Green Innovation
an empirical analysis of technology, skills and policy
François Perruchas
PhD tutors: Davide Consoli, Nicolò Barbieri
Introduction
Increasing concern about environmental impact of human activities
- 1992: Rio de Janeiro summit - creation of the IPCC
- 1997: Kyoto protocol
- 2012: creation of the IPBES
- 2015: Paris Agreement
⇒ Goal: contain increase in global average temperature below 2ºC
Motivation
Greenhouse gases (GHG) emissions
- 2018: 37.1 Gtons of CO2 [↑ 2.7% since 2017]
- ≈ 2070: net 0 emissions of GHG
↳ Green Innovation needed to...
- decrease emissions (mitigation)
- deal with temperature increases (adaptation)
- protect biodiversity and eco-systems
- accommodate social changes triggered
⇒ Goal: sustainable societies in 50 years
↪ Thesis: Empirical analysis of green innovation
Background
- Innovation rests on the ability to generate, diffuse, access and re-use knowledge (Romer, 1994; Weitzman, 1998)
- Interaction among agents crucial but individuals have limited access to information, and imperfect capacity to absorb, process, and respond to information (Nelson & Winter, 1982; Cohen & Levinthal, 1990)
- The latent potential of innovation depends on circumstances of time and place that facilitate (or not) the application of useful knowledge to specific ends (Atkinson & Stiglitz, 1969; David, 1974; Antonelli, 2002)
Background
- Successful innovation requires a combination of physical (i.e. materials) and social (i.e. know-how) inputs (Nelson and Sampat, 2001)
- Technology development tends to happen unevenly across domains of use i.e. Medicine, Education (Nelson, 1977); Aeronautics (Vincenti, 1990)
- Hard to strike a balance between the application of (codified) technical blueprints and (tacit) know-how that emerges (or not) in particular contexts of use (Rosenberg, 1974; David, 1974)
Background
Green technology (GT) is one component among others of "green innovation"
- GT are not immune to these “imperfections”...
- cover a wide range of domains of use (i.e. waste, water, air, etc)
- more complex, radical, pervasive and impactful than non-GT (Barbieri et al. 2018) and therefore require a wide range of competences that (sometimes) are far from established know-how (Marchi, 2012; Ghisetti et al. 2015)
- GT development is space-bound: geographical areas exhibit differential
- Exposure to climate events
- Ability to adapt to climate events
Motivation
Piecemeal evidence but no comprehensive study of GT over time, across countries, and domains of specialisation
Gaps in the litterature:
- Where and when do GT emerge?
- What is the relative state of development across domains of GT
- Which factors spur or impede GT development?
Goals
| Gaps | Contributions |
|---|---|
| Where and when do GT emerge? | A complete, scalable dataset on GT over time, geographical space and domain of specialisation |
| What is the relative state of development across domains of GT? | Heuristic identification of technology life cycle to articulate the physical and social components of GT |
| Which factors spur or impede GT development? | Focus on specific location characteristics |
Dataset creation
- Development of GT observed using patent data.
- Unit of analysis : INPADOC patent families from PATSTAT 2016a
- Identified with ENV-TECH classification (OECD, 2016)
patent family with at least 1 green IPC/CPC code → green patent family -
Geolocalisation of inventors' address at city level
→ geo-scalable (countries, states, commuting zones...)
ENV-TECH classification
Identify in the patent classification systems (IPC, CPC) codes associated with GT.
