million computers times, say, three hundred million
instructions/second is ninety quadrillion--a four-billion-fold increase in the
world's raw automated computational power in forty years, an average
annual rate of growth of 56 percent per year.
There is every reason to believe that this pace of productivity growth in
the leading sectors will continue for decades. More than a generation ago
Intel Corporation co-founder Gordon Moore noticed what has become
Moore's Law--that improvements in semiconductor fabrication allow
manufacturers to double the density of transistors on a chip every eighteen
months. The scale of investment needed to make Moore's Law hold has
grown exponentially along with the density of transistors and circuits, but
8
Moore's Law has continued to hold, and engineers see no immediate
barriers that will bring the process of improvement to a halt anytime soon.
The new economy will have more examples of very high fixed costs and
very low marginal costs. Such a pattern can produce positive-feedback:
rising demand will often produce higher efficiency and higher returns,
drives and lower prices, leading to yet higher demand. The old economy is
driven by negative feedback: rising demand leads to higher prices, which
leads producers when prices rise, to produce more and, consumers to buy
less, which restores and equilibrium at a lower level of demand. By
contrast, in an information economy, In that sense, if the agricultural and
industrial economies were "Smithian," the new economy may well be
"Schumpeterian." But will that make any difference for medium-run
productivity growth?
Past "new economies," past economic "revolutions" have also seen
extraordinary growth in technology, the rise to dominance of new
industrial sectors, and the transformation. The fifty years after the
invention of electricity, 1880-1930, saw an increase in the mechanical
horsepower applied to U.S. industry of perhaps a hundredfold and an
9
enormous increase in the flexibility of factory organization--a rate of
technological progress of more than nine percent per year (David (1990)).
The hundred years from 1750 to 1850, the core of the (technological)
industrial revolution itself, saw British textile output multiply thirtyfold; in
the middle of the eighteenth century it took hand-spinning workers 500
hours to spin a pound of cotton, but by the early nineteenth century it took
machine-spinning workers only 3 hours to perform the same task--a rate of
technological progress of ten percent per year sustained across half a
century (Freeman and Louca (2001)).
These earlier transformations revolutionized their economies' leading
industries and created "new economies": they changed the canonical
sources of value and the process of production. The industrial revolution
itself triggered sustained increases in median standards of living for the
first time, a shift to an manufacturing- and then to a services-heavy
economic structure, changed what people's jobs were, how they did them,
and how they lived more completely than any previous economic shift
save the invention of agriculture and the discovery of fire. The economic
transformations of the second industrial revolution driven by
electrification and other late nineteenth-century general-purpose
10
technologies were almost as far reaching: mass production, the large
industrial enterprise, the continent- and then world-wide market in staple
manufactured goods, the industrial labor union, the social insurance state,
even more rapid sustained increases in median living standards, and the
middle-class society.
How important will the information economy—the sectors and industries
that have extremely rapid productivity growth driven by the enormous and
ongoing technological revolutions in data processing and data
communications—turn out to be? Will this wave of innovation and
technological development have consequences similar to the trio of steam
power, metal forging, and automatic machinery that powered the original
British Industrial Revolution and transformed economies and societies
beyond recognition? Or will it turn out to have a much smaller impact on
long-run economic growth, as did previous leading sectors like civil
aviation, illumination, and chemical engineering—leading sectors that
produced astonishing leaps in productivity in their relatively narrow
sectors, but that had little long-run influence on the structure of the rest of
the economy or the rate of overall productivity growth?
11
Where Did the Post-1995 Productivity Speed-Up
Come From?
Percentage Points per year
Difference in growth (1995-2000 minus 1973-95)
Oliner-Sichel
Output per hour
1.08
BailyLawrence
1.26
Jorgenson et
al
0.92
Contributions from
Capital
IT Capital
Other Capital
Labor Quality
Multifactor Productivity
Computer Sector MFP
Other MFP
0.34
0.44
0.52
0.59
-0.25
0.04
0.72
0.47
0.25
0.59
-0.15
0.04
0.82
0.18
0.64
0.44
0.08
-0.11
0.51
0.27
0.24
The long-run economic impact of the "new economy" is likely to be very
large indeed for two reasons. First, the pace of technological progress in
the leading sectors driving the "new economy" is very rapid indeed, and
will continue to be very rapid for the foreseeable future. Second, the
computers, switches, cables, and programs that are the products of today's
leading sectors are general-purpose technologies, hence demand for them
is likely to be extremely elastic. Rapid technological progress brings
rapidly falling prices.
12
Rapidly falling prices in the contest of extremely elastic demand will
produce rapidly-growing expenditure shares. And the economic salience
of a leading sector--its contribution to productivity growth--is the product
of the rate at which the cost of its output declines and the share of the
products it makes in total demand. Thus unless Moore's Law ceases to
hold or the marginal usefulness of computers and communications
equipment rapidly declines, the economic salience of the data processing
and data communications sectors will not shrink, but grow.
13
II. "Measured" vs. "True" Productivity Growth:
Omissions and New Goods
--Boskin Commission and other studies
14
III. "Measured" vs. "True" Productivity Growth: The
Dot-Com Bubble
--Overinvestment theories of booms
--If the economy is highly productive at making investment goods,
overinvestment
will produce rapid productivity and output growth
--But to the extent that we value these investments put into place at
their "real"
value, the boom is significantly smaller. (By how much?)
15
16
IV. Computer Capital and Economic Growth
--A model with one final good
--The final good produced by labor, regular capital, computer capital,
and residual
--Take investment in these two types of capital as exogenous
--Moore's Law
--Conclusion: any belief that productivity growth will not be rapid in
the 2000s
rests on a belief that the exponent on computer capital is about to
fall off
a cliff.
