Economics of Innovation Jaffes model Manuel Trajtenberg 2005

1
Economics of InnovationJaffes model Manuel
Trajtenberg2005
2
Technological Opportunity and Spillovers of RD
Evidence from Firms Patents, Profits, and Market
ValuebyAdam JaffeAER, Vol. 76 (5), Dec 1986,
984-1001
3
Introduction

• The main goal is to assess the effect of key
  supply side factors,
• spillovers from other firms, and
• technological opportunity, and the
  corresponding tech position of firms
• on the productivity of own RD, and profits
• Key issue how to define these concepts, how to
  actually measure them?

• .

4
The key concepts

• Spillovers
• presumed to have a positive technological
  effect,
• but a potentially negative economic effect
  through competition.
• Tech opportunity exogenous variations in
  costs and difficulty of innovation in different
  areas.
• Tech position of firms vis tech opportunities
  endogenous, but slowly moving
• Concepts not directly observable use patent
  data to measure them.

5
Front page of a patent (partial)
Frohman-Bentchkowsky, et. al. May 13,
1980 Electrically programmable and erasable MOS
floating gate memory device employing tunneling
and method of fabricating same Inventors
Frohman-Bentchkowsky Dov (Haifa, IL) Mar
Jerry (Sunnyvale, CA) Perlegos George
(Cupertino, CA) Johnson William S. (Palo Alto,
CA). Assignee Intel Corporation (Santa Clara,
CA). Current U.S. Cl. 365/185.29 257/321
326/37 327/427 Field of Search 365/185, 189
307/238 357/41, 45, 304
References Cited 3,500,142 Mar.,
1970 Kahng 365/185 4,051,464 Sept.,
1977 Huang 365/185 Primary Examiner Fears
Terrell W. 16 Claims, 14 Drawing Figures
6
Patent Class Definitions examples

• CLASS 365, STATIC INFORMATION STORAGE AND
  RETRIEVAL
• CLASS DEFINITION This is the generic class for
  apparatus or corresponding processes for the
  static storage and retrieval of information. For
  classification herein, the storage system must be
  (1) static, (2) a singular storage element or
  plural elements of the same type, (3)
  addressable.
• CLASS 257, ACTIVE SOLID-STATE DEVICES E.G.,
  TRANSISTORS, SOLID-STATE DIODES)
• CLASS DEFINITION This class provides for active
  solid-state electronic devices, , usually
  semiconductors, which operate by the movement of
  charge carriers - electrons or holes - which
  undergo energy level changes within the material
  and can modify an input voltage to achieve
  rectification, amplification, or switching
  action, and are not classified elsewhere.

7
The Technological Location of Firm i

• Characterize each firm by the vector
• fij the of firm i patents in patent
  category j.
• The USPTO patent classification system (purely
  tech based, not industries)
• 400 patent classes (328 back in the 1980s).
• 150,000 patent sub-classes
• Jaffe aggregated the 328 patent classes into 49
  patent categories.

8
Measuring spillovers
A measure of technological proximity between
firms the angular separation (or uncentered
correlation) of the vectors Fi and Fj
The potential spillover pool RDj the RD
expenditure of firm j
9
Data

• All patents granted to 1700 manufacturing firms,
  for 1969-79 260,000 patents
• Firms linked to Compustat
• Two tech positions one based on patents up
  to 1972, the second based on patents after 1972.
• Thus each firm two 49 elements vector, one for
  each period.

Fi1
Fi2
1969
1979
1972
10
http//www.compustat.com/www/

• Compustat Data Packages Compustat
  OfferingsgtCompustat DatagtCompustat North
  AmericagtCompustat GlobalgtCompustat
  XpressfeedgtCompustat Xpressfeed
  LoadergtCompustat HistoricalgtCompustat
  Unrestated QuarterlygtResearch Insight on the
  WebgtResearch InsightgtMarket InsightgtStandard
  Poor's Custom Business Unit

11
Establishing the tech position of firms

• Use clustering algorithm identify firms with a
  similar tech focus, so that they face the same
  state of technological opportunity based again
  on the vectors Fi s.
• Found 21 clusters done twice, pre- and post
  1972.
• About 1/3 of firms change clusters between the
  2 periods.

