## The Impact of Environmental Policy on Innovation in Clean Technologies — Section 1–5 (wpiea2021213-print-pdf)

## Source details

**Canonical URL:** [The Impact of Environmental Policy on Innovation in Clean Technologies — Section 1–5 (wpiea2021213-print-pdf)](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021213-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021213-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021213-print-pdf.pdf.json)

---

### Introduction: motivation and scope
- Climate change is slow-moving; actions and consequences can be several decades apart, making technological progress unusually important.
- Social planner’s objectives: (i) avoid future climate-change related damages by reducing net carbon emissions, while (ii) minimizing transitional costs, including by stimulating technological progress among clean technologies.
- Research focus:
  - Whether and to what extent the overall tightening of policies has contributed to clean energy-related innovation.
  - How more detailed policies vary in terms of innovation performance.
  - Whether potential gains in clean innovation are offset by reductions in innovation in other (dirty) technologies.
- Empirical sample: 33 OECD countries from 1990 to 2016.
- Innovation proxy: climate-change mitigating patent families from PATSTAT.

### Conceptual and theoretical context
- Earlier integrated assessment models treated technology as exogenous; later work endogenized technology and showed climate policies can induce innovation.
- Key theoretical points:
  - In some models the optimal carbon tax can fully offset the pollution externality even with endogenous technology (e.g., Golosov et al. (2014)), reducing the first-best case for large R&D subsidies.
  - Path dependencies and market-size effects (Acemoglu et al. (2012, 2016); Fried (2018); Barrett (2021)) can create positive returns to scale and strengthen the case for R&D subsidies as a complement to carbon pricing.
  - Elasticity of substitution across energy sources and costs/payoffs of knowledge creation materially affect the role of innovation for emissions and welfare.

### Data, measurement, and empirical framework
- Sample coverage and measurement:
  - 33 countries, 1990–2016. Countries listed in source.
  - Innovation measured at the patent-family level (EPO DOCDB simple family), associated with priority date and inventor country.
  - Clean technology classification: Haščič and Migotto (2015) for climate-change mitigation technologies related to energy; Dechezleprêtre et al. (2017) for clean/grey/dirty electricity technologies.
- Policy measure:
  - OECD Environmental Policy Stringency (EPS) index: hierarchical, standardized, decomposable into market-based and non-market policies; values 0–6 at the most disaggregated level relative to sample distribution; aggregations are simple averages.
  - EPS data available up to 2014 (regressions use a one-period lag, limiting estimation to 2015).
- Estimation framework:
  - Innovation production: X_{i,t} = exp(α_i + δ_t + β ln(H_{i,t}) + γ ln(u_{i,t}) + ε_{i,t}).
  - Conditional fixed effects Poisson model with robust standard errors.
  - Technology-specific knowledge stock constructed via perpetual inventory with discount rate 10%.
  - Controls include overall innovation, oil and gas reserves (gas converted using 1 barrel of oil = 6,000 cubic feet of natural gas), household electricity prices, product and labor market regulation indicators, crude oil prices.
  - Country- and year-fixed effects absorb time-invariant country traits and common global dynamics; controls are lagged by one year.

### Key empirical findings — aggregate effects
- Aggregate EPS effect on clean energy innovation:
  - Estimated range: 0.15 to 0.25.
  - Baseline specification (Table 1, column 1): L.EPS = 0.184*** (4.46).
  - Other L.EPS coefficients and t-statistics:
    - Column (2): 0.249*** (6.25)
    - Column (3): 0.188*** (6.77)
    - Column (4): 0.232*** (6.74)
    - Column (5): 0.166*** (5.18)
    - Column (6): 0.203*** (5.48)
- Key covariates (selected, Table 1):
  - L.ln(tech stock): coefficients range from 0.479 to 0.744 (all ***) with t-statistics including (9.73), (10.30), (11.47), (6.31), (21.77), (4.93).
  - L.ln(all tech): coefficients include 0.415*** (6.70), 0.320*** (4.04), 0.573*** (5.98), 0.574*** (4.39), 0.165* (1.70), 0.452*** (2.66).
  - L.ln(electr. price HH, USD): notable coefficients 0.295*** (3.24) in column (3), 0.180* (1.84) in column (5), 0.357*** (3.52) in column (6).
  - L.ETCR electricity: -0.105** (-2.56) in column (4).
  - L.ln(oil & gas reserves, bb): e.g., Column (2): -0.136* (-1.66).
- Economic magnitude:
  - Counterfactual: change in average EPS between 1990 and 2010 contributed roughly 30 percent of the increase in clean energy innovation between 1990 and 2010.
  - Illustrative equivalence (specification including oil prices but no year-FE): tightening EPS ≈ USD 70 increase in crude oil price per barrel.
- Interpretation: effects materialize quickly (statistically significant almost immediately) and strengthen over the first 2 to 3 years.

### Dynamics: speed and persistence of effects
- Local projection results (Table 2, ∆EPS effects on log-difference in CCM Energy patent families):
  - t to t+1: ∆EPS = 0.106*** (2.78)
  - t to t+2: ∆EPS = 0.126** (2.05)
  - t to t+3: ∆EPS = 0.140* (1.80)
  - t to t+4: ∆EPS = 0.117 (1.06)
  - t to t+5: ∆EPS = 0.114 (1.04)
- Dynamics including oil & gas reserves (Section 5 dynamics):
  - EPS coefficients (t+1 to t+5): 0.101** (2.58), 0.115* (1.92), 0.135 (1.63), 0.110 (0.93), 0.113 (0.97).
  - ln(oil price) coefficients for t+1 to t+5: 0.217*** (3.73), 0.365*** (4.05), 0.331*** (3.12), 0.292*** (2.83), 0.147 (1.24).
- Dynamics including electricity prices (Section 5 dynamics):
  - EPS coefficients (t+1 to t+5): 0.117** (2.74), 0.200*** (3.55), 0.237*** (3.44), 0.266** (2.69), 0.232** (2.39).
  - ∆ln(all tech) coefficients generally larger in these specifications (e.g., t+1: 0.624*** (8.88)).
- Interpretation:
  - Innovation response increases for another two years to about ¾ of baseline Table 1 coefficient before becoming imprecise.
  - Negative coefficients on knowledge stock reflect maturation effects: higher innovation growth when technologies are new and slowing as they mature.
  - Patent timing (priority date) implies measured innovation may appear quickly; commercial availability lags.