In the nonfarm business sector—the part of the economy on which
productivity studies typically focus—output per labor hour rose between
1995 and 2000 at 2.5 percent per year, more than double the pace seen in
the preceding quarter century since 1970.6 This acceleration of
productivity growth raises for the first time in a generation the likelihood
of reasonably rapid and broad-based real income growth, if the jump in
productivity growth can be sustained. In the long run, productivity growth
and average income growth must correspond. An era like that of 19701995 in which productivity growth is slow must be, in Paul Krugman’s
(1989) phrase, an “age of diminished expectations.”
The case for attributing this acceleration in productivity growth to the
technological revolutions in information technology is now very strong. If
this attribution is correct, then this reacceleration of productivity growth is
6
Productivity growth starting in the early 1970s was anomalously and unexpectedly
low—a phenomenon called the “productivity slowdown.” It is depressing to note that
17
the most significant macroeconomic consequence of the “new economy,”
and one that all by itself justifies focusing much attention on computers
and communications.
Back before 1995 critics of visionaries who saw the computer as
transforming the world pointed to slow and anemic growth in aggregate
labor productivity. The end of the 1960s saw the American economy
undergo an aggregate productivity slowdown, in which the trend growth
rate of labor productivity fell by more than half. It seemed unreasonable
that what computer visionaries were touting as an extraordinary advance
in technological capabilities should be accompanied by a record-breaking
slowdown in economic growth. As Nobel Prize-winning MIT economist
Robert Solow posed the question, if the computer is so important "how
come we see the computer revolution everywhere but in the [aggregate]
productivity statistics?"
After Solow wrote, productivity performance worsened still further. In the
decade and a half before Solow asked his question in 1987 output per hour
grew at 1.1 percent per year. In the eight years after 1987 output per
worker grew at only 0.8 percent per year.
This "productivity paradox" was sharpened because at the microeconomic
level economists and business analysts had no problem finding that
investments in high technology had enormous productivity benefits. MIT
even now the causes of the productivity slowdown and of its persistence are not well
understood. See Fischer (1988).
18
economist Erik Brynjolffson and his coauthors found typical rates of
return on investments in computers and networks of more than fifty
percent per year. Firms that invested heavily in information technology
and transformed their internal structures so that they could use their new
technological capabilities flourished in the 1980s and 1990s--and their
lagging competitors did not.7
However, as Federal Reserve Board economists Oliner and Sichel (1994)
pointed out in the early 1990s, the then-failure to see the computer
revolution in the aggregate productivity statistics should not have come as
a surprise.8 In the 1970s and 1980s the computer industry was simply too
small a share of the economy and its output was not growing fast enough
for it to have a large impact on aggregate productivity. According to their
estimates, in the 1980s information technology capital—computer
hardware, software, and communications equipment—accounted for 3.3%
of the income earned in the economy, and the price-adjusted information
technology capital stock was growing at only 14% per year. You multiply
these two numbers together to get an estimate of the contribution of the
information technology sector to economic growth: in this case, a
contribution of 0.49% per year.
But beginning in 1992, the American economy began an extraordinary
investment boom. From 1992 to 2000, real business fixed investment grew
at 11% per year, with more than half of the additional investment going
7
8
See Erik Brynjolfsson and Loren Hitt (1996); Erik Brynjolfsson (1993).
An argument developed at greater length in Sichel (1997).
19
into computers and related equipment. And as the information technology
investment boom took hold, productivity growth and growth in real GDP
accelerated as well. Real GDP rose by an average of 3.9% per year
between 1995 and 2000. Nonfarm business measured labor productivity—
measured output per hour worked—grew at 2.7% per year.
20
Initially some economists—most prominent among them Robert J.
Gordon9—doubted that the acceleration in labor productivity growth in the
1990s was anything more than an unsustainable cyclical phenomenon.
Indeed, as Figure 2 shows, labor productivity can exhibit large swings
from year to year: the boom in productivity growth in 1992 was a onetime flash in the pan (although it did give rise to a series of papers on the
“jobless recovery”). The “Morning in America” boom in productivity
growth of 1983-1986 was also not sustained, at least in part because of
high government budget deficits that reduced capital accumulation. What
reason is there to believe that this boom in the second half of the 1990s is
different?
One of the most powerful reasons to believe that this acceleration of
aggregate productivity growth is permanent, and not a flash in the pan,
comes from the underlying growth accounting of the impact of the
9
See Gordon (2000). At one level the differences between Gordon and the others are
simply differences of emphasis: what is “large” enough for us to pay attention to? At
another level, a key difference revolves around how one accounts for the effect of the
business cycle, and what one would expect the effect of a fall in the natural rate of
unemployment to be on potential output. According to Oliner and Sichel’s growthaccounting model, a 2.5 percentage point fall in the natural rate would boost potential
output by the increase in employment times the share of income accruing to labor—a
boost to potential output of perhaps 1.5%. According to Gordon’s more macro-oriented
model, a 2.5 percentage point fall in the natural rate would boost potential output by the
Okun’s Law coefficient of 2.5 times the change in the unemployment rate—a boost to
potential output of 6.2%. I come down on the side of the first approach, largely because I
believe that the effect on potential output could not be as large as Okun’s law suggests
without generating markedly higher inflation, and we know that this decline in
21
information technology revolution. Consider the standard growthaccounting calculations applied to a model with one final good…
Back in the 1980s information technology capital accounted for 3.3% of
income earned in the economy; today according to Oliner and Sichel
(2000) it accounts for 7.0% of income earned. Back in the 1980s the
economy’s stock of information technology capital was growing at 14%
per year; today according to Oliner and Sichel (2000) it is growing at 20%
per year. Multiply these two sets of numbers together to find that the
increase in the economy’s information technology capital stock was
responsible for 0.5% per year of economic growth in the late 1980s, and
for 1.4% per year of economic growth today.10
At this growth-accounting level of analysis, all of these factors are highly
persistent. The rate of growth of the economy’s information technology
capital stock will not slow down rapidly or immediately. For one thing, the
same number of dollars spent on computers or communication equipment
today deliver perhaps three times as much in the way of real useful capital
as they did five years ago because of the extraordinary fall in computer
and communications equipment prices.11 Even simple use of amortization
funds to replace obsolete computers will generate enormous rates of
increase in the capital stock.
unemployment is associated with a fall in the natural rate and has not generated inflation.
