ICT as a GPT:
an Historical Perspective
Nicholas Crafts
The Solow Productivity Paradox
“You can see the
computer age
everywhere except in
the productivity
statistics”
Robert Solow, 1987
Solow Paradox Revisited
• Even before the mid-1990s ICT had a much
bigger impact than steam or electricity
• The Solow paradox was based on unrealistic
expectations…initially new technologies have a
small weight in the economy
• The growth potential of GPTs has been realised
more quickly over time
Steam as a GPT
• Steam Engines, Railways, Steamships
• James Watt’s invention: 1769
• Liverpool & Manchester Railway: 1830
• Steamship crosses the Atlantic: 1838
Steam Engine Technology
• Took a long time to become cost effective in most
sectors
• Coal consumption per hp per hour fell from 30 lb preWatt to 12.5 lb for Watt engine to 2 lb by 1900 when psi
reached 200 compared with 6 in1770
• The big breakthrough was not James Watt but the
move to the high pressure steam engine after 1850
• Cumulative effect on British labour productivity over 150
years similar to that of ICT on American labour
productivity over 35 years
Sources of Power, 1760-1907
(Thousand Horsepower)
1760
1800
1830
1870
1907
Steam
5
35
165
2060
9659
Water
70
120
165
230
178
Wind
10
15
20
10
5
Total
85
170
350
2300
9842
Source: Kanefsky (1979a, p338); not including internal combustion engines
Total Steam Contribution to Growth
of Labour Productivity (% per year)
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
1760-1800
1800-30
1830-50
Source: Crafts (2004); includes railway, steamships, steam engines
1850-70
1870-1910
Implications
• Small contribution of steam pre-1830 helps
explain Crafts-Harley view of industrial
revolution
• Strong contribution of steam power in second
half of 19th century says climacteric not
explained by hiatus between GPTs
• Puts Solow Paradox into perspective
The Computer and the Dynamo
(David, 1991)
• The big impact of commercial electricity on American
productivity was in the 1920s, 40 years after Edison
• TFP spillovers a key aspect of the 1920s’ productivity
surge; factory design improved through learning
externalities in transition from drive-shafts to wires
• Across manufacturing sectors, change in TFP growth
strongly correlated with change in electric motors
• Impact of TFP spillovers is to raise Y/L growth in
manufacturing by 2.4, whole economy by 0.7pp/year
Contributions to US Labor Productivity Growth
(Crafts, 2002; Oliner et al., 2007)
Electricity
1899-1919 0.40
ICT
1973-1995 0.74
1919-1929 0.98 (0.28) 1995-2006 1.45
Real Hedonic Price Index for Electric
Motors in Sweden (Edquist, 2010)
1901
100
1914
65.4
1925
61.5
1935
55.7
Source: Nordhaus (2007)
Capital Cost and Annual Cost per
Steam Horsepower per year (£ current)
Capital Cost
Annual Cost
1760
42
33.5
1800
56
20.4
1830
60
20.4
1850
37
13.4
1870
25
8.0
1910
15
4.0
Note: the estimates are for a benchmark textile mill in a low coal cost region like
Manchester
Impacts of GPTs on Growth
• ICT much bigger impact on American growth in
recent past than steam ever had on UK growth
• Costs of computing have fallen much faster than
did costs of electrical power or steam power
• ICT is historically remarkable
• Society seems to be getting better at exploiting
GPTs more rapidly
Does Innovation Generate
Supernormal Profits? (Nordhaus, 2004)
• Innovators capture about 2% of the total social
gain from technological progress
• The US stock market valuation of ‘new economy’
firms grew between 1995 and 2000 at a rate that
implied owners could capture 90% of the social
gain
• Yet the appropriability of gains from ICT unlikely to
match that of earlier technologies including
railways
Social Savings vs. Growth Accounting
• Alternative methodologies to estimate contribution of
new technology to economic growth
• Social savings pioneered by Fogel (1964) for US
railroads; measures reduction in cost
• Social savings amounts to a subset of growth
accounting; net vs. gross effects
• Social saving focuses attention on the benefits; generally
speaking, in the long run users get the benefits of a
new technology (Nordhaus, 2004)
Social Savings of Railways
• Transport benefits = total economic benefits
• Fogel’s upper bound measure: A + B + C + D + E
• Under perfect competition real cost of rail transport falls
at rate of TFP growth (price-dual measure)
• Compared with growth accounting includes own TFP
contribution but no spillovers and no capital deepening
• Capital deepening not ‘unique’; in absence of railway
other normal-returns investments
Figure 2
PT
PT0
D
A
B
E
C
PTI
D1
D
D2
T0
T1
Transportation
Social Gains from Railways, 1912
(%GDP)
14
12
10
8
Britain
Spain
6
4
2
0
-2
User Benefits
Net Revenue
Supernormal
Profits
Welfare Benefits
• Growth accounting focuses on increase in