### Effects by policy instrument (aggregate disaggregation)
- Main finding (Table 3, Table 5 series): both market-based and non-market policy tools induced clean innovation; carbon taxes are an exception.
- Selected coefficients (Table 3, column 6 examples):
  - Trading schemes: 0.0367*** (3.37)
  - Feed-in tariffs: 0.0283*** (2.71)
  - Emission limits: 0.0385* (1.66)
  - R&D subsidies: 0.0486** (2.39)
  - Carbon taxes: -0.0255 (-0.88) — statistically insignificant
- Additional specifications (Section 5 individual-policy tables):
  - Trading schemes: e.g., 0.0548*** (4.11) and 0.0337*** (3.69) in different columns.
  - Feed-in tariffs: e.g., 0.0383*** (3.43); 0.0409*** (4.10) where household electricity prices included.
  - Emission limits: 0.0564*** (3.03); 0.0632** (2.24) in other specifications.
  - R&D subsidies: 0.0621*** (6.06); 0.0478*** (2.91) in electricity-price specifications.
- Interpretation and caveats:
  - Carbon taxes show no significant effect largely because only slightly more than 10 percent of countries used carbon taxes in 2015; limited cross-country and time variation reduces statistical power.
  - Positive effects of trading schemes and electricity prices align with literature on relative price changes in favor of cleaner technologies driving innovation.
  - Results suggest substitutability among policy tools for stimulating innovation; multiple tools or combinations can promote clean innovation comparably.

### Composition versus addition of electricity innovation
- Dechezleprêtre et al. (2017) classification used to split electricity innovation into clean, grey, and dirty.
- Key coefficients (Table 4 / Section 4; fixed effects Poisson; N = 743):
  - Clean (column 1): L.EPS 0.0625*** (2.64)
  - Grey (column 2): L.EPS 0.150*** (3.39)
  - Dirty (column 3): L.EPS -0.0402** (-2.14)
  - Total (column 4): L.EPS 0.201*** (3.49) with ln(all electr) coefficient constrained at 1.00
- Interpretation:
  - Environmental policies shifted electricity-related innovation away from dirty toward clean and especially grey technologies.
  - Effect on total electricity innovation is net positive — increased clean and grey innovation was not offset by decreases in dirty innovation.
  - Magnitude of grey innovation response is large but requires cautious interpretation due to lower initial levels and sensitivity in some specifications.
  - Continued innovation in dirty technologies, even if slower, can reduce the cost competitiveness of clean technologies and slow the energy transition; nonetheless induced innovation reduces macroeconomic transition costs if it accelerates technological progress in clean options.

### Robustness and sensitivity analyses
- Innovation measure robustness:
  - Alternatives: patent families, all applications, international patent families.
  - Triadic patent families produce 44% zeroes versus 10% zeroes with “international” patent family.
  - Results qualitatively robust across measures; coefficients decline somewhat with international patent families.
- Dropping key innovator countries (Section 5):
  - L.EPS remains significant when dropping CHN, DEU, JPN, KOR, USA with coefficients: CHN dropped: 0.191*** (6.09); DEU dropped: 0.190*** (4.68); JPN dropped: 0.129*** (3.95); KOR dropped: 0.227*** (5.60); USA dropped: 0.168*** (4.16).
- Controls and sample sensitivity:
  - Including oil and gas reserves weakens results slightly.
  - Including electricity prices tends to reinforce results, especially 2–4 years after policy changes, though this partly reflects sample changes.
  - Controlling for oil and gas reserves and electricity prices does not materially change policy-variable coefficients except for carbon prices.
  - Removing individual large innovators does not materially change main conclusions.

### Policy implications and research agenda
- Policy implications:
  - Environmental policy tightening has made a statistically and economically significant contribution to increased clean technology innovation.
  - Both market-based and non-market policy tools can induce innovation and can be used or combined guided by political feasibility and expected emissions effects.
  - Because induced innovation largely complements rather than fully replaces innovation in traditional technologies, policymakers should account for continued dirty-technology innovation when designing transition strategies.
- Recommendations for further research:
  - Compare effectiveness and costs of different policy tools more directly.
  - Assess greenhouse gas emission implications of induced innovation.
  - Determine which policy combinations most decisively shift energy-sector innovation toward clean options.

### Key quantitative summaries (selected)
- Aggregate EPS effect on clean energy innovation: range 0.15 to 0.25 (statistically significant).
- Baseline EPS coefficient (Table 1, col 1): L.EPS = 0.184*** (4.46).
- Contribution to 1990–2010 increase in clean energy innovation: roughly 30 percent attributed directly to EPS change.
- Illustrative pricing equivalence: tightening EPS ≈ USD 70 increase in crude oil price per barrel (in estimation without year-FE).
- Short-run dynamics (Table 2): ∆EPS t to t+1 = 0.106*** (2.78); t to t+3 = 0.140* (1.80).
- Policy instrument examples (column 6, Table 3 / Section 3 and 5 examples):
  - Trading schemes: 0.0367*** (3.37) or up to 0.0548*** (4.11) in some specs.
  - Feed-in tariffs: 0.0283*** (2.71) to 0.0409*** (4.10) depending on spec.
  - Emission limits: 0.0385* (1.66) to 0.0632** (2.24).
  - R&D subsidies: 0.0486** (2.39) to 0.0621*** (6.06).
  - Carbon taxes: negative or insignificant (e.g., -0.0255 (-0.88); -0.0327 (-1.04)) with limited sample variation.

*Source: wpiea2021213-print-pdf — Sections 1–5*

### Section 1

### The Impact of Environmental Policy on Innovation in Clean Technologies — Section 1

### Introduction: motivation and scope
- Climate change is slow-moving; actions and consequences can be several decades apart, making technological progress unusually important.
- If clean alternatives become as cheap and efficient as dirty technologies, the energy transition can be done at relatively low costs in terms of growth and inclusiveness.
- From a social planner’s perspective the objective is to (i) avoid future climate-change related damages by reducing net carbon emissions, while (ii) minimizing transitional costs, including by stimulating technological progress among clean technologies.
- This paper studies empirically how climate change mitigating policies induce innovation in clean energy technologies using a panel of countries and patent-based innovation measures.

