The Role of the IT Revolution in Knowledge Diffusion, Innovation … · 2019. 8. 15. · Explore...

76
Introduction Empirical Evidence Model Quantitative Analysis Conclusion The Role of the IT Revolution in Knowledge Diffusion, Innovation and Reallocation Salome Baslandze Einaudi Institute (EIEF) November 24, 2015

Transcript of The Role of the IT Revolution in Knowledge Diffusion, Innovation … · 2019. 8. 15. · Explore...

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    The Role of the IT Revolution in

    Knowledge Diffusion, Innovation and Reallocation

    Salome BaslandzeEinaudi Institute (EIEF)

    November 24, 2015

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Introduction

    The IT Revolution: large penetration of Information andCommunications Technologies (ICT) over the last four decades.

    What is the impact of the IT revolution on aggregate productivity andsectoral reallocation of economic activities in the U.S.?

    • ”Direct Impact”: main focus of the literature.

    • Cost reductions or changes in business organizations.

    • ”Indirect Impact”: new channel in this paper.

    • Diffusion of knowledge, flow of ideas.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Introduction

    The IT Revolution: large penetration of Information andCommunications Technologies (ICT) over the last four decades.

    What is the impact of the IT revolution on aggregate productivity andsectoral reallocation of economic activities in the U.S.?

    • ”Direct Impact”: main focus of the literature.

    • Cost reductions or changes in business organizations.

    • ”Indirect Impact”: new channel in this paper.

    • Diffusion of knowledge, flow of ideas.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Introduction

    The IT Revolution: large penetration of Information andCommunications Technologies (ICT) over the last four decades.

    What is the impact of the IT revolution on aggregate productivity andsectoral reallocation of economic activities in the U.S.?

    • ”Direct Impact”: main focus of the literature.

    • Cost reductions or changes in business organizations.

    • ”Indirect Impact”: new channel in this paper.

    • Diffusion of knowledge, flow of ideas.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Tradeoff from Knowledge Diffusion

    Two opposing effects from increased knowledge diffusion:

    • Positive (Information) “Learning effect”:knowledge flowing to you;

    • Negative (Information) “Competition effect”:knowledge flowing to potential competitors.

    Paper’s Hypothesis:The overall effect from ICT should depend on an industry’stechnological characteristic, external knowledge dependence(Ex: electronic test & measurement instruments vs fabrics industry)

    I IT Revolution: Biased towards externally dependent industries.Induces sectoral reallocation.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Tradeoff from Knowledge Diffusion

    Two opposing effects from increased knowledge diffusion:

    • Positive (Information) “Learning effect”:knowledge flowing to you;

    • Negative (Information) “Competition effect”:knowledge flowing to potential competitors.

    Paper’s Hypothesis:The overall effect from ICT should depend on an industry’stechnological characteristic, external knowledge dependence(Ex: electronic test & measurement instruments vs fabrics industry)

    I IT Revolution: Biased towards externally dependent industries.Induces sectoral reallocation.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    This Paper

    Study the impact of the IT revolution on reallocation and growth1. Empirics

    • Explore knowledge diffusion with ICT using citations data.• Develop a novel measure of external knowledge dependence.• Document reallocation towards externally dependent industries.

    2. Theory• General equilibrium model of technological change.• Firms heterogeneous with respect to knowledge dependence.• IT revolution increases information accessibility and competition

    3. Quantitative Analysis• The IT revolution accounts for (1976-2003):

    • at least 3/4 of the observed sectoral reallocation;• 76% of an observed increase in LP growth. Dominant sustainable

    channel: ”indirect” impact from ICT.• Negative competition channel significantly dampens LP growth

    from ICT.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Literature/Contribution• Empirical literature on the impact of ICT on productivity

    (Stiroh (2002), Jorgenson, et al (2005), Kleis, et al (2012), Dedricket al (2003), Acemoglu, et al (2014), Brynjolfsson, Hitt (2000),among others).

    • New empirical facts; study a new channel through knowledgediffusion.

    • Endogenous growth and firm dynamics (Aghion, Howitt(1992), Klette, Kortum (2004), Akcigit, Kerr (2010), Lentz,Mortensen (2008), Aghion, et al (2014), Cai, Li (2012), amongothers ).

