Transcript Notes on growth

How economists measure
the impact of the digital
economy on growth
Jonathan Haskel
Digital Economy seminar
1 June 16:00-17:30
Level 1 Lecture Theatre, Imperial College Business School
Purpose of seminar: to try to explain to Digital Economy research
partners some of the work that economists are doing to investigate
economic growth and the digital economy.
Research questions
•
What is the digital economy?
–
–
•
Digitisation refers to new methods of encoding and transmitting information
A technological revolution, since such new methods are better, faster, cheaper, analgous to mechanisation
One related question: what is the impact of ICT on productivty growth?
–
Many technological revolutions in the past, how does ICT compare?
•
•
–
ICT is
•
•
•
•
Hardware - usually taken to be computers
Software
Comms equipment –the internet
What are the policy implications of this?
–
–
•
ICT a general purpose technology
E.g. Industrial revolution, electricity
Example: should there be tax breaks for software as there are for R&D?
Should we build more internet capacity? Have more hardware companies?
Other interesting related questions, not looked at here
–
–
Analysis of specific ICT/digital companies e.g. how, especially new start ups, grow, finance
and organise themselves etc.
Regulation and competition in the digital market place e.g. copyright, net neutrality, mobile
phone regulation, standard setting
Why are we interested in growth?
•
Definition:
–
–
–
economic growth = growth in GDP per persons
Productivity growth = growth in GDP per worker
(relation between the two: workers per person = labour force participation,
•
•
Points:
–
improvements in economic growth raise living standards
•
•
•
–
•
Lab force participation related to 1-unemployment rate
Compare developed and developing
Compare economics over time
(But, what about bads from economic growth, pollution etc?)
It turns out that economic growth is driven by productivity growth not participation
The fruits of economic growth in one table:
–
–
Higher productivity leads to higher incomes for all (though distribution might change). So
higher productivity per worker leads to higher incomes per worker (unless capital incomes
rise)
In fact: the average American’s annual income in 2000 was
• five times higher than that in 1890 and
• 12 times higher than in the 1850s.
– One way to see this: the number of hours a worker would have to work to obtain the
same items in 1895 versus 2000.
Multiplication of Productivity 1895–2000: Time Needed for an Average Worker
to Earn the Purchase Price of Various Commodities
Commodity
Horatio Alger (6
vols.)
One-speed bicycle
Cushioned office
chair
100-piece dinner set
Hair brush
Cane rocking chair
Solid gold locket
Encyclopedia
Britannica
Steinway piano
Sterling silver
teaspoon
Time to Earn Productivity
in 2000 (hours) Multiple
Time to Earn in
1895 (hours)
21
260
0.6
7.2
35
36.1
24
2
12
44
16
8
28
3.6
2
1.6
6
12.2
8
5
4.7
140
2,400
33.8
1,107.60
4.1
2.2
26
34
0.8
Source: Montgomery Ward catalogue, table in
http://www.econlib.org/library/Enc/StandardsofLivingandModernEconomicGro
wth.html
So what does very long run growth
look like?
The most powerful force in the
universe
0
10000
20000
30000
USA - RealGDP per Capita (1990 US$)
1900
1950
Year
2000
The visual power of logs
8
8.5
9
9.5
10
10.5
USA - Log RealGDP per Capita (1990 US$)
1900
1950
Year
lusa
2000
Fitted values
Real GDP per hour worked,
Europe versus the US, 1870-2000
GDP per hour worked varies relative to per capita
GDP catch up
Monday, July 20, 2015
8
So what is the contribution of
ICT/digital to all this?
• Hard to see in diagrams, but in US
– Post 1995, acceleration in growth
– Followed a long ICT computer investment
boom, and just when internet was getting