### Conceptual and theoretical context
- Traditional integrated assessment models treated technology as exogenous; later work (e.g., Nordhaus (2002), Popp (2004), Bosetti et al. (2009), Golosov et al. (2014), Hassler et al. (2020)) endogenized technology and showed climate policies can induce innovation.
- Key theoretical issues:
  - The optimal carbon tax can, in some models, fully offset the pollution externality even with endogenous technology (Golosov et al. (2014)), making high R&D subsidies unnecessary in a first-best situation.
  - Path dependencies and market-size effects (Acemoglu et al. (2012, 2016); Fried (2018); Barrett (2021)) can create positive returns to scale and strengthen the case for R&D subsidies as a complement to carbon pricing.
  - The elasticity of substitution across energy sources and the costs/payoffs of knowledge creation materially affect the role of innovation for emissions and welfare.

### Research questions and empirical approach
- The paper investigates:
  - Whether and to what extent the overall tightening of policies has contributed to clean energy-related innovation.
  - How more detailed policies vary in terms of innovation performance.
  - Whether potential gains in clean innovation are offset by reductions in innovation in other (dirty) technologies.
- Empirical strategy:
  - Panel estimation with country- and year-fixed effects to control for constant country-specific and time-varying global factors.
  - Controls include overall patenting across all technologies, commodity prices, education and innovation policies, and changing patenting culture.
  - Innovation proxied by climate-change mitigating patent families from PATSTAT.
  - Sample: 33 OECD countries from 1990 to 2016.

### Key empirical findings (aggregate and dynamics)
- Main conclusions (three-fold):
  - Tightening environmental policies since the early 1990s made a statistically and economically significant contribution to increased innovation in clean technologies.
  - At the aggregate level, the change in the average environmental policy stance between 1990 and 2010 had an effect roughly equivalent to a permanent increase in oil prices of USD 70 per barrel.
  - Effects materialize quickly: statistically significant almost immediately and strengthen over the first 2 to 3 years after the policy change.
- Policy-type findings:
  - Both market-based policies (including trading schemes and feed-in tariffs) and non-market policies (including emission limits on power plants and R&D subsidies) made positive, statistically significant, and roughly comparable contributions to clean innovation.
  - Results suggest substitutability among policy tools for stimulating innovation; multiple policy tools or combinations can promote clean innovation in comparable ways.
  - Estimated effects of carbon prices are not statistically significant in this sample, likely because carbon taxes were used by a small minority of countries and exploitable empirical variation is limited.
- Composition vs. addition of innovation (electricity technologies):
  - Environmental policies contributed to a shift in the composition of electricity-related innovation away from dirty towards clean and “grey” technologies.
  - The estimated effect on total electricity innovation is net positive — increased clean and grey innovation was not offset by a decrease in dirty innovation.
  - Continued innovation in dirty technologies, even if slower, reduces the cost competitiveness of clean technologies and thus can slow the energy transition; nevertheless, additional innovation reduces macro-economic transition costs if it translates into faster technological progress.

### Empirical contributions relative to the literature
- Advances on prior work:
  - Uses a more recent sample, better capturing the dramatic increase in clean innovation in the early 2000s and the flattening/partial reversal since 2010.
  - Employs a more standardized policy classification (OECD Environmental Policy Stringency, EPS) allowing analysis at different aggregation levels (aggregate EPS indicator and individual policy types).
  - Includes country- and time-fixed effects and controls for time-varying patenting behavior and global factors, improving identification relative to earlier cross-country studies (e.g., Johnstone et al. (2010)).

### Data and empirical framework (overview)
- Sample: 33 OECD countries, 1990–2016.
- Innovation measure:
  - Proxy: number of climate-change mitigating patent families associated with a country and year.
  - Constructed from PATSTAT: (i) associate each patent family with a country and year; (ii) classify whether technology codes justify inclusion as “clean”.
- Policy measure:
  - OECD’s Environmental Policy Stringency (EPS) index is hierarchical and standardized; indices reflect relative stringency over time and across countries and allow comparison of contributions across times or effects of changes relative to the historical distribution.
- Sensitivity analyses:
  - Robustness checks reported in the appendix show results are robust to different choices of innovation measures and empirical specifications.

*WP/21/213 — The Impact of Environmental Policy on Clean Innovation, Johannes Eugster; IMF Working Paper, August 2021; canonical URL: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021213-print-pdf.pdf*

### Section 2

### Section 2

### Measurement of innovation: patent families and sample
- Innovation is measured at the “patent family” level using the EPO’s DOCDB simple patent family classification to ensure each technology is counted once and to reduce bias from strategic patenting.
- The patent family is associated with:
  - the year of the first application (the “priority date”), and
  - the most common country of residence of the first inventors (the inventor country, not the applicant).
- Sample coverage and scope:
  - The sample covers 33 countries from 1990-2016.
  - Countries included: Australia, Austria, Belgium, Brazil, Canada, China, Czech Republic, Germany, Denmark, Finland, France, Greece, Hungary, Indonesia, India, Ireland, Italy, Japan, Korea, Netherlands, Norway, Poland, Portugal, Russia, Slovakia, Slovenia, Spain, Sweden, Switzerland, Turkey, the United Kingdom, the USA and South Africa.
- Rationale and robustness:
  - Using patent families mitigates double-counting from multiple national filings and continuation patents.
  - Potential disadvantage: patent family counts do not control for the economic or technological value of innovations. This is addressed empirically by including total innovation as a denominator and country- and year-fixed effects.
  - Sensitivity analyses consider broader counts (all applications) and narrower counts (“international” patent families with applications in at least two countries). Note: triadic patent families produce 44% zeroes versus 10% zeroes with “international” patent family.