OECD Env-tech identifies 4+1 groups:
- Environmental management
- Water-related adaptation technologies
- Capture, Storage and sequestration of greenhouse gases (GHG)
- Climate Change Mitigation Technologies (CCMT), divided in 5 sectors
- ... and Biodiversity protection and ecosystem health (listed but no content yet)
Evolution of Green Patent Families [1970 - 2010]
Distribution of Green Patent Families [1970 - 2010]
Number of patent families (x 1000)
Evolution of patent families per country [1970 - 2010]
Fitness ranking of a selection of countries
Adapted from figure 3.2 (chapter 3)
Technology Life Cycle
An extension of Abernathy & Utterback (1978) and Vona & Consoli (2015) to articulate technology life cycle stages through the lenses of the knowledge base
Policy & Methodologies
-
Importance of TLC for policies
→ stages should match territory's capacities
→ policies to promote the use of GT should take into account technology maturity -
No existing methodologies to assess TLC in a large set of technologies
(Haupt et al., 2007; Gao et al., 2013; Chang and fan, 2016)
Theoretical framework
An articulation of 2 concepts:
- the physical evolution of technologies (Abernathy & Utterback, 1978)
-
the social evolution (Vona & Consoli, 2015):
technology advances when suitable skills are needed and, in turn, change future demand for skills
Life cycle stages
| Stage | Description |
|---|---|
| Emergence | exploration, experimentation, competition between different designs, highly localised |
| Development | standardisation of design, shake out of inferior variants, some diffusion |
| Diffusion | stabilisation of design, wider geographical diffusion |
| Maturity | dominant design, high standardisation, widest geographical diffusion |
Proposed approach
Identification of TLC stages
Identification of TLC stages - example
Renewable energy generation
An example of a technology in diffusion stage
Enabling technologies in transport
An example of a technology in emergence stage
Maturity ranking per group
| Group | Technological field | Ranking |
|---|---|---|
| 5 | Capture, storage, sequestration or disposal of GHG | 1 (Less mature) |
| 6 | CCMT Transportation | 2 |
| 4 | CCMT Energy generation, transmission or distribution | 3 |
| 9 | CCMT Production or processing of goods | 4 |
| 2 | CCMT Water-related adaptation technologies | 5 |
| 8 | CCMT Wastewater treatment or waste management | 6 |
| 7 | CCMT Buildings | 7 |
| 1 | Environmental management | 8 (More Mature) |
Coherent with grey literature, i.e. OECD 2011, Towards Green Growth
GT in space and time
Which factors are associated with the production of green patents?
3 perspectives:
- countries development [chapter 5]
- knowledge base [chapter 6]
- policy [chapter 7]
Background
Agglomeration economies
- Marshall (1920): repeated interaction and proximity of goals and of competences
- Jacobs (1969): diversity of competences
Evolutionary economic geography
- Jacobs externalities do not merely lead to a more efficient division of labour within regions (Gleaser et al., 1992) → a diversified environment increases the opportunities for innovation
- Diversification per se does not fully capture the mechanism that brings about regional economic growth (Frenken et al., 2007)
- Knowledge flows within regions require a balance of cognitive distance to avoid lock-ins and of cognitive proximity to enable effective learning (Nooteboom, 2002; Iammarino and Boschma, 2009)
⇒ Related industries share some cognitive structures that enhance learning opportunities and knowledge spillovers
Countries development
countries development & capacities
technology complexity
↔
prob. of diversification / specialisation in a technology
Research design
based on Petralia et al. (2017), we estimate 2 models :
⇢ probability of diversification in a technology $_j$ = to have an RTA without one before