But it is not yet conclusively clear that Gordon’s analysis is wrong.
10
Oliner and Sichel’s conclusions are very similar to those of Jorgenson and Stiroh
(2000). Both are backed up and strengthened in a very interesting series of papers by
Nordhaus (2000a, 2000b, 2001) that I have not yet had a chance to fully digest.
11
See Triplett (1999a, 1999b).
22
Moreover, there is every reason to think that the fall in computer and
communications equipment prices will continue. The pace of
technological advance in information technology has been well-described
for three decades by what has come to be called "Moore's Law"--the rule
of thumb that Intel cofounder Gordon Moore's set out a generation ago
that the density of circuits we can place on a chip of silicon doubles every
eighteen months with little or no significant increase in cost. Moore's Law
has held for thirty years; it looks like it will hold for another ten at least.
Moore's Law means that today’s computers have 66,000 times the
processing power of the computers of 1975. It means that in ten years
computers will be approximately 10 million times more powerful than
those of 1975 at roughly constant cost. The installed base of information
processing power has increased at least million-fold since the end of the
era of electro-mechanical calculators in the 1950s. Such extraordinary
increases in productivity in data processing and data communications
equipment manufacture have the potential to have a large impact on
overall productivity growth as long as the share of total income
attributable to computer capital does not collapse.
Will the share of total income attributable to computer capital collapse?
Probably not. One might wonder whether rapid improvements in a
particular branch of industry will rapidly run into diminishing returns. The
first candlepower of light one can produce after dark--with a candle or an
oil lamp steady enough to read by--is a really big deal. The ten-thousandth
is not. The share of total income attributable to computer capital will
23
remain constant only if the productive value of the marginal computer
declines no more rapidly in percentage terms than the total computer
capital stock increases. In theory there is no reason that the productive
value of the marginal computer might not decline very rapidly indeed.
In theory the marginal returns to investment in computers could diminish
very rapidly. In practice this seems very unlikely to be the case. As John
Zysman has pointed out, one thing that makes computers likely to fit
Bresnahan and Trajtenberg’s (1995) definition of a true engine of growth,
a true general purpose technology, is that each sequential fall in the price
of computers has been accompanied by an exponential increase in the
demand for computers because it makes feasible a whole new set of
capabilities and uses.
24
V. Demand for High-Tech Goods and Economic
Growth
--A model with two final goods
--Technological change produces shifts in the relative prices of these
goods
--Demand elasticity is key to the productivity growth trend: is the
elasticity
of demand for the goods whose prices are falling greater than one
or
less than one?
--Assessing the demand elasticity for high-tech goods with rapidly
falling prices.
--Conclusion: any belief that productivity growth will not be rapid in
the 2000s
rests on a belief that the high-tech sectors have close to exhausted
their
set of potential uses.
An alternative approach would be to look not at a model with one final
good produced by labor, non-IT capital, and IT capital, but to consider a
model with two final goods—IT products and non-IT products—in which
the underlying dynamic of Moore’s Law produces a sharp ongoing fall in
the relative price of IT products. What then determines whether
productivity growth accelerates or not as the technological revolution in
the leading sector proceeds?
If total factor productivity growth in the rest of the economy is growing at
a rate R, and if total factor productivity in the leading industries and
sectors is growing at a faster rate L, then total factor productivity growth
in the economy as a whole will be equal to:
25
(1)
 = (L) + (1-)(R)
where  is the share of total expenditure on the goods produced by the
economy’s fast-growing technologically-dynamic leading sectors.
As the process of innovation and technological revolution in the leading
sectors proceeds, we would not expect the leading sector share  of total
expenditure to remain constant. If the goods produced by the leading
sectors are superior (or inferior) goods, the share  will rise (or fall) as
economic growth continues: only if the income elasticity of demand I for
its products is one will changes in the overall level of prosperity leave the
leading sector share unchanged. If the goods produced by the leading
sector have a high (or low) price elasticity of demand, the falls over time
in their relative prices will boost (or reduce) the share of total expenditure
: only if the price elasticity of demand P is one will the fall in the
relative price of leading sector products produced by the technological
revolutions leave the leading sector share unchanged.
Moreover, the leading sector share of total expenditure  matters only as
long as the leading sector remains technologically dynamic. Once the
heroic phase of invention and innovation comes to an end and the rate of
total factor productivity growth returns to the economy’s normal
background level R, the rate of productivity growth in the economy as a
26
whole will return to that same level R and the leading sector share of
expenditure  will no longer be relevant.
Thus five pieces of information are necessary to assess the aggregate
economic impact of an explosion of invention and innovation in a leading
sector:

The initial share of expenditure on the leading sector’s products, 0.

The magnitude of the relative pace of cost reduction, L – R, during
the leading sector’s heroic age of invention and innovation.

The duration of the leading sector’s heroic age of invention and
innovation.

The income elasticity of demand I for the leading sector’s products.

The price elasticity of demand P for the leading sector’s products.