    • Here, introduce technological interdependence, industryheterogeneity, knowledge diffusion.

    • Structural change and transformation (Kongsamut, et al. (2001),Ngai, Pissarides (2007), Herrendorf, et al. (2014), Duarte,Restuccia (2010), Buera, Kaboski (2012), among others)

    • Here, new IT technologies are biased towards externallydependent industries. Change in endogenous productivitiesarise due to differential impact by ICT.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Empirical Evidence

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Data

    • ICT data from BEA non-residential fixed capital dataset by typeof assets and industry. NIPA price indices for each asset.

    • Patents and citations data from NBER, 1976-2003.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Patent: Compact Hand-Held Video Game System

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Patent: Compact Hand Held Video Game System

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Patent: Compact Hand Held Video Game System

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Evidence for Knowledge Diffusion

    Idea: Keep track of knowledge flows through patent citations:

    Technology classes cite external patents more intensively over time.

    Citations Input-Output Matrix

    1976-1978

    Cited Technology Class (ranked)

    Citi

    ng T

    echn

    olog

    y C

    lass

    2001-2003

    Cited Technology Class (ranked)

    Citi

    ng T

    echn

    olog

    y C

    lass

    0

    0.1

    0.2

    0.3

    0.4

    0.5

    0.6

    >0.7

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Evidence for Knowledge Diffusion

    Idea: Keep track of knowledge flows through patent citations:Technology classes cite external patents more intensively over time.

    Citations Input-Output Matrix

    1976-1978

    Cited Technology Class (ranked)

    Citi

    ng T

    echn

    olog

    y C

    lass

    2001-2003

    Cited Technology Class (ranked)

    Citi

    ng T

    echn

    olog

    y C

    lass

    0

    0.1

    0.2

    0.3

    0.4

    0.5

    0.6

    >0.7

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Evidence for Knowledge Diffusion

    Idea: Keep track of knowledge flows through patent citations:Technology classes cite external patents more intensively over time.

    Citations Input-Output Matrix

    1976-1978

    Cited Technology Class (ranked)

    Citi

    ng T

    echn

    olog

    y C

    lass

    Nuclear X-ray industry

    Nuclear X-ray

    Measuring & Testing

    2001-2003

    Cited Technology Class (ranked)

    Citi

    ng T

    echn

    olog

    y C

    lass

    0

    0.1

    0.2

    0.3

    0.4

    0.5

    0.6

    >0.7

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Evidence for Knowledge Diffusion

    Idea: Keep track of knowledge flows through patent citations:Technology classes cite external patents more intensively over time.

    Citations Input-Output Matrix

    1976-1978

    Cited Technology Class (ranked)

    Citi

    ng T

    echn

    olog

    y C

    lass

    Nuclear X-ray industry

    Nuclear X-ray

    Measuring & Testing

    2001-2003

    Cited Technology Class (ranked)

    Citi

    ng T

    echn

    olog

    y C

    lass

    Nuclear X-ray industry

    Nuclear X-ray

    Measuring & Testing Organic Compounds

    Biotech

    0

    0.1

    0.2

    0.3

    0.4

    0.5

    0.6

    >0.7

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Evidence for Knowledge Diffusion

    Idea: Keep track of knowledge flows through patent citations:Technology classes cite external patents more intensively over time.

    Citations Input-Output Matrix

    1976-1978

    Cited Technology Class (ranked)

    Citi

    ng T

    echn

    olog

    y C

    lass

    2001-2003

    Cited Technology Class (ranked)

    Citi

    ng T

    echn

    olog

    y C

    lass

    0

    0.1

    0.2

    0.3

    0.4

    0.5

    0.6

    >0.7

    Self-citations share ↓ by 16% (finer classes: ↓ 26% & # new classes cited ↑ 4.4 times)Finer classes Standardized Reg. ICT

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Index of External Knowledge Dependence

    Define External Knowledge Dependence for a class j (412 classes)based on 2001-2003 patents:

    EKD Index =[# classes cited]∀ patent i∈j

    N0

    .51

    1.5

    2Density

    0 .2 .4 .6 .8ExternalKnowledgeDependenceDistribution of External Knowledge Dependence Examples Alternatives

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Evidence for Sectoral Reallocation in the Data

    • Since the IT revolution, more externally dependent industries• have increased patenting activities,• have increased shares in real activities.