$$ S_{cjt} = \beta_{1} Density_{cjt-1} + \beta_{2} ITC_{jt} + \beta_{3} Maturity_{jt} + \beta_{4} GDP_{ct} + Controls $$
⇢ probability of specialization in a technology $_j$ = to have an RTA
similar equation but interacting $Size_{jt}$, $HI_{jt}$, $ITC_{jt}$ and $Maturity_{jt}$ with $GDP_{ct}$
with:
- $ S_{cjt} = 1$ if $RTA_{cjt}$ > 1
- $ITC_{jt}$: complexity of a technology (Hidalgo, 2007)
- $Density_{jt}$: how close is a technology to country's portfolio
- $Maturity_{jt}$: life cycle stage of a technology, from 1 to 4
- $Controls$: $HI_{jt}$(Herfindhal Index), $Size_{jt}$(size of a technology)
Regression results overview
| Diversification (1) | Diversification (2) | Specialisation | |
|---|---|---|---|
| Density of neighbouring technologies | 0.02746*** | 0.11723*** | 0.13872*** |
| (0.01) | (0.01) | (0.01) | |
| Tech-Level Variables | |||
| Technology Complexity (ITC) | 0.00169*** | 0.00389*** | 0.00464*** |
| (0.00) | (0.00) | (0.00) | |
| Maturity | 0.02568*** | 0.04474*** | 0.04034*** |
| (0.00) | (0.00) | (0.00) | |
| GDP | 0.00012 | 0.00016 | 0.00523*** |
| (0.00) | (0.00) | (0.00) | |
| GDP x ITC | -0.00008 | ||
| (0.00) | |||
| GDP x Maturity | 0.00037*** | ||
| (0.00) | |||
| Controls | Yes | Yes | Yes |
| Tech, Country & Time Fixed Effects | Yes | Yes | Yes |
| Obs. | 51149 | 70547 | 77065 |
(1) RTA < 0.1 in the previous year, (2) RTA < at the beginning of the sample
Which type of local knowledge correlates with GP production along the life cycle
territories capacities to recombine knowledge
↔
green/non-green patent families production
Variety of knowledge recombination of the territory
We measure variety within the territory knowledge base using patent families and IPC codes (Castaldi, 2015).
3 levels of variety:
- UV → recombination of unrelated knowledge
(e.g. Chemistry ↔ Mechanical Engineering) - SRV → recombination of semi-related knowledge
(e.g. Combustion Engines ↔ Gas-Turbine Plants) - RV → recombination of related knowledge
(e.g. Engines characterised by fuel-air mixture compression ↔ Engines characterised by separate admission)
Evidences from US federal states [1980-2010]
Empirical model:
$$ GP^{L}_{jt} =\beta_{1} UV_{jt} +\beta_{2} SRV_{jt} +\beta_{3} RV_{jt} + Controls_{jt}$$
Where:
- $GP^{L}_{jt}$: N. of green patent families per millions inhabitants
- $UV_{jt}$: Unrelated Variety
- $SRV_{jt}$: Semi-Related Variety
- $RV_{jt}$: Related Variety
- $Controls_{jt}$: R&D expenditure, % of people with bachelor degree or higher, HC and R&D in neighbour states, Pop. density
Regression Results
| Green Pat | All Pat | |
|---|---|---|
| UV (log) | 1.386*** | -0.931* |
| (0.421) | (0.507) | |
| SRV (log) | 0.286 | 0.193*** |
| (0.184) | (0.0668) | |
| RV (log) | 0.392** | 0.515*** |
| (0.154) | (0.0987) | |
| Controls | YES | YES |
| State FE | YES | YES |
| Time Dummies | YES | YES |
| Random growth | YES | YES |
| Obs. | 1466 | 1470 |
Regression results including life cycle
| Green Pat | Emergence | Development | Diffusion | Maturity | |
|---|---|---|---|---|---|
| UV (log) | 1.386*** | 0.958* | 1.214** | 0.597 | 0.716 |
| (0.421) | (0.523) | (0.590) | (0.786) | (0.473) | |
| SRV (log) | 0.286 | -0.356 | 0.783*** | 0.166 | -0.205 |
| (0.184) | (0.338) | (0.201) | (0.249) | (0.147) | |
| RV (log) | 0.392** | 0.421 | 0.516*** | 0.434* | 0.554*** |
| (0.154) | (0.313) | (0.157) | (0.247) | (0.0848) | |
| Controls | YES | YES | YES | YES | YES |
| State FE | YES | YES | YES | YES | YES |
| Time Dummies | YES | YES | YES | YES | YES |
| Random growth | YES | YES | YES | YES | YES |
| Obs. | 1466 | 1392 | 1371 | 1424 | 1452 |
Role of Policy: Public procurement and local labour markets
Characterising Territory's Occupations
Task-based framework (Autor, Levy, Murnane, 2003; Autor and Dorn, 2013)
→ focus on relative importance of occupations.
match: main work tasks ↔ skills needed.