To gain a sense of the importance of these factors, let’s consider a few
simulations with sample parameter values. For simplicity’s sake, set the
initial share of expenditure on the leading sector’s products 0 equal to
0.02, set the income elasticity of demand for the leading sector’s products
I equal to 1.0, set the heroic age of invention and innovation to a period
40 years long, and set the background level of total factor productivity
growth R to 0.01 per year, one percent per year. Consider three values for
the price elasticity of demand P: 0.5, 2.0, and 4.0. And consider two
values for the wedge in the annual rate of technological progress between
the leading sector and the rest: 0.03, and 0.05.
27
With a price elasticity of demand of 0.5, the expenditure share of the
leading sectors declines from its original value of 2% as technology
advances and the prices of leading-sector goods fall. With a productivity
wedge of 5% per year, the initial rate of growth of economy-wide
productivity growth is 1.1% per year—1% from the background growth of
the rest of the economy, and an extra one-tenth of a percent from the faster
productivity growth in the one-fiftieth of the economy that is the leading
sector. By the twelfth year the expenditure share on leading sector
products has fallen below 1.5%. By the twenty-eighth year the expenditure
share has fallen below 1.0%. By the fortieth year the expenditure share has
fallen to 0.7%.
28
The low initial and declining share of the leading sector in total
expenditure means that 40 years of 6% per year productivity growth in the
leading sector has only a very limited impact on the total economy. After
forty years total productivity in the economy as a whole is only 2.54%
higher than had the leading sector not existed at all. Rapid productivity
growth in the leading sector has next to no effect on productivity growth
in the economy as a whole because the salience of the leading sector falls,
and the salience of other sectors resistant to productivity improvement
rises as technology advances. This is Baumol and Bowen’ (1966) “cost
29
disease” scenario: innovations become less and less important because the
innovation-resistant share of the economy rises over time. Indeed, as time
passes the rate of aggregate growth converges to the rate of growth in the
productivity-resistant rest of the economy.
By contrast, with a price elasticity of 4 the expenditure share of the
leading sectors grow rapidly from their original value of 2%. With a
productivity growth wedge of 5% per year, the leading sector share of
spending surpasses 10% by year 12, 30% by year 20, and reaches 89% by
30
year 40. As the spending share of the leading sectors rise, aggregate
productivity growth rises too: from 1.1% per year at the start to 1.4% per
year by year 10, 2.4% per year by year 20, 4.2% per year by year 30, and
5.4% per year by year 40. The impact on the aggregate economy is
enormous: total factor productivity after 40 years is 113% higher than it
would have been had the leading sector never existed.
In these simulations, there is only one reason for the sharp difference in
the effects of innovation in the leading sector: the different price
31
elasticities of demand for leading-sector products in the two scenarios.
The initial shares of leading sector products in demand, the rate of
technology improvement in the leading sector, and the duration of the
technology boom are all the same. But when demand for leading sector
products is price-elastic, each advance in technology and reduction in the
leading sector’s costs raises the salience of the leading sector in the
economy and thus brings the proportional rate of growth of the aggregate
economy closer to the rate of growth in the leading sector itself. By the
end of the 40 year period of these simulations, the scenario with the price
elasticity of 4 has seen the leading sectors practically take over the
economy, and dominate demand. This is the “economic revolution”
scenario: not only does productivity growth accelerate substantially and
material welfare increase, but the structure of the economy is transformed
as the bulk of the labor force shifts into producing leading-sector products
and the bulk of final demand shifts into consuming leading-sector
products.
32
33
VI. Can We Build a Model in Which Productivity
Growth in the 2000s Will Be Slow?
--Okun's law and the falling NAIRU: how much of rapid productivity
growth
can be attributed to simply a high-pressure economy?
--The dot-com bubble--causing overinvestment, and artificially
generating a
"false" high value for the elasticity of demand for high-tech goods.
--Learning about productivity growth: maybe the NAIRU will rise
rapidly.
--Conclusion: if everything goes wrong at once--if workers' wage
growth
aspirations rise quickly to return the NAIRU to its pre-1995 level,
if the high-pressure economy unwinds according to Okun's law,
and
if "true" elasticity of demand for high-tech goods is relatively lowthen we should expect productivity growth to be at a rate of X
Back at the start of the 1990s most macroeconomists estimated that the
economy’s natural rate of unemployment was between 6.5 and 7.0
percent. If unemployment fell below that level, it was argued, inflation
would begin to accelerate. Thus a Federal Reserve that wished to avoid
major recessions by maintaining the public’s confidence in its lack of
tolerance for inflation could not afford to let the unemployment rate fall
below 6.5 percent. These estimates were based on long historical
experience, summarized in Figure 3 which shows the track of inflation and
unemployment in the U.S. economy since 1960. In the 1960s inflation
increased when the unemployment rate fell below 5.5%. In the early
34
1970s, it seemed as though inflation fell when the unemployment rate rose
above 5.5%, but then came the major accelerations in inflation produced
by the oil price shock of 1973, and by the late 1970s it seemed as though it
required an unemployment rate of 6.5% or more to put downward pressure
on inflation.
35
In the 1980s, the workings of the labor market seemed worse: only when
unemployment rose above 7% in the early 1980s did inflation fall
noticeably. And in the late 1980s and early 1990s it seemed as though
inflation rose whenever the unemployment rate fell below 6.5%, and fell
36
when the unemployment rate rose above 6.5% percent.12 Just about the
time in the mid-1990s when the aggregate rate of productivity growth
began to boom, the comovements of inflation and unemployment went off
track. The fall in unemployment to 6% in the mid-1990s did not lead to
any acceleration in inflation, nor did the fall in unemployment to 5% and
then 4.5% in the late 1990s. Only in the last year and a half or so, as the
unemployment rate has fallen to 4%, have there been any signs of rising
inflation. In the early 1970s most macroeconomists thought the NAIRU
was in the range of 5 to 5.5 percent. By the early 1990s most
macroeconomists thought the NAIRU was in the range of 6 to 7 percent.
