Competition and innovation

Nick Bloom
Innovation
Nick Bloom, Macro Topics, Spring 2007
Two approaches to innovation work
Macro:
• Growth, but tough to come up with new angle, unless take
an interesting new micro-macro angle
• Business cycle – key question is does innovation change at
a BC frequency (Comin & Gertler, 2006 AER)
Micro:
• Labor: estimating innovation functions using patents, R&D
as function of demand, taxes, laws etc..
• IO: Modelling industry systems as fully endogenous
Theory
• Models of innovation
I have worked a lot on innovation historically but am doing less
now (at least on R&D and patents) as it is a well trodden path
Nick Bloom, Macro Topics, Spring 2007
Philippe Aghion, Nick Bloom, Richard Blundell,
Rachel Griffith and Peter Howitt
“Competition and Innovation: An Inverted-U
Relationship”
Quarterly Journal of Economics (2005)
Nick Bloom, Macro Topics, Spring 2007
Overview
Estimates an Aghion-Howitt style model using micro-macro data
The fundamental idea was to:
• Set up a stylized model with empirical predictions
• Carry out non-linear estimation of innovation-competition
relationship
• Use a clean IV strategy and estimate on micro data
An important paper:
(i) One first papers to estimate growth models using micro data
(ii) Very well stylized and accessible
Nick Bloom, Macro Topics, Spring 2007
Well cited paper (for its vintage)…
Nick Bloom, Macro Topics, Spring 2007
Although nothing compare to the Aghion & Howitt
(1992)
Nick Bloom, Macro Topics, Spring 2007
Conventional wisdom, theory and empirics have
conflicting views on competition & innovation link
• Conventional wisdom provides mixed views
“Competition effect”:
“... from Adam Smith to Richard Caves: the belief is that
competition is good...” (Nickell, 1996 JPE)
“Schumpeterian effect”:
“....anti-trust discourages innovation.......” (Bill Gate’s
lawyers, frequently…)
• Economic theory often supports the Schumpeterian effect of a
negative competition effect on innovation
• Empirical work typically finds a positive effect – i.e. Nickell
(1996 JPE), Blundell, Griffith & Van Reenen (1999 RES)
Nick Bloom, Macro Topics, Spring 2007
Paper develops a stylised model of competition,
innovation and growth across industries (1)
•
•
The economy contains many industries. Two firms in each
Industries are either:
• “neck-and-neck” as firms have the same technology
• “leader-follower” as firms have different technologies
•
Under low competition “neck-and-neck” firms earn
moderate profits, so limited incentive to innovate, so:
• “neck-and-neck” firms undertake little innovation, so
equilibrium has mainly “neck-and-neck” industries
• so increasing competition raises innovation as “neckand-neck” firms increase innovation in response to
more competition
“Escape competition effect”
Nick Bloom, Macro Topics, Spring 2007
Paper develops a stylised model of competition,
innovation and growth across industries (2)
•
Under high competition “neck-and-neck” profits are low, so
the rewards to innovating to become a leader are high, so:
• “neck-and-neck” firms undertake a lot of innovation
• equilibrium has mainly “leader-follower” industries
• so further increases in competition lower the profits for
followers to innovate and become “neck-and-neck”
“Schumpeterian effect”
•
This generates an inverted-U as competition first increases
innovation (as mainly “neck-&-neck” firms) then reduces
innovation (as mainly “leader-follower” firms)
Nick Bloom, Macro Topics, Spring 2007
The Model Provides Three Empirical Predictions
Model
Share of
“neck-&neck”
industries
Predictions
high
3. This higher the share of
“neck-and-neck” industries
the more positive the effect
of competition
low
2. The share of “neck-andneck” industries will decline
as competition increases
low
Competition
high
high
Innovation
low
1. Innovation and competition
will have an inverted Ushape relationship
low
Competition
Nick Bloom, Macro Topics, Spring 2007
high
Estimate This Using (UK) Accounting Data
• Measure innovation using patents (NBER file) matched to
UK firm-level accounts data
• Measure competition using a Lerner index (P-MC)/MC
– Again use accounts data by assuming AC ≈ MC
P  MC P  AC P  Q  AC  Q


P
P
PQ
– Endogenous so instrument using policy changes of
Privatizations and European Single Market Program
(note: this was the justification for using UK data)
Nick Bloom, Macro Topics, Spring 2007
Competition and innovation – raw data (figure 1)
•Like Syversson (2004) had a basic figure early in the paper
Suggested by the editor (Glaeser), and very good idea
Nick Bloom, Macro Topics, Spring 2007
There are some fancy econometrics
Basically, confirmed the inverted-U also held using splines
Introduced splines so latter on could use non-linear IV
Nick Bloom, Macro Topics, Spring 2007
The control function approach for non-linear IV
Non-linear IV is a good technique to use and control-function
approach is a good way, easy & intuitive way to do this.