connected
– Puzzle: no such acceleration in the EU
• So need a framework to assess how ICT
affects productivity
The sources of growth framework
Labour productivity growth
Driven by:
Physical capital growth
Human capital growth
Knowledge capital growth
Example: Ryanair
Increase in pax flown per Ryanair employee
Better planes flying faster
Better trained and qualified staff
Software for booking and rostering
Ticketless boarding
Overall comments
1. So what do we want to measure?
–
–
–
–
output (GDP = final output. Intermediates make this complicated to measure)
human capital
physical capital (capital services from planes, computers etc)
knowledge capital (capital services from software, business processes etc.)
2. At least some knowledge is embodied in machines.
Example of the framework in action:
1.
2.
intensive versus extensive growth.
suppose you added to an economy more and more machines (planes): what
would happen to growth?
Case study of intensive margin
growth: the USSR
GDP per capita,. 1948-89, USA and USSR
26,668
16,668
USSR
USA
11,668
6,668
1988
1986
1984
1982
1980
1978
1976
1974
1972
1970
1968
1966
1964
1962
1960
1958
1956
1954
1952
1950
1948
1,668
1946
1190 $ pe head
21,668
1988
1986
1984
1982
1980
1978
1976
1974
1972
1970
1968
1966
1964
1962
1960
1958
1956
1954
1952
1950
1948
1946
1190 $ pe head (log scale)
Extensive margin growth:
leads to diminishing returns
GDP per capita (log scale)
16,680
USSR
USA
1,668
How do we measure capital
services?
•
•
•
Measure “effective” capital stock
Broadly= additions to capital stock = volume of new investment – deprecation
Perpetual inventory model:
– K(t)=I(t)+(1-d)K(t-1)
•
•
How to measure I = investment?
Consider a pencil economy
– We really want the “volume” of pencils in the economy.
– In practice statisticians
•
•
•
•
Do an investment survey: how much are firms spending on pencils? (=Pp*Ip) = value of
investment
Do a price survey: how much does the average pencil cost? (=Pp)
Derive “volume” of pencil investment Ip = Pp*Ip/Pp
Computers
•
•
Do an investment survey: gives Pc*Ic
But then you need Pc. With rapidly evolving goods, that is not how much the price of the
average computer, since the average computer is vastly more powerful, more memory etc. than
before. So need to “quality adjust” prices for new computers
What’s been happening to
computer prices?
• Moore’s law
– The number of transistors on a chip doubles every
12-24 months. So “quality adjusted” prices have
been falling hugely. Now about 1/250th of level in
2000
• What is the equivalent of such a price fall in
the prices of everyday goods?
– 1978: a commercial flight between New York and
Paris cost $900. If the implied cost fall were
applied to the airline industry, cost = 4c.
– 1970: price of a new car in 1970 (in 2000 $) was
$16,000. prices now: $64.
Relative Prices of Computers, Communications, Semiconductors, and Software and
Computer Services Industry Output, 1960-2005
1000
10
1
Computers
Communications Equipment
Source: Jorgenson (2007)
Electronic Components
20
04
20
02
20
00
19
98
19
96
19
94
19
92
19
90
19
88
19
86
19
84
19
82
19
80
19
78
19
76
19
74
19
72
19
70
19
68
19
66
19
64
19
62
0.1
19
60
Log Scale (2000=1)
100
Software and Computer Services
Output Shares of Information Technology by Type, 1948-2003
Share of current dollar gross domestic product.
5.0
4.5
4.0
3.5
%
3.0
2.5
2.0
1.5
1.0
0.5
0.0
1948
1953
1958
1963
1968
1973
1978
1983
1988
1993
1998
2003
Source: Jorgenson
Computers
Communications
Software
Total
How to measure effects on
output?
• Let output come from
– capital
– Labour
– Ideas
• Output has risen hugely: how much is from capital, from labour, from ideas?
• Assume
capital labour ideas
Y  F( K , L , R )
Thus % change in all output:
elasticity of
Y due to K
Y Y / Y