    Log Patent Counts

    -.50

    .51

    1976 1980 1985 1990 1995 2000

    Coef

    ficien

    t on

    Year

    Dum

    my

    Year

    Top 25% external dep.

    Bottom 25% external dep.

    Share of Value Added in GDP

    910

    1112

    13S

    hare

    of V

    alue

    Add

    ed

    1975 1985 1995 2005year

    Top 25% external dep.

    Bottom 25% external dep.

    Log Patn = β0 + βtYeart + jFE + ε OldClasses NoIT

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Evidence for Sectoral Reallocation in the Data

    • Since the IT revolution, more externally dependent industries• have increased patenting activities,• have increased shares in real activities.

    Log Patent Counts

    -.50

    .51

    1976 1980 1985 1990 1995 2000

    Coef

    ficien

    t on

    Year

    Dum

    my

    Year

    Top 25% external dep.

    Bottom 25% external dep.

    Share of Value Added in GDP

    910

    1112

    13S

    hare

    of V

    alue

    Add

    ed

    1975 1985 1995 2005year

    Top 25% external dep.

    Bottom 25% external dep.

    Log Patn = β0 + βtYeart + jFE + ε OldClasses NoIT

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Summary of the Empirics

    1. Increased knowledge diffusion with ICT;

    2. Heterogeneity of industries in their external knowledgedependence;

    3. Reallocation of economic activities towards more externallydependent industries;

    4. Increased competition with ICT; Details

    5. Heterogeneous technological spillovers from ICT. Details

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    MODEL

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Overview

    A new model of endogenous technological change:• Quality ladder;• Schumpeterian creative destruction.

    New features:• R&D combines external knowledge with R&D expenses;

    • Industry heterogeneity wrt knowledge dependence;

    • ICT governing the access to external knowledge;

    • Higher diffusion leads to:• ”Learning” effect,• ”Competition” effect.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model

    • Representative household with logarithmic utility

    U =∞∫

    0

    exp(−ρt)logC(t)dt,

    • Final good produced using continuum of intermediate goods

    logYt =∫ 1

    0logy(j, t)dj

    • Perfect competition in the final good sector. Price of Y(t)normalized to 1.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Intermediate Goods

    • Intermediate good j is produced using a linear technology

    y(j, t) = q(j, t)f (ICT(j, t))l(j, t),

    bla

    • Competing follower able to produce at q−i(j, t)⇒firm i produces if qi(j, t) > q−i(j, t).

    • Limit pricing at p(j, t) = w(t)q−i(j,t)f (ICT−i(j,t))• Πi(j, t) = (p(j, t)−MCi(j, t))y(j, t)

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Intermediate Goods

    • Intermediate good j is produced using a linear technology

    y(j, t) = q(j, t)f (ICT(j, t))l(j, t),

    q- endogenous labor productivity term

    • Competing follower able to produce at q−i(j, t)⇒firm i produces if qi(j, t) > q−i(j, t).

    • Limit pricing at p(j, t) = w(t)q−i(j,t)f (ICT−i(j,t))• Πi(j, t) = (p(j, t)−MCi(j, t))y(j, t)

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Intermediate Goods

    • Intermediate good j is produced using a linear technology

    y(j, t) = q(j, t)f (ICT(j, t))l(j, t),

    f (ICT)-direct productivity impact of ICT

    • Competing follower able to produce at q−i(j, t)⇒firm i produces if qi(j, t) > q−i(j, t).

    • Limit pricing at p(j, t) = w(t)q−i(j,t)f (ICT−i(j,t))• Πi(j, t) = (p(j, t)−MCi(j, t))y(j, t)

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Intermediate Goods

    • Intermediate good j is produced using a linear technology

    y(j, t) = q(j, t)f (ICT(j, t))l(j, t),

    l- labor input

    • Competing follower able to produce at q−i(j, t)⇒firm i produces if qi(j, t) > q−i(j, t).