### Selection and characterization of clean technologies
- Two classification approaches are used to identify clean innovations:
  - Haščič and Migotto (2015): environment-related technologies (ERT) based on technological codes to construct “climate change mitigation technologies related to energy generation, transmission or distribution” — proxy for “clean energy innovation”.
  - Dechezleprêtre et al. (2017): distinguishes dirty (fossil fuel combustion), clean (alternative energy sources), and “grey” (improving pollution efficiency of dirty) electricity technologies.
- Focus of analysis:
  - The paper focuses on clean technologies related to energy (rather than all climate-change mitigating technologies) for three reasons:
    - The energy sector is particularly affected by many climate policies captured by the EPS measure.
    - According to the IPCC (Bruckner et al. 2014), the energy supply sector was with 35% the largest contributor of global GHG emissions in 2010, of which almost ¾ was attributable to the production of electricity and heat.
    - Energy has been a major and particularly dynamic source of clean innovation: average share of total climate mitigating innovation related to energy increased from 16% in 1990 to 35% in 2010.
- Empirical patterns:
  - The share of clean energy innovation increased dramatically in the early 2000s, peaked toward the end of the first decade, and significantly declined in most countries since then.
  - Within electricity innovation, the rise and fall in the clean share was driven mainly by within-group changes (increase then decline in the clean share within electricity technologies).

### Environmental Policy Stringency (EPS) indicator
- EPS data from the OECD (Botta and Kozluk (2014) methodology):
  - Aggregates assessments over 4 hierarchical levels.
  - The most aggregate EPS can be decomposed into market-based policies (taxes, trading-schemes, feed-in tariffs) and non-market policies (emissions limits, R&D subsidies).
  - At the most disaggregated level, each policy tool receives a value by country-year relative to the sample distribution: 0 if not used and 6 if among the most stringent uses; intermediate thresholds based on in-sample distribution. Aggregations are simple averages.
- Interpretive limits:
  - EPS values are standardized relative to the sample; a given numeric value or change on one indicator is not directly comparable to the same value on another indicator in absolute policy terms (e.g., CO2 reductions or carbon-tax-equivalent USD amounts).
  - Illustrative comparisons are done by contrasting movements on distributions (e.g., from the 10th to the 90th percentile) or by changes between years (e.g., between 1990 and 2010).
- Observed policy trends:
  - General trend toward more stringent regulation across countries, with variation by policy tool.
  - Emissions limits tightened substantially and gradually and were already used by the majority of countries in the early 1990s.
  - R&D subsidies show a similar but less clear pattern.
  - Market-based policies (trading schemes and feed-in tariffs) were rare in the early 1990s and later gained popularity, affecting the average stringency measure.
  - Carbon taxes remained very sporadic through the sample period, limiting the empirical ability to detect significant effects of carbon taxes.
- Coverage note:
  - Data on environmental policy stringency exists up to 2014; with a one-period lag in regressions this limits estimation to 2015.

### Conceptual and empirical framework
- Production function and estimation:
  - Innovation X_{i,t} is modeled as a multiplicative production function: X_{i,t} = θ_{i,t} (H_{i,t})^α (u_{i,t})^γ, where u_{i,t} is accessible stock of knowledge and H_{i,t} is research effort.
  - Rewritten for estimation: X_{i,t} = exp(α_i + δ_t + β ln(H_{i,t}) + γ ln(u_{i,t}) + ε_{i,t}), with α_i and δ_t country- and time-fixed effects.
  - Estimation uses a conditional fixed effects Poisson model with robust standard errors.
- Key controls and construction:
  - Technology-specific stock of knowledge constructed as the discounted sum of total patent applications using a discount rate of 10% and the perpetual inventory method.
  - Overall innovation included to control for patenting culture and policies related to education and research.
  - Other control variables (used where data permit) include:
    - Proven oil and gas reserves as proxies for country-specific supply of fossil fuels (gas converted using 1 barrel of oil = 6,000 cubic feet of natural gas).
    - Household electricity prices as proxy for market-based incentives for innovation.
    - Product and labor market regulation indicators (e.g., electricity specific ETCR from OECD; labor market stringency from the Fraser Institute), noting the labor regulation variable constrains sample size and is not energy specific.
    - Crude oil prices to proxy global changes in fossil fuel supply and demand (often absorbed by year fixed effects).
- Fixed effects and identification:
  - Country- and year-fixed effects absorb time-invariant country characteristics and common dynamics (global economic/financial conditions, commodity prices), so policy coefficient identification relies on relative variation across countries over time.
  - All control variables are lagged by one year to account for knowledge production lags.

### Results overview (preview)
- Aggregate findings summarized at the start of Section III:
  - The aggregate EPS index significantly stimulates clean innovation.
  - Effects are economically sizeable, materialize relatively quickly, and are highly insensitive to different specifications.
  - Subsequent subsections will show:
    - (i) Most detailed policy tools support the strong aggregate effects.
    - (ii) The stimulus in clean innovation is not offset by a decline in innovation related to traditional technologies.

*Source: wpiea2021213-print-pdf - Section 2*

### Section 3

### The Effects of Environmental Policy Overall

### Aggregate Effects on Clean Energy Innovation
- Estimated effects of environmental policy stringency (EPS) on clean energy innovation are positive, highly statistically significant, and reasonably stable, ranging from 0.15 to 0.25.
- Baseline specification (column 1, Table 1) coefficient:
  - L.EPS = 0.184*** (4.46)
- Other specifications (Table 1 L.EPS coefficients and t-statistics):
  - Column (2): L.EPS = 0.249*** (6.25)
  - Column (3): L.EPS = 0.188*** (6.77)
  - Column (4): L.EPS = 0.232*** (6.74)
  - Column (5): L.EPS = 0.166*** (5.18)
  - Column (6): L.EPS = 0.203*** (5.48)
- Key control variables from Table 1 (coefficients and t-statistics):
  - L.ln(tech stock) coefficients:
    - (1): 0.600*** (9.73)
    - (2): 0.744*** (10.30)
    - (3): 0.507*** (11.47)
    - (4): 0.632*** (6.31)
    - (5): 0.552*** (21.77)
    - (6): 0.479*** (4.93)
  - L.ln(all tech) coefficients:
    - (1): 0.415*** (6.70)
    - (2): 0.320*** (4.04)
    - (3): 0.573*** (5.98)
    - (4): 0.574*** (4.39)
    - (5): 0.165* (1.70)
    - (6): 0.452*** (2.66)
  - L.ln(oil & gas reserves, bb):
    - Column (2): -0.136* (-1.66)
    - Column (5): -0.122 (-1.29)
    - Column (6): -0.131 (-1.41)
  - L.ln(electr. price HH, USD):
    - Column (3): 0.295*** (3.24)
    - Column (5): 0.180* (1.84)
    - Column (6): 0.357*** (3.52)
  - L.ETCR electricity:
    - Column (4): -0.105** (-2.56)
    - Column (6): -0.0615 (-1.16)
  - L. Labor market reg. (EFW):
    - Column (4): 0.0152 (0.69)
    - Column (6): -0.0394 (-1.35)
- Sample sizes and fixed effects (Table 1):
  - Country FE: Yes (all columns)
  - Year FE: Yes (all columns)
  - N by column: 781, 743, 608, 579, 424, 352
- Note on baseline: Column 1 used as baseline because inclusion of additional controls reduces sample size; its EPS coefficient is towards the lower end of the 0.15–0.25 range.