Basic idea is two-step approach:
Step 1: regress X (Competition) on the Z instruments (policy) in a
non-linear fashion to soak out all the exogenous variation in X
Xi,t = α + f(Zi,t) + vi,t
Step 2: regress Y (innovation) on X plus the residuals vi,t (these
contains the endogenous bit of X) to identify from the exogenous
parts of X only
Yi,t = α + β1Xi,t + g(vi,t) + ei,t
Nick Bloom, Macro Topics, Spring 2007
Including/excluding controls still get a inverted-U
Note the standard
results set-up. Start
very simple on LHS
and then slowly add
junk as controls.
Want to show robust
in the simple and full
specifications…
Nick Bloom, Macro Topics, Spring 2007
Including/excluding controls still get a perverted-U
Note the full
footnotes – always
do this: (i) makes the
paper clearer, (ii)
signals you are more
serious
We actually forgot to
mention why column
(3) has only 67 obs
Nick Bloom, Macro Topics, Spring 2007
Also confirm the additional predictions that higherdispersion associated with both higher level of
competition, and higher response to competition
One thing to note is the styling of the paper. The QJE editor
(Glaeser) correctly requested we changed paper order from:
“Theory then Empirics”
To
“Basic empirics, basic theory, additional theory, additional
empirics”
That is you can easily change the theory-empirics ordering
around, or even integrate them in this case.
Best way is to trial this out using presentations. Initial guideline is
start
if much
stronger, otherwise start with theory
Nickwith
Bloom,empirics
Macro Topics, Spring
2007
Innovation Stylized Facts Overview
Nick Bloom, Macro Topics, Spring 2007
Empirical tests typically use one of two data sources
– R&D and patents
The standard (and more traditional) measure of macro, industry
and firm-level innovation is R&D – best starting place:
• Good: Measure $ of innovation inputs, time series
• Bad: Often not available (small/private/European firms)
More recently large number of papers using the NBER patents
database on around 6 millions patents plus their citations
• Good: Huge amounts of details on innovation, plus citations
(so can map out innovation process)
• Bad: Patents very stochastic (cite weighting helps), hard to
map to industries, hard to map to macro (patent office-cycles)
Note this data has been heavily used – “Blood out of stones”
Nick Bloom, Macro Topics, Spring 2007
Innovation has positive spillovers, particularly
geographically locally and within technology fields
Using the patent data a couple of key papers showed:
Innovation is spills over locally (Jaffe, Trajtenberg and Henderson,
QJE 1993).
• Use patents data to show citations concentrated in State
and Metropolitain area
• Used a control group of patents in same technology class
and year to control for agglomeration
Innovation spills over within industries (Jaffe, AER 1986)
• Firms patents allocated to technology classes
• Cross-firm correlation of allocation provides proxy of
technology closeness (do they overlap in technology space)
• R&D weighted by technology distance important in firm
Nick Bloom, Macro Topics, Spring 2007
performance (spills-over most from technology neighbours)
The first paper is in the Econ top 100 cited papers
Nick Bloom, Macro Topics, Spring 2007
Market size is an important determinant of
innovation – Schmookler (rather than Schumpeter)
Firms targets larger markets as rewards from innovation higher
Blundell, Griffith and Van Reenen (1999, RES) show that market
share plays a very significant role in determining innovation:
• Important to control for competition
Aghion and Linn (2006, QJE) show that market size plays an
important role in determining pharmaceutical innovation
• Instrument this with demographics to get identification
So in long-run growth models need something to offset size effects
(Jones, 1995 QJE)
Nick Bloom, Macro Topics, Spring 2007
The direction of technological change is
endogenous – Hick’s “induced innovation”
Refinement of the market size story is technology is multidimensional and can alter course
Acemoglu (1998 QJE, 2002 RESTUD ) model and estimate
technology change responding to the increased supply of skills
Newell, Jaffe and Stavins (QJE, 1999) show innovations in airconditioners respond to energy prices
Popp (2002, AER) shows a strong link between energy saving
patents and energy prices
Nick Bloom, Macro Topics, Spring 2007