- abstract - require creativity, problem-solving, intuition...
- routine - execution of codified instructions with minimal discretion.
- manual - demand visual and language recognition, personal interaction and physical dexterity.
Evidences from US Commuting Zones [2005-2011]
Empirical models:
$Y_{j,t}=\beta_{1}GPP_{j,t-1}+\beta_{2}RI_{j,t-1}+\beta_{3}GPP_{j,t-1}\times RI_{j,t-1} + Controls_{j,t}$
$Y_{j,t}=\beta_{1}GPP_{j,t-1}+\beta_{2}AI_{j,t-1}+\beta_{3}GPP_{j,t-1}\times AI_{j,t-1} + Controls_{j,t}$
$Y_{j,t}=\beta_{1}GPP_{j,t-1}+\beta_{2}MI_{j,t-1}+\beta_{3}GPP_{j,t-1}\times MI_{j,t-1} + Controls_{j,t}$
where
- $Y_{j,t}$: Green Patent stock
- $GPP_{j,t-1}$: Green Public Procurement
- $RI_{j,t-1} = 1$ if CZ is routine intensive
- $AI_{j,t-1} = 1$ if CZ is abstract intensive
- $MI_{j,t-1} = 1$ if CZ is manual intensive
Regression Results
| (I) | (II) | (III) | |
|---|---|---|---|
| Total GPP | 0.039*** | 0.021** | 0.048*** |
| (0.009) | (0.008) | (0.009) | |
| Routine Ind. | 0.004 | ||
| (0.012) | |||
| GPP x RI | -0.000 | ||
| (0.011) | |||
| Abstract Ind. | 0.017 | ||
| (0.015) | |||
| GPP x AI | 0.042*** | ||
| (0.010) | |||
| Manual Ind. | 0.001 | ||
| (0.010) | |||
| GPP x MI | -0.037*** | ||
| (0.012) | |||
| controls | YES | YES | YES |
| R2 | 0.458 | 0.469 | 0.464 |
| N | 3851 | 3851 | 3851 |
Empirical Analysis - Findings
- Capacities in neighbouring technologies are as important as GDP to diversify/specialize in GT
- Capacities to recombine unrelated knowledge are more important in GT, especially in the emergence stages
- Green Public Procurement is positively associated with the development of GT, in particular in abstract skills intensive territories
Conclusions
Main contributions
| Gaps | Contributions | |
|---|---|---|
| no transversal studies across time, countries, technologies | ✔ | A dataset about green patent families, scalable in space and technologies |
| lack of research on dynamics of GT | ✔ | An heuristic articulation of technology life cycle |
| lack of research on territory's characteristics correlated with GT development | ✔ | Empirical analysis of territory characteristics that foster green patents production |
Contribution: Dataset about green patent families
- From 1855 to 2010
- Located in 141 countries - scalable to any geographic level
- 36 green technologies - scalable to finer-grained technologies
- A website to visualize it: www.greentechdatabase.com
Contribution: Heuristic articulation of technology life cycle
- Observe empirically and systematically unevenness across and within domains
- Establish correspondence between technical “exploring” (volume of patenting) and “broadening” (geographical spread)
- Provide a detailed policy tool to know the relative state of advance of technology-region pairs, information available on www.greentechdatabase.com
Contribution: Empirical analysis of territory characteristics that foster green patents production
- Establish correspondence between country capabilities and specialisation / diversification
- Account for the connection between knowledge base and TLC
- Assess the role of policy mediated by human capital
Policy implications
- Betting on GT to mitigate climate change should account for TLC stage:
- they may be not yet mature when needed (e.g. GHG sequestration)
- one component of the "technological ecosystem" may be not mature (e.g. electric vehicles / enabling tech. in transport and in energy)
- Policy makers should also account for territory's characteristics when promoting the development of GT, besides their life cycle stage.
Limitations
- Patents do not capture all technological innovations, with differences across technologies
- Intrinsec value of green patents is not considered
- Life Cycle methodology identifies only relative development stages
- Green innovation is not only GT
Future avenues and developments
- Explore new territories, like the European Union and China.