### Economic Magnitude and Interpretation
- Counterfactual comparison (predicted clean energy innovation in 2010 using actual EPS vs. EPS unchanged since 1990) suggests the change in EPS directly contributed roughly 30 percent of the increase in clean energy innovation between 1990 and 2010.
- Illustrative estimation including oil prices (but no year-FE) indicates tightening environmental policies was roughly equivalent to a USD 70 increase in the price of crude oil per barrel.
- Year-fixed effects absorb common trend toward tighter regulation; actual effect might be higher than implied by coefficients.
- Second-round effects via increased future knowledge stock are present but appear of second-order importance.

### Robustness and Additional Controls
- Oil and gas reserves coefficient not statistically significant in aggregate but has expected sign and may be more relevant for disaggregated electricity innovation.
- Labor market regulation coefficient is insignificant and changes sign across specifications; labor regulation not specific to energy sector.
- ETCR electricity (product market regulation) shows that less stringent regulation (lower ETCR) tends to support innovation, though significance declines in the most complete specification; potential multicollinearity with electricity prices noted.

---

### Dynamics: Speed of the Effect (Table 2)
- Local projection estimation shows positive effects on innovation materialize already in the year tightening becomes effective.
- ∆EPS coefficients from Table 2 (log-difference in CCM Energy patent families between t−1 and t+j):
  - t to t+1: ∆EPS = 0.106*** (2.78)
  - t to t+2: ∆EPS = 0.126** (2.05)
  - t to t+3: ∆EPS = 0.140* (1.80)
  - t to t+4: ∆EPS = 0.117 (1.06)
  - t to t+5: ∆EPS = 0.114 (1.04)
- L.ln(tech stock) in Table 2 (coefficients and t-statistics) — showing negative sign due to maturation effects:
  - t to t+1: -0.199*** (-4.29)
  - t to t+2: -0.351*** (-4.28)
  - t to t+3: -0.492*** (-4.47)
  - t to t+4: -0.586*** (-5.10)
  - t to t+5: -0.658*** (-5.69)
- ln(oil price) coefficients (Table 2):
  - t to t+1: 0.222*** (3.86)
  - t to t+2: 0.376*** (4.30)
  - t to t+3: 0.340*** (3.32)
  - t to t+4: 0.295*** (2.90)
  - t to t+5: 0.147 (1.23)
- ∆ln(all tech) coefficients (Table 2):
  - t to t+1: 0.545*** (4.30)
  - t to t+2: 0.428** (2.47)
  - t to t+3: 0.372* (1.76)
  - t to t+4: 0.388* (1.93)
  - t to t+5: 0.394*** (2.96)
- Table 2 sample sizes (N) and r2:
  - N: 649, 616, 584, 552, 520
  - r2: 0.285, 0.289, 0.339, 0.387, 0.422
- Interpretation:
  - Innovation response increases for another two years to about ¾ of the baseline Table 1 coefficient before becoming imprecise.
  - Negative coefficient on knowledge stock reflects higher innovation growth when technologies are new and slowing as they mature.
  - Patent timing conventions (priority date) imply measured innovation may appear quickly though commercial availability lags; front-loading of patent applications could move measured innovation earlier but unlikely to change qualitative result that innovation reacts rapidly.

### Effects of More Detailed Climate Policies (Table 3)
- Analysis uses EPS sub-indicators to assess specific policy instruments.
- Main finding: Both non-market policies (emission limits, R&D subsidies) and market policies (trading schemes, feed-in tariffs) made statistically significant contributions to clean innovation; carbon taxes are a notable exception with highly insignificant effect.
- Table 3 coefficients and t-statistics (selected):
  - L.ln(tech stock) (columns 1–6): coefficients range from 0.552*** to 0.655*** with t-statistics (9.31) to (14.62) to (12.08) across columns.
    - Examples:
      - Column (1): 0.592*** (11.62)
      - Column (2): 0.569*** (14.62)
      - Column (3): 0.655*** (11.41)
      - Column (4): 0.584*** (10.89)
      - Column (5): 0.552*** (9.31)
      - Column (6): 0.582*** (12.08)
  - L.ln(all tech) (columns 1–6):
    - Column (1): 0.445*** (6.47)
    - Column (2): 0.464*** (7.98)
    - Column (3): 0.374*** (6.01)
    - Column (4): 0.398*** (6.70)
    - Column (5): 0.535*** (7.59)
    - Column (6): 0.443*** (8.29)
  - Co2 taxes:
    - Column (1): 0.00595 (0.27)
    - Column (6): -0.0255 (-0.88)
  - Trading schemes:
    - Column (2): 0.0377** (2.30)
    - Column (6): 0.0367*** (3.37)
  - Feed-in tariffs:
    - Column (3): 0.0326*** (5.02)
    - Column (6): 0.0283*** (2.71)
  - Emission limits:
    - Column (4): 0.0559*** (2.63)
    - Column (6): 0.0385* (1.66)
  - R&D subsidies:
    - Column (5): 0.0592*** (4.03)
    - Column (6): 0.0486** (2.39)
- Table 3 sample sizes and fixed effects:
  - Year FE: Yes (all columns)
  - Country FE: Yes (all columns)
  - N: 807, 804, 807, 807, 807, 804
- Interpretation and caveats:
  - Carbon taxes show no significant effect largely because only slightly more than 10 percent of countries used carbon taxes in 2015; limited variation over time and across countries reduces power to detect effects.
  - Trading schemes (which put a variable price on carbon) and electricity prices show significant innovation-inducing effects, aligning with literature finding relative price changes in favor of cleaner technologies drive innovation.
  - Historical contribution graphs (Figure 5 described qualitatively) use column 6 coefficients to show:
    - Left graph: contribution relative to 1990 levels.
    - Middle graph: effect of a one-standard-deviation change over entire sample.
    - Right graph: effect of moving from 10th to 90th percentile in cross-sectional distribution of 2010.
  - R&D subsidies had modest historical contribution because their average score did not change much over time, but their effect scales up when measured by total variation or cross-sectional variation.
  - Tightening of emission limits has been among the more powerful historical drivers of clean innovation, though estimates may be noisier due to heterogeneity in how limits are defined and adjusted.
  - The independent positive effects (except carbon taxes) of different policy tools support integrated assessment models with endogenous growth that market-based policies targeting emissions can strengthen innovation incentives, and they temper fears that non-market mechanisms inherently provide muted incentives to innovate.