    • Limit pricing at p(j, t) = w(t)q−i(j,t)f (ICT−i(j,t))• Πi(j, t) = (p(j, t)−MCi(j, t))y(j, t)

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Intermediate Goods

    • Intermediate good j is produced using a linear technology

    y(j, t) = q(j, t)f (ICT(j, t))l(j, t),

    text

    • Competing follower able to produce at q−i(j, t)⇒firm i produces if qi(j, t) > q−i(j, t).

    • Limit pricing at p(j, t) = w(t)q−i(j,t)f (ICT−i(j,t))• Πi(j, t) = (p(j, t)−MCi(j, t))y(j, t)

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Intermediate Goods

    • Intermediate good j is produced using a linear technology

    y(j, t) = q(j, t)f (ICT(j, t))l(j, t),

    text

    • Competing follower able to produce at q−i(j, t)⇒firm i produces if qi(j, t) > q−i(j, t).

    • Limit pricing at p(j, t) = w(t)q−i(j,t)f (ICT−i(j,t))• Πi(j, t) = (1− q−iqi )Y(t)

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Productivity Evolution

    – Law of motion upon innovation:

    qi(j, t + ∆t) = λqi(j, t), where λ > 1

    – i innovates n times after −i:

    qi(j, t) = λnq−i(j, t)

    – nj – technology gap.– Πi(j, t) = (1− λ−n)Y(t)

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Productivity Evolution

    – Law of motion upon innovation:

    qi(j, t + ∆t) = λqi(j, t), where λ > 1

    – i innovates n times after −i:

    qi(j, t) = λnq−i(j, t)

    – n – technology gap.– Πi(j, t) = (1− λ−n)Y(t)

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Productivity Evolution

    – Law of motion upon innovation:

    qi(j, t + ∆t) = λqi(j, t), where λ > 1

    – i innovates n times after −i:

    qi(j, t) = λnq−i(j, t)

    – n – technology gap.

    – Πi(j, t) = (1− λ−n)Y(t)

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Types of R&D

    • Innovation is a stochastic outcome from R&D activities.

    • Two types of innovation:

    • Vertical (xj) – own quality improvement:Tech. gap becomes nj(t + ∆t) = nj(t) + 1.

    • Horizontal (zj) – improves quality in other line:Tech. gap becomes nl(t + ∆t) = 1.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. R&D

    • To conduct R&D, firms combine external knowledgewith R&D expenditures.

    • Product lines have heterogeneous knowledge productiontechnologies – defines their type s.

    • Product line j has an external knowledge dependencedistribution with density fj distributed on (j−

    δj2 , j +

    δj2 ).

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Technological Circle and Knowledge Dependence

    Large

    extern

    al

    depend

    ence Small external

    dependence

    ε – exogenous knowledge diffusion parameter.

    Ē(j, ε) =∫ j+ εt2

    j− εt2fjdj

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Technological Circle and Knowledge Dependence

    ε – exogenous knowledge diffusion parameter.

    Ē(j, ε) =∫ j+ εt2

    j− εt2fjdj

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Technological Circle and Knowledge Dependence

    ε- exogenous knowledge diffusion parameter.

    Ē(j, ε) =∫ j+ εt2

    j− εt2fjdj – share of knowledge utilized.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Vertical R&D

    • For a firm of type s, to achieve a Poisson arrival rate of verticalinnovation x, it needs to invest:

    Costverts,t =αs

    Ē(s, t)φxγλ−n,

    αs- technological efficiency of type s; Ē(s, t) ≤ 1.

    • “Learning effect”: ICT increases knowledge diffusion ε –positive technological externalities: Ē(s, t)→ 1.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Horizontal R&D• For a firm of type s, to achieve a Poisson arrival rate of

    horizontal innovation z, it needs to invest:

    Costhorizs,t =αsβ

    Ē(s, t)ψzγ

    αs- technological efficiency of type s; Ē(s, t) ≤ 1.• Innovate on a random product line i ∈ (j− ε2 , j +

    ε2 ).

