Expert workshop on the creation and uses of combined environmental and economic performance datasets at the micro-level - 10-11 July 2018 - OECD, Paris
Measuring the Impact of Clean Technologies Using French Firm-Level Data
1. Measuring Clean Technologies at
the Firm Level
Damien Dussaux (PSL Mines ParisTech Cerna)
OECD Workshop
« Combining economic and environmental microdata »
Paris, July 10/11
2. Why is it important?
o Evaluating environmental policies
Long run effect on competitiveness (Green race and first mover
advantage)
Clean Air Act’s Best Available Control Technology
o Analyse diffusion of clean technologies
Trade policies
Intellectual Property Right
o Evaluating innovation policy
Ademe finance clean investments via Programme d’Investissements
d’Avenir
Bpifrance finance SMEs via Fonds Ecotechnologies or Prêt Vert
3. Typology of indicators
1. Innovation in clean technologies
o Patents in clean technologies (PATSTAT)
o Introduction of material or energy saving innovation (CIS)
2. Adoption of clean technologies
o Pollution Abatement Cost and Expenditure (Antipol)
o Investment in energy efficiency (Insee)
3. Consequences of technology adoption
o Total Factor Productivity and Material productivity
o Energy intensity and Pollution intensity
o Green goods and services
4. Community Innovation Survey (CIS)
o Conducted in several European countries:
1997, 2001, 2005, 2007, 2009, 2011, 2013, 2015
Industry, wholesale, transport, some services.
o Stratified (activity, size) sample of 20,000 French firms >=
10
o Set of questions evolve
o Some surveys include questions regarding the importance
of:
material saving innovations
pollution abatement
labour saving innovations and others
o For all waves, firms can answer
1. not relevant,
Weak panel
structure
Discrete
Choice Model
Fixed-effects
not feasible
5. CIS panel structure
Number of
years
available
Number
of firms
Percentage
1 58,713 70%
2 15,723 19%
3 5,046 6%
4 2,032 2%
5 1,092 1%
6 554 < 1%
7 424 < 1%
8 228 < 1%
Only a few firms
are sampled
several years
Almost no
panel
structure
Source: French CIS, manufacturing.
6. Variable Obs Mean Std. Dev. Min Max
Process innovation (0/1) 20,856 0.95 0.2 0 1
Product innovation (0/1) 15,666 0.98 0.1 0 1
Material saving 46,177 0.96 1.1 0 3
Labour saving 33,491 1.5 1.2 0 3
Flexibility 51,344 1.7 1.1 0 3
Environmental impacts 46,179 1.0 1.1 0 3
Market share increase 33,534 2.2 1.0 0 3
Range of products 33,532 2.2 1.0 0 3
Quality improvement 33,493 2.2 1.0 0 3
CIS summary statistics
Source: French CIS, manufacturing
7. Discrete Choice Model Estimations are
biased
o « Fixed-effects » Probit
Inclusion of dummy variables for each unit
Incidental parameters problem persistent bias
decreasing in T (Neyman and Scott, 1948)
o Solutions exist but require good panel structure
Conditional MLE in the binary logit model is a solution
(Chamberlain, Review of Economic Studie, 1980)
Carro (Journal of Econometrics, 2007)’s Modified MLE
reduces bias from O(1/T) to O(1/T²)
8. Pollution Abatement Cost and
Expenditure (PACE)
o Antipol (Insee) survey asks plants (siret) about their investment in capital
or knowledge to protect the environment:
Waste water, air pollution, solid waste,
Noise, Soil, Biodiversity,
Energy efficiency not covered
o Around 11,000 industrial plants sampled yearly:
o all plants ≥ 250 employees surveyed
o plants between 20 and 249 employees randomly sampled
o Sampled stratified on economic activity and size
o 80% of the plants respond
o Current expenditure are collected every 3 year since 2004
o Distinction between end of pipe and integrated technologies
9. Antipol panel structure
Number of
years available
Number
of firms
Percentage
1 10,980 27%
2 6,720 17%
3 4,662 12%
4 4,554 11%
5 2,528 6%
6 2,039 5%
7 1,402 3%
8 1,101 3%
9 952 2%
10 750 2%
Etc…
Only a few plants
are sampled
several years
Weak
panel
structure
Source: Antipol
10. Variable Obs Mean Std. Dev. Min Max
Capital expenditure 131,222 238 1,775 0 295,292
End of pipe investment 131,335 197 1,599 0 295,292
Integrated investment 131,261 41 621 0 82,000
End of pipe – Air 105,677 41 475 0 37,896
Integrated – Air 92,735 18 278 0 37,383
End of pipe – Solid
waste
105,987 19 280 0 63,745
Integrated – Solid waste 91,943 2 39 0 4,373
Current expenditure 9,516 284 1,351 0 79,541
Environmental tax paid 17,621 148 730 0 54,805
Antipol summary statistics
Source: French Antipol. Thousand
euros.
11. Variable Obs Mean Std. Dev. Min Max
Capital expenditure 131,222 238 1,775 0 295,292
End of pipe investment 131,335 197 1,599 0 295,292
Integrated investment 131,261 41 621 0 82,000
End of pipe – Air 105,677 41 475 0 37,896
Integrated – Air 92,735 18 278 0 37,383
End of pipe – Solid
waste
105,987 19 280 0 63,745
Integrated – Solid waste 91,943 2 39 0 4,373
Current expenditure 9,516 284 1,351 0 79,541
Environmental tax paid 17,621 148 730 0 54,805
Antipol summary statistics
Source: French Antipol. Thousand
euros.
12. From plants to firm-level
o Antipol data and EACEI data are at the plant level
o But most economic performance indicators are at the firm level
o Also, some decisions are taken at the level of the firm
o Several ways to reconcile the datasets:
Aggregate plant data at the firm level using employment data and merge at the
firm level
Compute RHS that are weighted average of plant-level variables where the
weights are the plants’ share of total employees
o Within-firm reallocations are important
Is the policy generating net employment loss or reallocation within firms?
A good panel structure is key to do that
13. Firm-Level Productivity
o Estimation of production function
o 𝑦𝑖𝑡 = 𝛼𝑙 𝑙𝑖𝑡 + 𝛼 𝑘 𝑘𝑖𝑡 + 𝛼 𝑚 𝑚𝑖𝑡 + 𝜇𝑖 + 𝛿𝑡 + 𝜔𝑖𝑡 + 𝜖𝑖𝑡
o 𝜔𝑖𝑡 is Total Factor Productivity
o 𝑦𝑖𝑡 is logged output usually measured by turnover
𝜔𝑖𝑡 is contaminated by firms mark-up key to understand policy impacts
Physical quantity allows to distinguish technological change from market
power
But firms are multi-products
Quality ?
o Material physical quantity are not available either material
14. Conclusion
o Getting better but some problems remain important
o Inconsistency in the units sampled generate problems for econometric
estimations but also for aggregate analyses (output and employment
reallocation, firms entry and exit)
o Continuous measures are needed to use good models
o Some dimensions are still absent :
R&D in green technologies,
investment in energy efficient technologies,
material and output physical quantities.
oEnvironmental policies can be tricky to measure e.g. (i) numerous
exemptions for fossil fuel tax (ii) direct R&D subsidies