### Key Quantitative Summaries and Notes
- Overall EPS effect range: 0.15 to 0.25 on clean energy innovation (statistically significant).
- Baseline EPS coefficient (Table 1, col 1): 0.184*** (4.46).
- Policy contribution to innovation 1990–2010: roughly 30 percent of the increase in clean energy innovation attributed directly to EPS change.
- Illustrative pricing equivalence: tightening EPS ≈ USD 70 increase in crude oil price per barrel (in estimation without year-FE).
- Short-run dynamics: ∆EPS = 0.106*** (2.78) for t to t+1, peaking near 0.140* before losing statistical precision after two-to-three years.
- Specific policy instrument effects (column 6, Table 3, examples):
  - Trading schemes: 0.0367*** (3.37)
  - Feed-in tariffs: 0.0283*** (2.71)
  - Emission limits: 0.0385* (1.66)
  - R&D subsidies: 0.0486** (2.39)
  - Carbon taxes: -0.0255 (-0.88) — statistically insignificant, limited variation in sample.

*Source: wpiea2021213-print-pdf - Section 3*

### Section 4

### Section 4

### Effect of Individual Environmental Policy Tools
- Comparing movements on the stringency-distribution of individual policy indicators makes estimated effects independent of magnitude of the indicator, but estimated effects depend on how much the use of a policy tool has differed over the two dimensions of the sample.
- Positive and individually significant effects of different policy tools point to some substitutability among them as far as innovation is concerned.
- Policy implication: climate policy tools can be used and potentially combined focusing on political feasibility and expected effect on carbon emissions.

### Clean, Grey, Dirty and Total Electricity Innovation
- Purpose: expand analysis from clean technologies to total innovation by technology type within electricity, using the Dechezleprêtre et al (2017) classification of clean, grey, and dirty electricity technologies.
- Clean electricity innovation identified by Dechezleprêtre et al. (2017) is highly correlated with OECD clean energy innovation: the average country-specific correlation over time is 95%.
- Model differences versus baseline:
  - Controls for available knowledge stocks of specific technology and stock of electricity technology overall.
  - Columns (1)–(3) impose a coefficient of 1 on total patenting in electricity innovation (sum of clean, grey, and dirty). Interpretation: elasticity of the particular type of innovation controlling for electricity innovation overall.
  - Inclusion of oil and gas reserves; these affect clean and dirty electricity innovation differently.
- Key empirical results (Table 4 coefficients from a fixed effects Poisson estimation; t statistics in parentheses):
  - Clean (column 1):
    - L.EPS 0.0625*** (2.64)
    - L.ln(all tech) 0.0572 (0.58)
    - L.ln(oil & gas reserves) -0.103** (-2.03)
    - L.ln(all electr stock) -0.182 (-0.95)
    - L.ln(spec electr stock) 0.125 (0.92)
  - Grey (column 2):
    - L.EPS 0.150*** (3.39)
    - L.ln(all tech) -0.352*** (-3.40)
    - L.ln(oil & gas reserves) -0.0836 (-1.37)
    - L.ln(all electr stock) 0.171 (0.80)
    - L.ln(spec electr stock) 0.513*** (6.67)
  - Dirty (column 3):
    - L.EPS -0.0402** (-2.14)
    - L.ln(all tech) 0.0520** (2.28)
    - L.ln(oil & gas reserves) 0.0361 (1.33)
    - L.ln(all electr stock) -0.189 (-0.47)
    - L.ln(spec electr stock) 0.131 (0.32)
  - Total (column 4):
    - L.EPS 0.201*** (3.49)
    - L.ln(all tech) 0.416*** (4.07)
    - L.ln(oil & gas reserves) -0.0419 (-0.31)
    - L.ln(all electr stock) 0.638*** (4.14)
    - ln(all electr) coefficient constrained at 1.00
- Observations and fit:
  - Year FE: Yes; Country FE: Yes
  - N 743 for each column
- Findings and interpretation:
  - Environmental policies increased shares of clean innovation and increased grey innovation even more (both coefficients statistically significant at least at the 5% level).
  - Magnitude of grey innovation stands out but requires cautious interpretation due to lower initial levels and reduced robustness of the grey coefficient in some specifications.
  - Column 4: effect on total electricity innovation is strong and positive — the relative decrease in dirty innovation (column 3) was more than offset by increased innovation in clean and grey categories.
  - Implication: a complete crowding out of innovation in traditional technologies may be too pessimistic; induced innovation can be additional rather than purely substitutive.

### Robustness and Sensitivity Analysis
- Robustness checks focus on different measures of innovation and dropping key innovator countries.
- Measures of innovation:
  - Tables A1 and A2 compare patent families, all patents, and international patent families (patent families with applications in at least two patent offices).
  - Top innovating countries hardly change with the measure, but relative sizes do (e.g., going from patent families to all applications increases innovation score mostly for the US and Japan; going from all families to only international patent families reduces importance of China).
  - Results remain qualitatively highly robust; coefficients stay statistically significant except in few cases where additional controls cut sample by almost half. Coefficients decline somewhat, particularly with international patent families.
- Dropping key innovator countries:
  - Table A3: dropping any key innovator does not materially change results. Coefficient increases slightly when Korea is dropped and decreases somewhat when Japan is dropped; statistical significance remains unaffected.
- Dynamics of the effect:
  - Dynamic results use baseline specification and alternative specifications controlling for oil and gas reserves and electricity prices.
  - Inclusion of oil and gas reserves weakens results very slightly.
  - Electricity prices, though themselves statistically insignificant, tend to reinforce results substantially, particularly two to four years after policy change; sample change (reduced coverage of electricity prices for emerging markets) explains part of reinforcement.
- Effects of specific policies:
  - Tables A6 and A7: controlling for oil and gas reserves and electricity prices at household level does not materially change results. With the exception of carbon prices, coefficients of policy variables generally remain statistically significant and very similar in magnitude.
- Robustness for types of electricity innovation:
  - Tables A8 and A9:
    - A8: unconstrained coefficient on total electricity innovation is close to 1 for dirty and clean innovation but significantly larger for grey innovation. Precision slightly reduced; overall similar results.
    - A9: removing oil and gas reserves makes the grey coefficient insignificant; including electricity prices reduces sample by about 20% and makes the dirty coefficient insignificant while clean and grey coefficients remain very close to those in Table 4.