So nearly all macroeconomists have been surprised by the stunningly swift
fall in the NAIRU down to somewhere in the neighborhood of 4.5 percent
by the late 1990s.
12
For a more formal econometric analysis of the time-varying natural rate of
unemployment—one that stresses the uncertainty surrounding our estimate of the natural
rate at any moment in time—see Staiger, Stock and Watson (1997).
37
To what extent might the productivity boom of the late 1990s be the result
of the fall in the NAIRU? According to Okun’s Law a 2 percentage point
fall in the unemployment rate would be linked to a five percentage point
rise in the level of output relative to potential output—enough to by itself
drive a one percentage-point acceleration of economic growth over a fiveyear period.
38
It is, however, possible that the natural rate of unemployment is linked to
the rate of economy-wide productivity growth. The era of slow
productivity growth from the mid-1970s to the mid-1990s saw a relatively
high natural rate. By contrast, rapid productivity growth before 1973 and
after 1995 has been associated with a lower natural rate. If workers'
aspirations for real wage growth themselves depend on the rate of
unemployment and do not depend directly on productivity growth, then a
speedup in productivity growth will reduce the natural rate.
If productivity growth is relatively slow, then a low rate of unemployment
will lead workers to demand high real wage increases—real wage
increases above the rate of productivity growth. But firms cannot
continuously grant real wage increases higher than the rate of productivity
growth and still remain profitable. Long before their profits disappear they
will respond to the higher real wage growth aspirations and demands by
economizing on workers. Unemployment will rise until the average
unemployment rate is high enough to curb worker aspirations for real
wage growth to a level consistent with trend productivity growth.
39
With a higher rate of productivity growth, firms can afford to pay higher
real wage increases without going bankrupt. The unemployment rate
consistent with real wage growth aspirations that match productivity is
lower. Hence an economy with higher productivity growth has a lower
natural rate of unemployment.
The attribution of the fall in the NAIRU in the 1990s to the “new
economy”—as an indirect consequence of the acceleration in productivity
growth—is plausible and enticing, but far from proven.
If we look far back in history at the long bull runs of the American stock
market—1890-1910, or 1920-1930, or 1950-1970—we see that for each
10% that the real value of dividends rise over a twenty-year period, the
real value of stock prices tends to rise by 15%. But if we look just at the
most recent bull market—the one that started in 1982—we find that a
market-wide rise in dividends of 10% produces not a 15% but a 26%
increase in stock prices. The runup in stock prices during the 1920s was
extraordinary, but in real terms the increase in dividends paid out in the
1920s, and the increase in corporate profitability, was more than half of
the increase in real stock market values. The runup in stock prices during
the 1950s and 1960s was extraordinary too, but in real terms increases in
40
dividends and in earnings were two-thirds as large as the increase in real
values.
The most recent bull market, as measured by the S&P composite index, is
the largest: a more than seven-fold increase in real values. Yet real
dividends paid on a pro-rata share of the S&P composite index have risen
by less than 30% since the early 1980s. And earnings on a pro-rata share
have increased by less than 50 percent.
Any economist examining this pattern must reach one of two conclusions
(or hedge his or her bets by taking a position between them). The first is
that for a century the stock market has been grossly underpriced—has
discounted the risk associated with owning equities at a much too high
rate. It is only now that equity valuations are “fair” in the sense of
promising expected real returns on stocks akin to those on bonds plus a
small extra risk premium. The second is that the stock market today is
subject to irrational exuberance on a scale never before seen in America.13
13
The conclusion reached by Robert Shiller (2000), who backs up his quantitative
estimates of fundamental values with a great deal of thick description of the thought
processes of market participants. Of course, only the thought processes of the marginal
agent are truly relevant to assessing the information content of prices.
41
If the second conclusion is correct, what role has the “new economy”
played in drawing tighter limits around the stabilizing potential of
42
arbitrage14 and in diminishing the information about fundamentals in the
hands of the marginal investor? Odean and Barber (2001) point out that
experimental economists have spelled out conditions under which markets
are most vulnerable to prolonged mispricing and to speculative bubbles,
and that our current stock market as it has been fueled by the growth of
online trading and online information appears to meet all of them. Stock
markets have managed to generate prolonged mispricing and spectacular
crashes in the absence of the internet in the past. But there is definitely
reason to worry that the extra information about and access to the stock
market provided by the information technology revolution has not led to a
more informed marginal investor, or to a market that is a better judge of
fundamental values.
If the future of the "new economy" is as bright as the previous section
suggests, then why have high-tech stock market values fallen so far in the
past year and a half? There is a strand of today's conventional wisdom that
holds that the crash of the NASDAQ reveals that the "new economy" was
smoke and mirrors. It was the irrational exuberance that often breaks out
14
A way of thinking about the problem of noise trading developed by Shleifer and
Vishny (1997).
43
at the peak of a boom, not any deeper or permanent change in the
economy. But it is more likely that the crash of the NASDAQ was the
result of the realization by investors that the "new economy" was, in most
sectors and for most firms, likely to lead not to large quasi-rents from
established market positions but to heightened competition and reduced
margins.