    Replace incumbent. Sell the line for price p (bargaining).• As a result, each product line faces creative destruction

    τjt =

    j+ εt2∫j− εt2

    zitεt

    di

    • “Competition effect”: τ increases with a rise in knowledgediffusion, ε.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Steady State

    • Finite number of industry types s ∈ 1...S having same δ and αs(so, equal Ē(s, t)).

    • Assumption: all types distributed evenly on the technologicalcircle⇒ Location j is not a state variable.

    • Consider a model in steady state.• (n, s) – state vector

    • Denote a normalized value of a firm by v(n, s).

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Firm’s Value Function

    ρv(n, s) = maxx(n,s),z(n,s)

    π(n)− αsĒ(s, t)φ

    x(n, s)γλ−n︸ ︷︷ ︸vertical R&D cost

    − αsβĒ(s, t)ψ

    z(n, s)γ︸ ︷︷ ︸horizontal R&D cost

    + x(n, s)(v(n + 1, s)− v(n, s))︸ ︷︷ ︸successful vertical innovation

    + z(n, s)p︸ ︷︷ ︸horizontal innovation

    − τv(n, s)︸ ︷︷ ︸creative destruction

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Steady State

    • Proposition: i) The value function is linearly separable:

    v(n, s) = A(s)− B(s)λ−n

    where

    A(s) =1 + zsp− αsβĒ(s,t)ψ z

    γs

    ρ + τ,

    B(s) =1 + xγs αsĒ(s,t)φ

    ρ + τ + xs λ−1λ,

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Steady State

    • Proposition: i) The value function is linearly separable:

    v(n, s) = A(s)− B(s)λ−n

    ii) Type-specific innovation: x(n, s) = x(s) and z(n, s) = z(s).

    • Stationary distribution over the state space:

    µ(n, s) =1S

    (xs

    xs + τ

    )n−1τ

    xs + τ

    • Equilibrium aggregate growth:

    g = logλ(

    ∑s xsS

    + τ

    )+

    ∆F(ICT)F(ICT)

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Model. Steady State

    • Proposition: i) The value function is linearly separable:

    v(n, s) = A(s)− B(s)λ−n

    ii) Type-specific innovation: x(n, s) = x(s) and z(n, s) = z(s).

    • Stationary distribution over the state space:

    µ(n, s) =1S

    (xs

    xs + τ

    )n−1τ

    xs + τ

    • Equilibrium aggregate growth:

    g = logλ(

    ∑s xsS

    + τ

    )+

    ∆F(ICT)F(ICT)

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Quantitative Analysis

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Setup

    • Estimate technological parameter δ (triangular knowledgedependence distribution). Four types of industries, S = 4.

    • Parameterize in two stages:• Outside the model;• Calibrate the model to the initial steady state in 1976-1978.

    • Estimate diffusion series εt from the data.

    • Feed in exogeneous series of εt and simulate the model overtime, 1976-2003.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Technological Distribution in the Data

    • δj estimated based on the latest period, 2001-2003.

    EKD Index =[# classes cited]∀ patent i∈j

    N

    0.5

    11.5

    2Density

    0 .2 .4 .6 .8Delta

    • 4 types: δ1, δ2, δ3, δ4 – quartiles of the distribution.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Estimated Direct Impact of ICT

    • Recall the production function

    y(j, t) = q(j, t)f (ICT(j, t))l(j, t),

    • Functional form: f (ICT) = ICTκ.

    • Convert to estimate the following regression:

    gLPjt = 0.02gICTjt + gPatentsjt + controls(0.009)

    • Point estimate comparable to the elasticities from the literature(Brynjolfsson and Hitt (2000), [0.01, 0.04]).