- Investigate if the use of GT really leads to sustainable societies.
- Compare life cycle of GT with that of other technologies
- Analyse other aspects of green innovation, and Delve into the interactions between GT, inequality, extreme climate events and societal resilience.
Thesis output
- 2018: Green Technologies Fitness, Angelica Sbardella, François Perruchas, Lorenzo Napolitano, Nicolò Barbieri and Davide Consoli, Entropy 2018, 20(10), 776
- Specialisation, Diversification and the Ladder of Green Technology Development, François Perruchas, Davide Consoli and Nicoló Barbieri, R&R in Research Policy
- Specialization, diversification and environmental technology life-cycle, Nicoló Barbieri, François Perruchas and Davide Consoli, R&R in Economic Geography
- Public procurement, local labour markets and green technological change: Evidence from US Commuting Zones, Gianluca Orsatti, François Perruchas, Davide Consoli and Francesco Quatraro, R&R in Environmental and Resource Economics
- Green innovation and income inequality: a complex system analysis, Lorenzo Napolitano, Angelica Sbardella, Davide Consoli, Nicolò Barbieri and François Perruchas, submitted to Research Policy (special issue on Economic Complexity)
Thank you
Further material
Dataset
Evolution of patent families per country [1970 - 2010]
1981 = base 100
Mapping of Green Technologies Evolution
- Fractional count of patent families per country / technology
- Creation of weighted matrix $W_{c,t}(y)$
- binary matrix to reflect $RCA_{c,t}(y)$
- EFC algorithm takes binary matrix as input
Country fitness: average complexity of its technologies
Technology complexity: lower when patented by low-fitness countries
green fitness ranking of 36 technologies
green fitness ranking of 61 active countries
green fitness ↔ GDP ↔ export fitness
Technology Life Cycle
Geographical Ubiquity
The indicator for each GT is given by the number of countries that exhibit RTA in a particular technology (Petralia et al., 2017; Balland & Rigby, 2017):
\[UBIQUITY_{jt} = \sum_{c}M_{cj}\]
where \(M_{cj} = 1\) if \(RTA > 1\)
Green diversification and specialization across countries
Research design
based on Petralia (2017), we estimate the probability of
⇢ diversification in a technology $_j$
$$ S_{cjt} = \Theta_{1} Density_{cjt-1} + \Theta_{2} Density_{cjt-1} \times GDP_{ct} + \beta_1 \log Size_{jt} + \beta_2 HI_{jt} \\ + \beta_3 ITC_{jt} + \delta_c D_c + \delta_j D_j + \delta_t D_t + \varepsilon_{cjt} $$
⇢ specialization in a technology $_j$
$$ S_{cjt} = \Theta_{1} Density_{cjt-1} + \beta_1 \log Size_{jt} \times GDP_{ct} + \beta_2 HI_{jt} \times GDP_{ct} \\ + \beta_3 ITC_{jt} \times GDP_{ct} + \delta_c D_c + \delta_j D_j + \delta_t D_t + \varepsilon_{cjt} $$
with:
- $ S_{cjt} = 1$ if $RTA_{cjt}$ > 1
- $ITC_{jt}$: complexity of a technology (Hidalgo, 2007)
- $Density_{jt}$: how close is a technology to country's portfolio
- $HI_{jt}$: Herfindhal Index: indicate how concentrated is tech. production
- $Size_{jt}$: size of the technology
Regression results
| Div. Eq. (1) | Div. Eq. (2) | Spec. Eq. | |