Y
K / K
elasticity of
Y due to L
K
K
% increase in K
Y / Y

L / L
elasticity of
Y due to R
L
L
% increase in L
Y / Y

R / R
R
R
% increase in R
•How do we measure the elasticities?
– suppose we move from L0 to L1 “units” of
Understanding
growth
labour
(say check-in staff/call
centres)? How
much extra output, Y0 to Y1 do we get out?
– Approach 1: engineering
– Approach 2: economics
• Cost of those extra workers= W*(L1-L0)
• Revenue from that extra output so generated =
Py*(Y1-Y0)
• Assumption: cost-minimising firms:
– If extra revenue> extra costs, hire more workers
– So in equilibrium: Py* (Y1-Y0)= W*(L1-L0)
– Thus in equilibrium the extra output is the price ratio
times the extra labour: (Y1-Y0) = (W/Py) * (L1-L0)
– In % changes (Y1-Y0)/Y0 = (WL0)/PyY0) * (L1-L0)/L0
Using prices to determine
elasticities
Engineering :
elasticity of
Y due to K
elasticity of
Y due to L
Y Y / Y

Y
K / K
K
K
Y / Y

L / L
% increase in K
elasticity of
Y due to R
L
L
Y / Y

R / R
% increase in L
R
R
% increase in R
Economics :
share of K
costs in PY
PK K
Y

Y
PY Y
share of L
costs in PY
K
K
% increase in K
PL L

PY Y
share of R
costs in PY
L
L
% increase in L
PR R

PY Y
R
R
% increase in R
Implementation
• Prices of labour and capital we can see
• Price of knowledge?
– Some costs: e.g. licence to use software
– Some free: = a spillover
• e.g. university knowledge, copying others,
information on the internet
– Some just too hard to measure: e.g. a unit of
management consulting
– Let’s assume its free..then growth is
Growth when knowledge is free
share of K
costs in PY
share of L
costs in PY
PK
Y
 K
Y
PY Y
K
K

% increase in K
PL L
PY Y
L
L
contrib of free knowledge
(plus mismeasurement)
 TFPgrowth
% increase in L
Additional assumption: by construction PYY=WL+PKK. So we can
write:
productivity growth
= growth output per worker
 Y L 



Y
L



 contrib of free knowledge

 (plus mismeasurement)
PK K  K L 


 TFPgrowth


PY Y
K
L
 capital deepening

 =growth cap per worker 
share of K
costs in PY
Applying to the USA
Capital cost share of value
added, USA
10000000
0.37
8000000
0.35
6000000
VA
0.33
4000000
LAB
0.31
2000000
CAP
0.29
0
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
$, '000
Nominal output, capital and
labour spend, USA, 19802004
CAP/VA_US
5 per. Mov. Avg. (CAP/VA_US)
Applying to the USA
Growth in output per worker and capital per worker, USA,
1980-2005 (5 yr moving averages)
0.10
0.08
Axis Title
0.06
0.04
0.02
0.00
1980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004
-0.02
-0.04
dln(y-l)_US
dln(k-l)US
5 per. Mov. Avg. (dln(y-l)_US)
5 per. Mov. Avg. (dln(k-l)US)
Applying to the USA
3
2.5
2
1.5
1
0.5
0
US, 1980-95
US, 1995-05
The EU/USA puzzle
3
2.5
2
1.5
1
0.5
0
US, 1995-05
EU 15, 1995-05
The EU/US puzzle
(1)
(2)
(3)
(4)
(5)
(6)
VA
L
K
KIT
KNIT
MFP
(1)=(2)+(5)
+(6)
(3)=(4)+(5)
1980-1995
USA
3.2
1.1
1.4
0.8
0.6
0.8
EU 15
2.1
0.0
1.1
0.4
0.7
1.0
USA
3.6
0.7
1.6
1.0
0.5
1.3
EU 15
2.2
0.6
1.2
0.6
0.6
0.4
1995-2005
Why the divergence? Possible answers:
a. lack of EU co-investment
b. Structural problems preventing EU services from TFP
growth
Source: EUKLEMS. EU15 are those for whom data available AUT, BEL, DNK, ESP, FIN, FRA, GER,
ITA, NLD & UK.
Research questions
• Why US v EU?
– Efficiency of service sector, regulation etc.
• Measurement
– Prices of ICT goods
– Better measures of knowledge rather then just
free
– More information from engineers on efficiency
e.g. potential gains in information technology
efficiency