The exuberance that pushed the NASDAQ so high in 1999 and early 2000
rested on the belief technological leap forward in data processing and data
communications technologies had created a large host of winner-take-all
markets in which increasing returns to scale were the dominant feature. An
information good--a computer program, a piece of online entertainment, or
a source of information--the work only needs to be done once and then it
can be distributed to a potentially unlimited number of consumers for
pennies: producing at twice the scale gains you nearly a 50% cost
advantage. Moreover, information goods produced at larger scale are more
valuable to consumers. The version with the largest market share becomes
the standard. It is the easiest to figure out how to use, the easiest to find
support for, and the one that works best with other products (which are, of
course, designed to work best with it).
44
In that part of the new economy dominated by supply-side economies of
scale and demand-side economies of scope, a firm that establishes a
market-share lead gains a nearly overwhelming position. Its products are
most valuable to customers. Its cost of production is the least. Unless its
competitors are willing to take extraordinary and extraordinarily costly
steps--like those Microsoft took against Netscape, pouring a fortune into
creating a competitive product and then distributing the competing
Internet Explorer for free--the first firm to establish a dominant market
position will reap high profits as long as its sector of the industry lasts.
But increasing returns to scale and winner-take-all markets are not the
only or even the primary consequence of high-tech's technological
revolution. It is at least as likely that innovations in computer and
communications technologies are competition's friends. Theythe frictions
that in the past gave nearly every producer in the economy a little bit of
monopoly power. They enable swift searches that reveal the prices and
qualities of every single producer, while in the past such information could
only be acquired by a lengthy, costly, painful process. In the past you
45
could find comparison-shop only by trudging from store to store. In the
present you can use the world wide web.
Thus in the "new economy" more markets will be contestable.
Competitive edges based on past reputations or brand loyalty or
advertising footprints will fade away. As they do so profit margins will
fall: competition will become swifter, stronger, more pervasive, and more
nearly perfect.
Consumers will gain and shareholders will lose. Those products that can
be competitively supplied will be at very low margins. The future of the
technology is bright; the future of the profit margins of businesses--save
for those few that truly are able to use economies of scale to create
mammoth cost advantages--is dim. Is it really possible for anyone to
acquire significant economies of scale by writing a single suite of software
that will cover the heterogeneous purchasing requirements of millions of
businesses seeking to streamline their operations by using the internet? Is
it really possible for anyone to acquire significant economies of scale by
using the internet to distribute information about groceries? Ther
NASDAQ crash was the result of the marginal investor's realizing that the
46
odds were heavily against. But the NASDAQ crash tells us little about the
future of the underlying technologies, or about their true value.
The end of a period of high euphoria and extravagant boom will inevitably
bring a reduction in investment in the economy's leading sectors. This
reduction will not by itself bring about a Great Depression--or even more
than a period of "readjustment"--as long as other sources of demand are
present and able to absorb the slack in productive resources created by the
end of high euphoria. However, managing this expenditure-switching is a
very delicate macroeconomic task.
Moreover a euphoric boom is a period during which people stop thinking
as intensely about problems of macroeconomic management and the
business cycle. Ironically, it is precisely during euphoria that
countercyclical policy becomes less important, but it is in the aftermath of
euphoria that countercyclical policy becomes more important than at any
other time. For example, nobody in Japan in the late 1980s paid any
attention to problems of business cycle management. Few in Japan in the
early 1990s paid sufficient attention to the business cycle. And the
47
Japanese economy and the world economy today are suffering from that
lapse.
48
VII. Conclusion
What determines whether demand for a leading sector’s products is priceinelastic—in which case we are in Baumol and Bowen’s “cost disease”
scenario in which technological progress in the leading sector barely
affects the aggregate economy at all—or price-elastic—in which case we
are in the “economic revolution” scenario, and everything is transformed?
What determines the income and price elasticities of demand for the hightech goods that are the products of our current leading sectors?
The more are high-tech products seen as "luxury" goods, and the greater is
the number of different uses found for high-tech products as their prices
decline, the larger will be the income and price elasticities of demand--and
thus the stronger will be the forces pushing the expenditure share up, not
down, as technological advance continues.
Modern silicon and fiber-based electronics technologies may well fit
Bresnahan and Trajtenberg's (1995) definition of a "general purpose
technology"--one useful not just for one narrow class but for an extremely
wide variety of production processes, one for which each decline in price
appears to bring forth new uses, one that can spark off a long-lasting major
economic transformation. Such general purpose technologies are, as
Bresnahan and Trajtenberg say, “engines of growth”: precisely because
they have a wide range of potential uses, and are complementary to a large
proportion of other inputs, their price elasticity of demand is likely to be
high.
49
The possibility of the demand-side externalities called “network effects”—
Metcalfe’s law, the idea that the value of any network is proportional to
the square of the number of connected nodes—raises the likely elasticity
of demand still further. (However, offsetting the point that value is
proportional to the square of the size of the network is the point that the
most valuable nodes are likely to be connected to the network first—a
point that Paul Krugman (2000) has made and called “DeLong’s law.”)
The wide potential domain of use of information technology is one sign
that it is truly a high-elasticity general purpose technology. Selling plastic
doghouses in warehouse stores in middle America is not usually thought
of as a high-tech enterprise. Yet Wal-Mart’s extraordinary efficiency
advantage over other retailers in the 1980s and 1990s can be credited in
large part to its early investments in modern information technology, and
to careful thought and skilled execution of how modern information
technology can achieve economies of distribution. As Wal-Mart founder
Sam Walton (1992) wrote in his autobiography:
Nowadays, I see management articles about information
sharing as a new source of power in corporations. We’ve
been doing this from the days when we only had a handful
of stores. Back then, we believed in showing a store
manager every single number relating to his store, and
eventually we began sharing those numbers with the
department heads in our stores. We’ve kept doing it as
50
we’ve grown. That’s why we’ve spent hundreds of millions
of dollars on computers and satellites--to spread all the
little details around the company as fast as possible. But
they were worth the cost. It’s only because of information
technology that our store managers have a really clear
sense of what they’re doing most of the time.