|---|---|---|---|
| Density | 0.02746*** | 0.11723*** | 0.13949*** |
| (0.01) | (0.01) | (0.01) | |
| Density x GDP | 0.00049 | 0.00030 | |
| (0.00) | (0.00) | ||
| Tech-Level Variables | |||
| Log Size | 0.00497** | 0.00658*** | 0.01495*** |
| (0.00) | (0.00) | (0.00) | |
| Herfindahl Index | -0.01898*** | -0.04997*** | 0.01416* |
| (0.01) | (0.01) | (0.01) | |
| ITC | 0.00169*** | 0.00389*** | 0.00484*** |
| (0.00) | (0.00) | (0.00) | |
| GDP | 0.00012 | 0.00016 | 0.00736*** |
| (0.00) | (0.00) | (0.00) | |
| GDP x Log Size | -0.00049*** | ||
| (0.00) | |||
| GDP x Herfindahl Index | -0.01453*** | ||
| (0.00) | |||
| GDP x ITC | -0.00012** | ||
| (0.00) | |||
| Tech Fixed Effects | Yes | Yes | Yes |
| Time Fixed Effects | Yes | Yes | Yes |
| Country Fixed Effects | Yes | Yes | Yes |
| Obs. | 51149 | 70547 | 77065 |
| (1) RTA < 0.1 in the prev. period. (2) RTA < at the beg. of the sample. | |||
Regression results including tech. maturity
| Div. Eq. (1) | Div. Eq. (2) | Spec. Eq. | |
|---|---|---|---|
| Density | 0.02746*** | 0.11723*** | 0.13872*** |
| (0.01) | (0.01) | (0.01) | |
| Density x GDP | 0.00049 | 0.00030 | |
| (0.00) | (0.00) | ||
| Tech-Level Variables | |||
| Log Size | 0.00497** | 0.00658*** | 0.01555*** |
| (0.00) | (0.00) | (0.00) | |
| Herfindahl Index | -0.01898*** | -0.04997*** | 0.00558 |
| (0.01) | (0.01) | (0.01) | |
| ITC | 0.00169*** | 0.00389*** | 0.00464*** |
| (0.00) | (0.00) | (0.00) | |
| Maturity | 0.02568*** | 0.04474*** | 0.04034*** |
| (0.00) | (0.00) | (0.00) | |
| GDP | 0.00012 | 0.00016 | 0.00523*** |
| (0.00) | (0.00) | (0.00) | |
| GDP x Log Size | -0.00053*** | ||
| (0.00) | |||
| GDP x Herfindahl Index | -0.01311*** | ||
| (0.00) | |||
| GDP x ITC | -0.00008 | ||
| (0.00) | |||
| GDP x Maturity | 0.00037*** | ||
| (0.00) | |||
| Tech Fixed Effects | Yes | Yes | Yes |
| Time Fixed Effects | Yes | Yes | Yes |
| Country Fixed Effects | Yes | Yes | Yes |
| Obs. | 51149 | 70547 | 77065 |
Countries characteristics
Diversification probabilities
Countries characteristics
Specialization probabilities
Co-evolution of GT and the knowledge base
Previous litterature
- GT require different competences that are far form traditional knowledge bases of industries (De Marchi, 2012; Ghisetti et al., 2015)
- GT are more complex, radical, pervasive and impactful than non-GT (Barbieri et al., 2018)
Knowledge structures underlying innovation
- Innovation: knowledge accumulation and recombination
(Aghion & Howitt,1992; Weitzman, 1998; Basalla, 1988; Arthur, 2007 - Location matters:
- Marshall (1920): innovation needs interaction, proximity of goals and of competences
- Jacobs (1969): diversity of competences
Related - Unrelated variety
- two kinds of diversity:
- within-industry: Related Variety (RV)
- between-industry: Unrelated Variety (UV)
- UV & RV are complementary (Castaldi, 2015):
- UV more associated with Radical Innovation
- RV more assoc. with Incremental Innovation
Measuring variety with Entropy
We measure variety with the territory knowledge base using patent families and IPC codes (Castaldi, 2015).