12
Technological clusters
13
The patenting equation

• Patents as a function of
• own RD - flow
• Spillovers Pool (RD of others, weighted by
  their tech proximity)
• Interaction between the two (absorptive
  capacity)
• Dummies for Tech clusters

14
The Profits Equation

• How to measure profits? Operating income before
  depreciation
• Profits as a function of
• own RD stock
• Spillovers Pool
• Interaction between the two
• Dummies for Tech clusters
• Capital
• Market share
• Market concentration C4

15
The Tobins q equation

• Tobins q Market value/Capital
• As a function of
• RD/Capital
• (RD/Capital) x Spillovers Pool
• Dummies for Tech clusters
• Market share
• Market concentration C4

16
The estimating equations
17
Statistics for Regression Variables
Note All nominal variables are millions of 1972
dollars. Market share and concentration are
percentages.
18
Data issues

• 432 firms, 1/3 of those that report RD, but
  they account for 95 of total RD (possible
  selectivity bias against small firms).
• Two cross-sections, centered on 1973 and 1979.
  Each cross section average of 3 years data,
  smoothing.
• RD stocks (for the profits and market value
  equations) computed assuming 15 depreciation
  of RD extrapolation into the past using
  average growth rate.
• Market shares only for 1972 proxies for 1979.

19
Econometric issues

• Endogeneity of RD, Capital, market share shock
  (e.g. unobserved management skill) may both
  lead to higher patents, profits, etc. and to
  higher investments and market share.
• Measurement error in some of the Xs, e.g. RD
  (because assume only contemporaneous effect for
  patents, the way the stock is constructed, etc.).

20
Estimation

• First estimates OLS, but as said possible
  endogeneity.
• Estimate first differences between two cross
  sections (like fixed effects) but just 2 cross
  sections, very imprecise, may exacerbate errors
  in variables problem
• Bring in instruments, estimate 3SLS like 2SLS,
  but also system of equations to take into account
  possible correlations of error terms across
  equations. White s.e.

21
Instruments

• Need IVs for RD, Capital and Market Share. Use
  Industry variables (given for the individual
  firm). Each firm belongs to a bunch of SIC,
  according to its sales, so take the weighted
  average that gives variation across firms even
  if similar.
• Industry RD, Sales, growth rate
• Industry MSE - minimum efficient scale (IV for
  Capital)
• Spillover pool (and interactions between it and
  the industry variables)

22
Table 5 3SLS Estimates - 432 obs(elasticities)
S spillover pool (zero mean)
23
Table 5 cont.
24
Results for the patents equation

• Elasticity of patents w.r.t. RD 0.88 for
  average firm, higher for those with
  above-average spillover pool.
• Elasticity w.r.t. to S-pool 0.51 0.35
  log(RD) for those with mean RD (1.8) 1.1
  very large!
• Thus, if everybody increases RD by 10, total
  patents increase by 20 (0.0880.035x(1.8x1.1)0
  .051).
• Return of 2 patents per million own RD,
  0.6 patents per 10 M of others RD.

25
Results for the profits equation
Compute gross rate of return to RD (or to
capital) start with estimated elasticity of RD
Gross rate of return to RD (mean ?/mean RD
stock) 208/121 1.7 Estimated elasticity x
1.7 0.18 x 1.7 0.31 (in Jaffe it is 28,
difference apparently because of mean of the
ratio, not ratio of the means)
26
Profit equation cont.

• Gross rate of return to capital
• Mean profits/mean capital 208/968 0.21
• times elasticity of capital 0.21 x 0.825 0.18
• (in Jaffe it is 15)
• Thus, returns to RD (0.31) almost 2 times larger
  than return to conventional capital (but much
  higher depreciation).

27
Profit equation cont. 2

• Spillover pool negative direct effect (-0.095,
  only marginally significant), but positive
  through interaction (0.058). Net effect for
  average log(RD)
• -0.095 0.058 (3.09) 0.084
• For firms with little RD (below half s.d. of
  mean log RD) negative net effect

28
Tobins q equation

• Similar to profit equation negative direct
  effect of S, positive interaction firms doing
  lots of RD benefit from spillovers pool.
• An RD dollar increases market value 3 times as
  much as a dollar invested in physical capital
  (see coefficient of RD/Capital).
• C(4) decreases market value, own market share
  increases q.

29
Pooled OLS Estimates for 1973 and 1979 (846 Obs)
30
OLS cont.
Lower coefficients for RD in OLS because
endogeneity expected positive bias, but errors in
measurement the other way.
31
Technological clusters tech opportunity

• High correlation over time of the tech dummy
  coefficients from the patents equation, low corr.
  for the profits or market value
• Technological persistence, but high profits in
  clusters in 1973 get competed away by 1979.
  Indeed,
• positive corr. between net entry into clusters,
  and size of coefficients in 1973
• negative corr. between change in dummies in
  Tobins q equation, and entry (i.e. entry lowers
  market value).

32
Key contributions

• Method to measure spillovers, tech position and
  tech proximity (used since).
• Findings
• Large impact of spillovers in 3 equations
  positive for patents, negative (direct) for
  profits and market value, but positive if do at
  least average RD.
• Importance of absorptive capacity.
• Large (private) returns from RD (twice as large
  as for physical investment).