In addition, the history of the electronics sector suggests that the income
and price elasticities tend to be high, not low. Each successive generation
of falling prices for computers, switches, and cables has produced
radically new uses for computers and communications equipment.
The first, very expensive, computers were seen as good at performing
complicated and lengthy sets of arithmetic operations. The first leadingedge applications of large-scale electronic computing power were military:
the burst of innovation during World War II that produced the first one-ofa-kind hand-tooled electronic computers was totally funded by the war
effort. The coming of the Korean War won IBM its first contract to
actually deliver a computer: the million-dollar Defense Calculator. The
military demand in the 1950s and the 1960s by projects such as Whirlwind
and SAGE [Semi Automatic Ground Environment]--a strategic air defense
system--both filled the assembly lines of computer manufacturers and
trained the generation of engineers that designed and built.
51
The first leading-edge civilian economic applications of large--for the
time, the 1950s--amounts of computer power came from government
agencies like the Census and from industries like insurance and finance
which performed lengthy sets of calculations as they processed large
amounts of paper. The first UNIVAC computer was bought by the Census
Bureau. The second and third orders came from A.C. Nielson Market
Research and the Prudential Insurance Company. This second, slightly
cheaper, generation was of computers was used not to make sophisticated
calculations, but to make the extremely simple calculations needed by the
Census, and by the human resource departments of large corporations. The
Census Bureau used computers to replace their electro-mechanical
tabulating machines. Businesses used computers to do the payroll, reportgenerating, and record-analyzing tasks that their own electro-mechanical
calculators had previously performed.
The still next generation of computers--exemplified by the IBM 360
series--were used to stuff data into and pull data out of databases in real
time--airline reservations processing systems, insurance systems,
inventory control. It became clear that the computer was good for much
more than performing repetitive calculations at high speed. The computer
was much more than a calculator, however large and however fast. It was
also an organizer. American Airlines used computers to create its SABRE
automated reservations system, which cost as much as ten airplanes (see
Cohen, Delong, and Zysman (2000)). The insurance industry automated its
back office sorting and classifying.
52
Subsequent uses have included computer-aided product design, applied to
everything from airplanes designed without wind-tunnels to
pharmaceuticals designed at the molecular level for particular
applications. In this area and in other applications, the major function of
the computer is not as a calculator, a tabulator, or a database manager, but
is instead as a what-if machine. The computer creates models of what-if:
would happen if the airplane, the molecule, the business, or the document
were to be built up in a particular way. It thus enables an amount and a
degree of experimentation in the virtual world that would be prohibitively
expensive in resources and time in the real world.
The value of this use as a what-if machine took most computer scientists
and computer manufacturers by surprise. None of the engineers designing
softare for the IBM 360 series, none of the parents of Berkeley UNIX,
nobody before Dan Bricklin programmed Visicalc had any idea of the
utility of a spreadsheet program. Yet the invention of the spreadsheet
marked the spread of computers into the office as a what-if machine.
Indeed, the computerization of Americas white-collar offices in the 1980s
was largely driven by the spreadsheet program's utility--first Visicalc, then
Lotus 1-2-3, and finally Microsoft Excel.
For one example of the importance of a computer as a what-if machine,
consider that today's complex designs for new semiconductors would be
simply impossible without automated design tools. The process has come
full circle. Progress in computing depends upon Moore's law; and the
53
progress in semiconductors that makes possible the continued march of
Moore's law depends upon progress in computers and software.
As increasing computer power has enabled their use in real-time control,
the domain has expanded further as lead users have figured out new
applications. Production and distribution processes have been and are
being transformed. Moreover, it is not just robotic auto painting or
assembly that have become possible, but scanner-based retail quick-turn
supply chains and robot-guided hip surgery as well.
In the most recent years the evolution of the computer and its uses has
continued. It has branched along two quite different paths. First,
computers have burrowed inside conventional products as they have
become embedded systems. Second, computers have connected outside to
create what we call the world wide web: a distributed global database of
information all accessible through the single global network. Paralleling
the revolution in data processing capacity has been a similar revolution in
data communications capacity. There is no sign that the domain of
potential uses has been exhausted. So far there are no good reasons to
believe that the economic salience of high-tech industries are about to
decline, or that the pace at which innovation continues is about to flag.
There is room for computerization to grow on the intensive margin, as
computer use saturates potential markets like office work and email. But
there is also room to grow on the extensive margin, as microprocessors are
used for tasks like controlling hotel room doors or changing the burn mix
54
of a household furnace that few, two decades ago, would ever have
thought of.
Thus the balance of probabilities is that the elasticity of demand for the
products of our current high-tech computer and communications leading
sectors is high, not low. Because of the general purpose nature of the
technology, it has an enormous number of potential uses, many of which
have not yet been developed. The way to bet is that our new economy will
have not a limited but an enormous impact on how we live.
55
56
Table 1
Growth Accounting and the Post-73 Productivity Slowdown
nonfarm business, percent per year
Period
Output per hour
Contributions froma
Capital
IT Capital
Other Capital
Labor quality
Multifactor Prod.
MFP from R&D
1948-73
1973-95
Difference
2.9
1.4
-1.5
0.8
0.1
0.7
0.2
1.9
0.2
0.7
0.4
0.3
0.2
0.4
0.2
-0.1
0.3
-0.4
0.0
-1.5
0.0
Source: Bureau of Labor Statistics USDL 01-125 May 3, 2001.
Note: a Contributions do not add exactly to the total because of rounding and because
growth rates compound multiplicatively.