$$UV_{it} = \sum_{k} s_{k,it} ln \left ( \frac{1}{s_{k,it}} \right )$$
$$SRV_{it} = \sum_{l} s_{l,it} ln \left ( \frac{1}{s_{l,it}} \right ) - \sum_{k} s_{k,it} ln \left ( \frac{1}{s_{k,it}} \right )$$
$$RV_{it} = \sum_{m} s_{m,it} ln \left ( \frac{1}{s_{m,it}} \right ) - \sum_{l} s_{l,it} ln \left ( \frac{1}{s_{l,it}} \right )$$
where
- $i$: Federal State
- $k, l, m$: Technological categories at IPC 1-digit, 4-digit, 8-digit
- $s_{k,it}$: Share of patents in category $k (l, m)$
Measuring variety with Entropy - Example
| Level I (k) - UV | Level II (l) - SRV | Level III (m) - RV |
|---|---|---|
| C Chemistry | C01B Non-metallic elements | C01B 6/10 Monoborane; Diborane; Addition complexes thereof |
| .. | .. | C01B 6/11 Preparation from boron or inorganic compounds containing boron and oxygen |
| .. | .. | C01B 6/13 Addition complexes of monoborane or diborane, e.g. with phosphine, arsine or hydrazine |
| .. | .. | C01B 6/15 Metal borohydrides; Addition complexes thereof |
| .. | C01C Ammonia | . |
| .. | C01D Lithium, sodium, potassium, etc | . |
| .. | C01F Compounds of the metals beryllium, magnesium | . |
| D Textiles | .. | .. |
| F Mechanical Engineering | .. | .. |
Evidences from US federal states
Empirical model:
$$ GP^{L}_{jt} =\beta_{1} Variety_{jt} +\beta_{2} R\&D_{jt} +\beta_{3} HC_{jt} + Controls_{jt} + \tau_{j} + \gamma_{t} + \delta_{jt} + e_{jt}$$
Where:
- $GP^{L}_{jt}$: N. of green patent families per millions inhabitants in US states over year
- $Variety_{jt}$: UV, SRV and RV
- $R\&D_{jt}$: R&D expenditure
- $HC_{jt}$: % of people with bachelor degree or higher
- $Controls_{jt}$: HC and R&D in neighbour states, Pop. density
- $\gamma_{t}$: state fixed effects
- $\delta_{jt}$: time fixed effects
Public procurement and local labour markets
Public procurement and local labour markets
Many works about the impact of public procurement on innovation (Nelson, 1982; Geroski, 1990; Ruttan, 2006)
↳ Few research on green innovation in particular (Ghisetti, 2017).
GT: double-externality issue (Jaffe, Newell, and Stavins, 2005)
- knowledge externalities
- environmental externalities
→ Public policy required
- demand-oriented tools: green public procurement
- supply-oriented tools: skills composition
Public procurement data
- Public information provided by USAspending.gov (data cover the period from 2000 to date)
- Location (5-digit zipcode) where the contract is executed
- Amount of resources dedicated (in 2010 USD)
- Whether the contract relates to product or service
- Green content of the contract, following Product and Service Codes Manual
Occupational data
Task-based framework (Autor, Levy, Murnane, 2003; Autor and Dorn, 2013)
→ focus on relative importance of occupations.
match: main work tasks ↔ skills needed.
- abstract-intensive - require creativity, problem-solving, intuition...
- routine-intensive - execution of codified instructions with minimal discretion.
- manual-intensive - demand visual and language recognition, personal interaction and physical dexterity.