### Aggregate Effect and Overall Conclusions
- Aggregate effect on clean energy innovation: evidence points to a statistically significant and economically important contribution, roughly equivalent to a permanent 70 USD increase in the price of a barrel of crude oil.
- Main conclusions:
  - Environmental policies induced additional innovation in clean technologies and shifted innovation away from dirty toward clean and grey technologies, with a net positive effect on electricity innovation overall.
  - Induced innovation reduces macro-economic costs of an ambitious climate agenda by (i) helping clean technologies become viable, cost-effective alternatives more quickly and (ii) increasing technological progress overall.
  - However, because induced innovation largely complements rather than replaces innovation in traditional technologies, continued innovation in dirty technologies may slow the energy transition if it postpones replacement by cleaner alternatives.
- Policy recommendation and research agenda:
  - Additional policy efforts are necessary and politically feasible because various environmental policy tools can induce additional clean-technology innovation and can be combined based on political preferences and expected effects on emissions.
  - Further research is needed to compare the effectiveness and costs of different policy tools, to consider greenhouse gas emission implications, and to determine which combinations most decisively shift energy-sector innovation toward clean options.

*Source: wpiea2021213-print-pdf - Section 4*

### Section 5

### wpiea2021213-print-pdf - Section 5

### Aggregate effect when dropping key innovator countries
- Dependent variable: CCM Energy patent families; N = 755 for all columns.
- L.EPS coefficients (with t statistics):
  - Country dropped CHN: 0.191*** (6.09)
  - Country dropped DEU: 0.190*** (4.68)
  - Country dropped JPN: 0.129*** (3.95)
  - Country dropped KOR: 0.227*** (5.60)
  - Country dropped USA: 0.168*** (4.16)
- L.ln(tech stock) coefficients (with t statistics):
  - CHN dropped: 0.549*** (9.17)
  - DEU dropped: 0.600*** (10.02)
  - JPN dropped: 0.555*** (9.95)
  - KOR dropped: 0.758*** (7.93)
  - USA dropped: 0.617*** (9.53)
- L.ln(all tech) coefficients (with t statistics):
  - CHN dropped: 0.514*** (4.28)
  - DEU dropped: 0.415*** (6.97)
  - JPN dropped: 0.428*** (7.89)
  - KOR dropped: 0.270*** (2.67)
  - USA dropped: 0.392*** (6.30)

### Dynamics of the effect, including Oil & Gas reserves (Log-difference in CCM Energy patent families between t+1 and t+j)
- Sample sizes and r2:
  - For t+1: N = 632, r2 = 0.290
  - For t+2: N = 601, r2 = 0.288
  - For t+3: N = 570, r2 = 0.336
  - For t+4: N = 539, r2 = 0.378
  - For t+5: N = 508, r2 = 0.410
- EPS coefficients (with t statistics):
  - t+1: 0.101** (2.58)
  - t+2: 0.115* (1.92)
  - t+3: 0.135 (1.63)
  - t+4: 0.110 (0.93)
  - t+5: 0.113 (0.97)
- L.ln(tech stock) coefficients:
  - t+1: -0.221*** (-4.03)
  - t+2: -0.385*** (-4.11)
  - t+3: -0.529*** (-4.26)
  - t+4: -0.622*** (-4.94)
  - t+5: -0.688*** (-5.15)
- ln(oil price) coefficients:
  - t+1: 0.217*** (3.73)
  - t+2: 0.365*** (4.05)
  - t+3: 0.331*** (3.12)
  - t+4: 0.292*** (2.83)
  - t+5: 0.147 (1.24)
- ∆L.ln(all tech) coefficients:
  - t+1: 0.542*** (4.29)
  - t+2: 0.428** (2.47)
  - t+3: 0.377* (1.77)
  - t+4: 0.388* (1.95)
  - t+5: 0.392*** (2.96)
- ln(oil & gas reserves, bb) coefficients:
  - t+1: -0.00209 (-0.05)
  - t+2: -0.0401 (-0.64)
  - t+3: -0.0891 (-1.09)
  - t+4: -0.110 (-1.11)
  - t+5: -0.169 (-1.56)
- Note: Country-fixed effects, a trend, EPS values between t and t+j, and three lags of EPS and growth of innovation included but not reported.

### Dynamics of the effect, including Electricity prices (Log-difference in CCM Energy patent families between t+1 and t+j)
- Sample sizes and r2:
  - t+1: N = 500, r2 = 0.324
  - t+2: N = 473, r2 = 0.339
  - t+3: N = 447, r2 = 0.377
  - t+4: N = 421, r2 = 0.461
  - t+5: N = 395, r2 = 0.471
- EPS coefficients (with t statistics):
  - t+1: 0.117** (2.74)
  - t+2: 0.200*** (3.55)
  - t+3: 0.237*** (3.44)
  - t+4: 0.266** (2.69)
  - t+5: 0.232** (2.39)
- L.ln(tech stock) coefficients:
  - t+1: -0.186*** (-3.55)
  - t+2: -0.305*** (-3.44)
  - t+3: -0.410*** (-3.23)
  - t+4: -0.516*** (-3.43)
  - t+5: -0.616*** (-3.97)
- ln(oil price) coefficients:
  - t+1: 0.229*** (3.14)
  - t+2: 0.388*** (3.61)
  - t+3: 0.353*** (2.88)
  - t+4: 0.328*** (3.27)
  - t+5: 0.186 (1.49)
- ∆ln(all tech) coefficients:
  - t+1: 0.624*** (8.88)
  - t+2: 0.552*** (6.03)
  - t+3: 0.512*** (4.95)
  - t+4: 0.473*** (4.57)
  - t+5: 0.389*** (4.11)
- ln(electr. price HH, USD) coefficients:
  - t+1: -0.0364 (-0.55)
  - t+2: -0.0461 (-0.41)
  - t+3: -0.00958 (-0.07)
  - t+4: 0.0163 (0.13)
  - t+5: 0.0549 (0.46)