productive potential but social savings highlights
user benefits
• Not necessarily the same in an open economy
• British railway-users got nearly all the benefits
• Cotton textiles in the industrial revolution: foreign
consumers took 60% of the benefits of British
technological progress (Harley,1999)
Transport Benefits May Not Capture
All the Economic Benefits
• Wider economic benefits may be relevant; cf. recent
developments in transport CBA
• Understanding the characteristics of the transport-using
sector matters
• Imperfect competition (freight)
• New goods (passengers)
• Agglomeration benefits from industrial re-location, bigger
cities etc = TFP spillovers (both)
City Size and Labour Productivity
• Elasticity of labour productivity with respect to city size in
range 0.04 to 0.11
• Robust result but could be one or more of several
reasons
• Effective city size is the key concept; population with 80
minutes of UK NUTS1 region has positive effect on
productivity (Rice & Venables, 2004)
• If railways increased effective city size, there would be a
TFP spillover
Better Transport: with
Productivity Externality
TFP Spillovers
• In danger of being unduly neglected; clearly
the really hard part of evaluating impact of GPT
• GPTs often have geographic implications and
that may be key source of spillovers
• Intra-sectoral correlation of capital in use and
TFP growth does not necessarily capture these
Growth in No-ICT-Production
Country (Oulton, 2010)
yc = Bh1-α –β(kc)α(kICT)β
Δp/p = μc – μICT < 0
Δyc/yc = μc + (1-α-β)Δh/h + αΔkc/kc + βΔkICT/kICT
User cost = marginal product, so
Δyc*/yc* = μc + (1-α-β)Δh/h + αΔyc */yc * + β(Δyc*/yc – Δp/p)
Δyc*/yc* = Δc*/c* = (μc – βΔp/p)/(1-α-β) + Δh/h
Implications
• So in open economy consumers get higher
growth from technological progress in foreign
ICT even though no domestic ICT production
• Projected ICT-use effects on long-run growth
generally much bigger than ICT-output effects
• Economies like Ireland which export ICT
production have a negative terms of trade effect
on real income
ICT-Use Effects vs. ICT-Output
Effects (% per year) (Oulton, 2010)
ICT-Use
ICT-Output
Finland
0.67
0.57
France
0.48
0.17
Germany
0.44
0.33
Ireland
0.39
0.51
Uk
0.60
0.16
USA
0.70
0.22
Special Features of Celtic Tiger
Growth
• Pivotal role for export-platform FDI and supplyside policy designed to attract it
• Exceptionally large share of ICT production
(Domar weight = 22.6%)
• GDP much bigger than GNP and terms of trade
adjustment to real income growth unusually
large
• Employment growth made a large contribution
to real GDP growth
Contributions to Labour Productivity
Growth, 1995-2001 (% per year)
Ireland
EU
Total ICT
4.27
0.69
TFP in ICT Production
3.62
0.27
Other TFP
-0.01
0.19
Other Capital Deepening
1.22
0.48
Labour Productivity Growth
5.48
1.37
Source: Timmer and van Ark
(2005)
Growth of Real GNP
(% per year)
GNP
TT-Adjusted
GNP/Head GNP/HW
GNP/Head
19731987
2.7
1.7
3.1
1.0
19872003
5.6
4.9
3.6
3.9
Europe and ICT
• European countries have generally not matched USA in
ICT contribution to growth; UK does relatively well
• “American diagnosis” is too much regulation too much
taxation, too little competition (the ‘corporatist legacy’)
• Less ICT production but also smaller ICT capitaldeepening contribution; the latter is the worrying
diagnostic
• ICT-using services (e.g., distribution) have been
Europe’s Achilles Heel and may be where regulation
has slowed diffusion most
Regulation and the contribution of ICT-using services to
aggregate productivity growth
ICT using services, 1996-2001
Correlation coefficient: -0.62
t-statistic: -3.35
MEX
USA
1.3
AUS
GBR
0.8
IRL
SWE
CAN
0.3
NLD
JPN
AUT
NOR
DNK
FIN
CHE
KOR
BEL
DEU
ESP
ITA
FRA
-0.2
0
0.5
1
1.5
2
Product market regulation (inward-oriented), 1998
2.5
3
3.5
Source: Nicoletti & Scarpetta (2005)
UK in the ICT Era
• Compared with big continental economies, UK
fares much better with ICT than with Fordist
manufacturing
• Thatcherite reforms have lagged pay off
• Competitive product markets and flexible labour
markets are favourable to relatively rapid
diffusion of ICT
• NB: downside is exposure to large and badlyregulated financial services sector
Social Capability and ICT
• Standard American criticisms of Europe at least
equally valid for 20 years before 1995
• Social capability depends on requirements of the
technological epoch
• It is not that there is more regulation but rather
that existing regulation is more costly in the ICT
world
Policy Lesson
• Social capability that facilitates diffusion is the
key to good productivity performance
• Technology transfer matters much more than
domestic R & D; 5/6 new technology comes from
R & D in ROW (Eaton & Kornum, 1999)
• Intangible capital rather than just R & D is key to
exploiting new technology (Haskel et al., 2009)