Table 2
57
Accounting for the Post-95
Productivity Speed-Up
percent per year
Difference in growth (1995-2000 minus 1973-95)
Output per hour
Contributions from Capital d
IT Capital
Other Capital
Labor quality
Multifactor Prod.
Computer sector MFP
Other MFP
Oliner-Sichela
1.08
0.34
0.59
-0.25
0.04
0.72
0.47
0.25
Notes: a Updated figures provided by Daniel Sichel, see Oliner and Sichel (2000).
Nonfarm business.
b
Updated figures based on the methodology used in Martin Baily and Robert
Lawrence (2001). Nonfarm business.
c
Dale Jorgenson, Mun Hoh and Kevin Stiroh (2001).
d
Contributions do not add exactly to the total because growth rates are
compounded multiplicatively.
Baily-Lawren
1.26
0.44
0.59
-0.15
0.04
0.82
0.18
0.64
Table 3: Labor Productivity Growth by Industry
GDP originating per full-time equivalent employee, average annual
percent changes, selected periods.
58
1989-95
1995-2000 Difference
a
Private Industries
0.88
1.97
1.09
Agriculture
0.34
2.75
2.41
Mining
4.56
-1.78
-6.34
Construction
-0.10
-0.66
-0.56
Manufacturing
3.18
4.45
1.27
Durables
4.34
6.77
2.43
Non-Durables
1.65
1.43
-0.23
Transportation
2.48
1.52
-0.96
Trucking and Warehousing
2.09
0.99
-1.10
Transportation by Air
4.52
2.19
-2.33
Other Transportation
1.51
1.51
0.00
Communication
5.07
2.19
-2.88
Electric / Gas / Sanitary
2.51
2.25
-0.26
Wholesale Trade
2.84
5.90
3.06
Retail Trade
0.68
4.74
4.05
FIRE
1.70
3.51
1.81
Finance
3.18
9.53
6.34
Insurance
-0.28
0.42
0.70
Real Estate
1.38
2.80
1.42
Services
-1.12
0.08
1.21
Personal Services
-1.47
0.66
2.13
Business Services
-0.16
1.12
1.28
Health Services
-2.31
-0.23
2.09
Other Services
-0.72
-0.24
0.47
ICT Intensive Half
2.43
4.15
1.72
Non-ICT Intensive Half
-0.10
1.05
1.15
Source: Based on data from Bureau of Economic Analysis.
Note a: Covers different periods, includes agriculture and non-business
sectors and so is not directly comparable to the non-farm business
sector results reported earlier.
59
Figure 1: Net foreign purchases of US assets
1100
Othe
900
Secu
Direct investment
billion dollars
700
Official assets
500
US Treasuries
and currency
300
100
-100
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
199
60
61
References
Baily, Martin Neil and Robert Z. Lawrence, “Do We Have a New E-conomy?” American
Economic Review, Papers and Proceedings, 91 (2), May 2001, 308-312.
Baily, Martin N., Robert D. Willig, Peter R. Orszag and Jonathan M. Orszag, An Economic
Analysis of Spectrum Allocation and Advanced Wireless Services, Sebago Associates,
Washington and San Francisco, October 2001.
Baily, Martin Neil, “Macroeconomic Implications of the New Economy,” Working Paper WP019, Institute for International Economics, September 2001. Paper presented at the Kansas City
Federal Reserve conference on the information economy, Jackson Hole Wyoming, Aug-Sept,
2001.
Ball, Laurence, and Robert Moffitt, “Productivity Growth and the Phillips Curve,” Photocopy.
Johns Hopkins University, January 2001.
Blinder, Alan, “The Internet and the New Economy,” Brookings Institution Policy Brief, 60, June
2000.
Council of Economic Advisers, Economic Report of the President, Washington, DC:
U. S. Government Printing Office, January 2001.
Gordon, Robert J., “Does the ‘New Economy’ Measure up to the Great Inventions of the Past?”
Journal of Economic Perspectives, Fall 2000, 14(4), 49-74.
Gordon, Robert J., “Did the Productivity Revival Spill Over from Manufacturing to Services?
Conflicting Evidence from Four Data Sources.” Paper presented at the NBER productivity
workshop, July 27, 2001b.
Hall, Robert E., “The Stock Market and Capital Accumulation,” American Economic Review, 91
(5), December 2001, 1185-1202.
Jorgenson, Dale W., Mun S. Ho, and Kevin J. Stiroh, “Projecting Productivity Growth: Lessons
from the U.S. Growth Resurgence,” paper prepared for the conference on “Technology, Growth
and the Labor Market,” sponsored by the Federal Reserve Bank of Atlanta and Georgia State
University, Mimeo, December 31, 2001.
Litan Robert E., and Alice M. Rivlin, The Economic Payoff from the Internet Revolution,
Brookings, Washington DC, 2001.
McKinsey Global Institute, US Productivity Growth 1995-2000, McKinsey Global Institute,
Washington DC, October 2001.
Nordhaus, William D., “Productivity Growth and the New Economy,” Working Paper 8096,
National Bureau of Economic Research, January 2001.
Oliner Stephen D., and Daniel E. Sichel, “The Resurgence of Growth in the Late 1990s: Is
Information Technology the Story?” Journal of 62Economic Perspectives, Fall 2000, 14(4), 3-32.
Oliner Stephen D., and Daniel E. Sichel, “Information Technology and Productivity: Where are
We Now and Where are We Going?” notes for a presentation at the conference on “Technology,
Growth and the Labor Market,” sponsored by the Federal Reserve Bank of Atlanta and Georgia
State University, Mimeo, January 7, 2002.
Stiroh, Kevin, “Information Technology and the U.S. Productivity Revival: What Do the
Industry Data Say?” Mimeo, Federal Reserve Bank of New York, December 2001;
forthcoming, American Economic Review.