### Effect of individual policies, including Oil & Gas reserves (Dependent: CCM Energy patent families)
- Sample sizes: N = 763 (columns 1, 3, 4, 5), N = 760 (columns 2, 6).
- L.ln(tech stock) coefficients (with t statistics): column-wise
  - 0.789*** (9.85); 0.726*** (9.00); 0.827*** (11.61); 0.759*** (7.91); 0.775*** (9.45); 0.753*** (9.03)
- L.ln(all tech) coefficients (with t statistics):
  - 0.254*** (2.95); 0.305*** (3.77); 0.234*** (3.05); 0.251*** (2.72); 0.328*** (3.75); 0.305*** (3.89)
- L.ln(oil & gas reserves, bb) coefficients:
  - -0.0165 (-0.22); -0.0225 (-0.30); -0.0586 (-0.89); -0.0883 (-0.97); -0.00301 (-0.04); -0.0894 (-1.31)
- Individual policy coefficients (selected):
  - trading schemes: 0.0548*** (4.11) in one column; 0.0337*** (3.69) in another
  - Feed-in tariffs: 0.0383*** (3.43) in one column; 0.0409** (2.56) in another
  - emission limits: 0.0564*** (3.03) and 0.0527*** (3.36)
  - R&D subsidies: 0.0621*** (6.06) and 0.0569*** (3.54)
- Co2 taxes coefficients shown: -0.00533 (-0.31) in one column and -0.0327 (-1.04) in another (not always present).

### Effect of individual policies, including Electricity prices (Dependent: All CCM Energy patent applications)
- N = 608 for all columns.
- L.ln(tech stock) coefficients (with t statistics): 0.526*** (11.79); 0.514*** (12.57); 0.581*** (13.12); 0.505*** (12.69); 0.490*** (8.19); 0.536*** (10.83)
- L.ln(all tech) coefficients: 0.598*** (5.14); 0.587*** (5.16); 0.553*** (5.71); 0.512*** (4.56); 0.684*** (5.14); 0.526*** (5.07)
- L.ln(electr. price HH) coefficients: 0.283** (2.17); 0.219* (1.76); 0.365*** (3.01); 0.221** (2.26); 0.299** (2.28); 0.265** (2.56)
- Individual policy coefficients (selected):
  - trading schemes: 0.0267** (2.34) in one column; 0.0295*** (3.11) in another
  - Feed-in tariffs: 0.0409*** (4.10); 0.0407*** (4.03)
  - emission limits: 0.0632** (2.24) and 0.0435 (1.37)
  - R&D subsidies: 0.0478*** (2.91) in one column; 0.0227 (1.59) in another
- Co2 taxes coefficients: 0.0220 (0.53) in one column and -0.0328 (-1.03) in another (not always present).

### Relative electricity innovation with unconstrained effects of total electricity innovation
- Dependent variables: Patent families related to different types of electricity technologies; N = 743 for all three columns.
- L.EPS coefficients (with t statistics):
  - clean: 0.0537* (1.86)
  - grey: 0.0911* (1.93)
  - dirty: -0.0372* (-1.83)
- L.ln(all tech) coefficients:
  - clean: 0.0177 (0.18)
  - grey: -0.484*** (-3.99)
  - dirty: 0.0634*** (2.90)
- L.ln(spec electr stock) coefficients:
  - clean: -0.0508 (-0.68)
  - grey: 0.575*** (8.56)
  - dirty: 0.0261 (0.09)
- L.ln(all electr stock) coefficients:
  - clean: -0.0343 (-0.19)
  - grey: -0.148 (-0.90)
  - dirty: -0.0807 (-0.27)
- L.ln(oil & gas reserves):
  - clean: -0.131** (-2.39)
  - grey: -0.0812** (-1.99)
  - dirty: 0.0394 (1.50)
- ln(all electr) coefficients: 1.092***, 1.361***, 0.982*** (t statistics not listed in excerpt).

### Relative electricity innovation with fewer or additional controls
- Two panels (N = 781 for columns 1–3; N = 579 for columns 4–6).
- L.EPS coefficients:
  - Column 1 (clean): 0.0487*** (3.00)
  - Column 2 (grey): 0.0786 (1.28)
  - Column 3 (dirty): -0.0393*** (-3.13)
  - Column 4 (clean): 0.0304 (1.39)
  - Column 5 (grey): 0.156*** (6.58)
  - Column 6 (dirty): -0.00441 (-0.46)
- L.ln(all tech) coefficients show variation across specifications (examples):
  - Column 1: -0.0880 (-1.33)
  - Column 2: -0.254*** (-2.90)
  - Column 3: 0.105*** (4.02)
  - Column 6: 0.0686*** (4.08)
- L.ln(spec stock) coefficients:
  - Column 1: 0.443*** (3.48)
  - Column 2: 0.508*** (8.39)
  - Column 3: 0.453* (1.81)
  - Column 4: 0.308** (2.24)
  - Column 5: 0.718*** (8.10)
  - Column 6: 0.473 (1.36)
- L.ln(all electr stock) coefficients:
  - Column 1: -0.382** (-2.44)
  - Column 2: 0.0161 (0.11)
  - Column 3: -0.561** (-2.45)
  - Column 4: -0.474*** (-2.60)
  - Column 5: -0.239 (-1.14)
  - Column 6: -0.473 (-1.39)
- Controls included in some specifications:
  - L.ln(oil & gas reser.) appears with coefficient -0.0745* (-1.66) in one column and -0.0656 (-1.35) in another.
  - L.ln(electr. price HH) appears with coefficients -0.110* (-1.77), 0.329*** (4.57), and 0.0294 (1.19) across specifications.
- Note: In these regressions, the coefficient on overall electricity innovation is constrained at 1. Robust standard errors in parentheses. Country- and year-fixed effects are included in related Poisson estimations elsewhere in the section.

*Italic: Source: wpiea2021213-print-pdf - Section 5*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021213-print-pdf.pdf_