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Moments and Estimates• Rest of the 9 parameters α1, α2, α3, α4, β, φ, ψ, γ, λ are calibrated

    to match the following moments:

    Moment Values

    Targets Data Model

    Aggregate Growth1976 1.31% 1.30%

    Average HorizontalInnovation1976

    0.22 0.19

    R&D Intensity1976 0.09 0.05Vertical Innovation

    Horizontal Innovation 1976 1.20 1.24[ Innov4Innov1

    , Innov3Innov1 ,Innov2Innov1

    ]1976 [2.25, 1.15, 1.53] [2.31, 1.09, 1.53][ Innov4

    Innov1, Innov3Innov1 ,

    Innov2Innov1

    ]1977 [2.34, 1.12, 1.68] [2.33, 1.05, 1.61][ Innov4

    Innov1, Innov3Innov1 ,

    Innov2Innov1

    ]1978 [2.56, 1.37, 1.85] [2.38, 1.18, 1.72]

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Moments and Estimates

    • Resulting estimates are:

    Calibrated Parameters

    Parameters Meaning Value

    [α1, α2, α3, α4] Scaling of vertical R&D [2.02, 0.5, 0.52, 0.05]

    β Scaling of horizontal R&D 6.47

    λ Step size of innovation 1.03

    φ Vertical R&D curvatureto external knowledge

    2.48

    ψ Horizontal R&D curvatureto external knowledge

    2.10

    γ R&D cost curvature 2.98

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Estimated Diffusion Parameter

    Estimate knowledge diffusion as a function of ICT:

    ε̄jt = β0 + β1ICTjt + β2ICTothersjt + pat stockjt + cap stockjt + j FE+ time+ ejt

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Estimated Diffusion Parameter

    Estimate knowledge diffusion as a function of ICT:

    εt = Meanj(β̂0 + β̂1ICTjt + β̂2ICTothersjt )

    .1.2

    .3.4

    .5.6

    Ave

    rage

    Eps

    ilon

    1975 1980 1985 1990 1995 2000Year

    Knowledge Diffusion over Time

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Experiment. Reallocation of Innovation Activities

    Total Innovation Trends by Type of Industry

    01

    23

    4T

    otal

    Inno

    vatio

    n

    1975 1980 1985 1990 1995 2000Year

    Top 25% (model) Bottom 25% (model)Bottom 25% (data) Top 25% (data)

    Time Change Data Model ExplainedTop 25%

    Bottom 25% 2.35 times 1.75 times 74%

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Experiment. Reallocation of Real Activities

    Sharet(s) =Qt(s)ICTκt (s)∑n λ

    −nµ(n, s)∑s Qt(s)ICTκt (s)∑n λ−nµ(n, s)

    ,

    Share of Total Output by Type of Industry12

    1416

    18S

    hare

    of O

    utpu

    t (V

    A)

    1976 1981 1986 1991 1996 2001Year

    Bottom 25% (model) Top 25% (model)Bottom 25% (data) Top 25% (data)

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Growth of Labor Productivity

    Growth of Labor Productivity. Data and Model

    -20

    24

    6G

    row

    th R

    ate

    of L

    abor

    Pro

    duct

    ivity

    1975 1980 1985 1990 1995 2000 2005Year

    Model Data

    Change in growth Data Model ExplainedBefore/After 1990 ↑ 45% ↑ 34% 75.6%

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Growth Decomposition

    Decomposition of ICT’s Impact

    1.2

    1.4

    1.6

    1.8

    22.2

    Growth

    1975 1980 1985 1990 1995 2000Year

    Before 1990 After 1990Average:28.7 bp

    �Indirect 24%Direct 76%

    Average:78 bp

    �Indirect 70%Direct 30%

    1

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Growth Decomposition

    Decomposition of ICT’s Impact

    1.2

    1.4

    1.6

    1.8

    22.2

    Growth

    1975 1980 1985 1990 1995 2000Year

    Direct channel from ICT

    Before 1990 After 1990Average:28.7 bp

    �Indirect 24%Direct 76%

    Average:78 bp

    �Indirect 70%Direct 30%

    1

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Growth Decomposition

    Decomposition of ICT’s Impact

    1.2

    1.4

    1.6

    1.8

    22.2

    Growth

    1975 1980 1985 1990 1995 2000Year

    Indirect channel from ICT

    Direct channel from ICT

    Before 1990 After 1990Average:28.7 bp

    �Indirect 24%Direct 76%

    Average:78 bp

    �Indirect 70%Direct 30%

    1

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Counterfactuals

    Counterfactual experiment to quantify negative competition effect:

    Growth in 2003. Three scenarios.

    No growthin ICT

    Total impact of ICT(direct+indirect)

    ICT withoutcompetitive spillovers

    1.16% 2.02% 2.60%

    Solow: ”You can see the computer age everywhere but in the productivitystatistics.”