Occupational data
For each type of task (A,R,M)
For each occupation: $ATI_{k}=ln(T_{k,1980}^{A})-ln(T_{k,1980}^{R})-ln(T_{k,1980}^{M})$
Then, for each CZ $j$: $ASH_{jt}=({\sum_{k=1}^{K}}L_{jkt}\cdot{{1}}[ATI_{k}>ATI^{P66}])({\sum_{k=1}^{K}}L_{jkt})^{-1}$
Finally, if the CZ $j$ is in the top third of national task share at time $t$: $AI_{jt}={{1}}\left[ASH_{jt}>ASH_{t}^{P66}\right]$
Empirical strategy
Using full sample of 722 CZs observed from 2000 to 2011. We investigate the relationship between GPP and local green technological activity:
$Y_{j,t}=\beta_{0}+\beta_{1}GPP{}_{j,t-1}+\boldsymbol{X_{j,t}^{\prime}}\beta_{2}+\epsilon_{j,t}$
where:
- $Y_{j,t}$: fractionalized stock of green patent families (weighted by forward citations) at time $_{t}$ filed by inventors resident in CZ $_{j}$;
- $GPP{}_{j,t-1}$: level of expenditures for GPP performed in CZ $_{j}$ at time $_{t−1}$ (2010 USD);
- $X_{j,t}^{\prime}$: controls for CZs' labor force and demographic composition
Regression results
| (I) | (II) | |
|---|---|---|
| product GPP | 0.053*** | |
| (0.013) | ||
| service GPP | 0.087*** | |
| (0.011) | ||
| pop density | -0.000 | -0.000 |
| (0.000) | (0.000) | |
| empl share | -0.000** | -0.000** |
| (0.000) | (0.000) | |
| N. of Firms | 0.000*** | 0.000*** |
| (0.000) | (0.000) | |
| R&D empl | 8.609** | 8.510** |
| (4.378) | (4.168) | |
| R2 | 0.472 | 0.498 |
| N | 7937 | 7937 |
Standard errors clustered at the level of State. Models include a constant, year dummies and geographic dummies (9 Census divisions). * p < . 1, ** p < . 05, *** p < . 01
Empirical strategy
To test for moderating effects of CZ occupational task compositions, we estimate 3 models:
$Y_{j,t}=\beta_{0}+\beta_{1}GPP_{j,t-1}+\beta_{2}RI_{j,t-1}+\beta_{3}GPP_{j,t-1}\times RI_{j,t-1}+\boldsymbol{X_{j,t}^{\prime}}\beta_{4}+\epsilon_{j,t}$
$Y_{j,t}=\beta_{0}+\beta_{1}GPP_{j,t-1}+\beta_{2}AI_{j,t-1}+\beta_{3}GPP_{j,t-1}\times AI_{j,t-1}+\boldsymbol{X_{j,t}^{\prime}}\beta_{4}+\epsilon_{j,t}$
$Y_{j,t}=\beta_{0}+\beta_{1}GPP_{j,t-1}+\beta_{2}MI_{j,t-1}+\beta_{3}GPP_{j,t-1}\times MI_{j,t-1}+\boldsymbol{X_{j,t}^{\prime}}\beta_{4}+\epsilon_{j,t}$
where $RI_{j,t-1}$, $AI_{j,t-1}$ and $MI_{j,t-1}$ are dummy variables indicating if CZ is routine, abstract or manual intensive tasks respectively.
Due to occupational data availability, we consider here 2005-2011.
REGRESSION RESULTS INCLUDING TASK COMPOSITION
| (I) | (II) | (III) | (IV) | (V) | (VI) | |
|---|---|---|---|---|---|---|
| tot GPP | 0.039*** | 0.039*** | 0.039*** | 0.021** | 0.040*** | 0.048*** |
| (0.008) | (0.009) | (0.008) | (0.008) | (0.008) | (0.009) | |
| RI | 0.003 | 0.004 | ||||
| (0.013) | (0.012) | |||||
| GPP x RI | -0.000 | |||||
| (0.011) | ||||||
| AI | 0.041*** | 0.017 | ||||
| (0.014) | (0.015) | |||||
| GPP x AI | 0.042*** | |||||
| (0.010) | ||||||
| MI | -0.013 | 0.001 | ||||
| (0.010) | (0.010) | |||||
| GPP x MI | -0.037*** | |||||
| (0.012) | ||||||
| controls | YES | YES | YES | YES | YES | YES |
| R2 | 0.458 | 0.458 | 0.464 | 0.469 | 0.461 | 0.464 |
| N | 3851 | 3851 | 3851 | 3851 | 3851 | 3851 |
GPP, RSH, ASH and MSH lagged 1-year. Standard errors clustered at the level of State. Models include a constant, year dummies and geographic dummies (9 Census divisions). * p < . 1, ** p < . 05, *** p < . 01