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Extension in Progress

    Cross-Country Analalysis:• Increased divergence in growth between the U.S. and EU from

    90s.• Could it be because of different effects the ICT technologies

    played?• EUKLEMS cross-country data on output, productivities, ICT by

    industries.• Look at the composition of industries in EU relative to the U.S.• Esimate evolution of εt in EU.• Similate the model under different scenarios: EU/US sectoral

    composition and EU/US ICT evolution.• Results:

    • contribution of different levels of ICT investment into differentialgrowth.

    • contribution of different sectoral composition to differential growth.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Conclusion

    • Study the role of ICT as a tool for knowledge dissemination inthe economy.

    • Explore the impact of this new role on productivity growth andsectoral reallocation.

    Other avenues:

    • Endogenizing the IT revolution.

    • Cross-country analysis.

    • Analysis of industrial policies.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Appendix

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Evidence for Knowledge DiffusionFiner technology categories- 412 (nclass)

    Citations Input-Output Matrix

    1976-1978

    Cited Technology Class (ranked)

    Citi

    ng T

    echn

    olog

    y C

    lass

    2001-2003

    Cited Technology Class (ranked)

    Citi

    ng T

    echn

    olog

    y C

    lass

    0

    0.1

    0.2

    0.3

    0.4

    0.5

    0.6

    >0.7

    Note: Each row represents one of the 412 technology categories and each cell depicts a share of citations which are given bypatents in citing technology class to corresponding cited technology class. All shares in each row add up to one. Citedtechnology classes on the horizontal axis are ranked by citation shares received.

    Back

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Evidence for Knowledge Diffusion

    Standardize citations tables using Sinkhorn-Knopp (1976)algorithm. Get same marginal distributions (2001-2003).

    Citations Input-Output Matrix. Standardized

    1976-1978

    Cited Technology Class (ranked)

    Citi

    ng T

    echn

    olog

    y C

    lass

    2001-2003

    Cited Technology Class (ranked)

    Citi

    ng T

    echn

    olog

    y C

    lass

    0

    0.1

    0.2

    0.3

    0.4

    0.5

    0.6

    >0.7

    Back

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Knowledge Diffusion and ICT

    Num. externalclasses cited

    Externalcitations share

    Log ICT0.151∗∗∗

    (0.003)0.042∗∗∗

    (0.012)Patent Stock X XCapital Stock X XNum. firms X XYear, Class FE X XObservations 11, 286 11, 286R2 0.86 0.85Method Poisson OLS

    Notes: Industry (nclass) × Year regressions. Standard errors clustered bytechnology classes. *p < 0.10, **p < 0.05, ***p < 0.01 Back

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    External Knowledge Dependence0

    .51

    1.5

    2D

    ensi

    ty

    0 .2 .4 .6 .8External Knowledge Dependence

    Textiles

    Wire,fabrics,and,structure

    Tuners

    Metal,tools,andimplements,making

    Railway,draft,appliances

    Wireworking

    ,,,,,,,,,,,,,,,,,FertilizersSingle,generator,,systems

    Compound,tools

    BridgesBooks,,strips

    Organic,compounds

    ElectricityMetallurgical,apparatus

    CryptographgyInduced,nuclear,reactions

    AcousticsTelevision

    Communications:,directive,radio,waves

    Data,processing:Static,info,storage

    Synthetic,resins

    Electrical,Computers,and,othersX-ray,or,gamma,ray,systemsEducation,and,demonstration

    Hidraulic,and,Earth,engineering

    Measuring,and,testing

    Data,processing:

    Communications:,electrical

    Molecular,biology

    Semiconductors

    Distribution of External Knowledge DependenceBack

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Reallocation of Innovation. Old patent classes

    • Identify some of the old classes from patents prior to 1960.• Distribution of external dependence for the old classes (245

    classes): (quartiles 1-12%, 2-25%, 3-30%, 4-33%)

    Log Patent Counts. Old Classes-.8

    -.6-.4

    -.20

    .2.4

    .6.8

    1C

    oeffi

    cien

    ts o

    n Y

    ear D

    umm

    ies

    1976 1980 1985 1990 1995 2000Year

    Top 25% external dep.

    Bottom 25% external dep.

    Back

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Reallocation of Innovation. Without ICT classes

    • Look at the trends excluding ICT technology classes.

    Log Patent Counts. No ICT-classes.-.6

    -.4-.2

    0.2

    .4.6

    .8

    1976 1980 1985 1990 1995 2000

    Top 25% external dep.

    Bottom 25% external dep.

    Back

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    ICT and Heterogenous Technological Spillovers

    Num. externalclasses cited

    Externalcitations share

    Log ICT0.062∗∗∗

    (0.004)−0.008∗∗(0.004)

    Log ICT × Ext. Dep. 0.045∗∗∗

    (0.007)0.028∗∗∗

    (0.003)Patent Stock X XCapital Stock X XNum. firms X XYear, Class FE X XObservations 11, 286 11, 286R2 0.89 0.88Method Poisson OLS

    Notes: Industry (nclass) × Year regressions. Standard errors clustered bytechnology classes. *p < 0.10, **p < 0.05, ***p < 0.01 Back

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    ICT and Competition

    # Entrants(new)

    # Entrants(other incumb.)

    Log (Herf index)

    Log ICT0.361∗∗∗

    (0.020)0.550∗∗∗

    (0.079)−0.389∗∗∗(0.028)

    Num. incumb. X X XYear, Class FE X X XObservations 9, 682 9, 682 11, 286R2 0.76 0.74 0.86Method Poisson Poisson OLS

    Notes: Industry (nclass) × Year regressions. Standard errors clustered bytechnology classes. *p < 0.10, **p < 0.05, ***p < 0.01 Back

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Alternative Indexes ofExternal Knowledge Dependence

    • Benchmark definition on 1976-2003:Spearman′s corr = 0.98

    • Benchmark definition on 1976-1978:Spearman′s corr = 0.81

    • Alternative definition:

    ∑patent i∈j # classes citedi412×Nj

    Spearman′s corr = 0.52Back

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Equilibrium

    Definition (Steady-State Equilibrium) Given the exogenous allocationof ICT across product lines and the parameter of the diffusion process, εt,an equilibrium of the economy consists of{x∗s , z∗s , µ(n, s)∗, p∗(n, s), y∗(n, s), Y∗, w∗, τ∗, g∗, r∗}s.t.: (i) Aggregate output Y∗t is given by final output specification;(ii) Intermediate goods prices and output p∗t (n, s), y

    ∗(n, s) satisfy demandand price setting equations;(iii) Wage w∗ clears the labor market;(iv) Innovation decisions x∗s , z∗s maximize a firm’s value;(v) Equilibrium creative destruction τ∗ is consistent with eq. horizontalinnovation;(vi) Distribution µ∗(n, s) is consistent with eq. inflow-outflow dynamics;(vii) r∗ satisfies the Euler equation;(viii) g∗ results from eq. horiz. and vert. innovations and ICT growth.

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    Vt(n, s) =

    [Πt(n)− αsĒ(s,t)φ xt(n, s)

    γλ−n

    − αsβĒ(s,t)ψ zt(n, s)γ

    ]∆t + o(∆t)

    +e−rt+∆t∆t

    (xt(n, s)∆t + o(∆t))Vt+∆t(n + 1, s)

    +(zt(n, s)∆t + o(∆t))(pt+∆t + Vt+∆t(n, s))+(τt∆t + o(∆t))× 0

    (1− xt(n, s)∆t− zt(n, s)∆t− τt∆t− o(∆t))Vt+∆t(n, s)

  • Introduction Empirical Evidence Model Quantitative Analysis Conclusion

    r(t)Vt(n, s)− V̇t(n, s) = maxxt(n,s), zt(n,s)

    Πt(n)− αsĒ(s,t)φ xt(n, s)

    γλ−n

    − αsβĒ(s,t)ψ zt(n, s)γ

    +xt(n, s)(Vt(n + 1, s)−Vt(n, s))+zt(n, s)pt − τtVt(n, s)

    IntroductionEmpirical EvidenceModelQuantitative AnalysisConclusion