RES EARCH

CLIMATE POLICY sures implemented over the past 2 decades

across 41 countries from six continents, whose
Climate policies that achieved major emission emissions altogether account for 81% of total
global emissions in 2019 (19). The aim of this
reductions: Global evidence from two decades large-scale, cross-country assessment is to guide
societies and decision-makers in effectively
Annika Stechemesser1,2,3*, Nicolas Koch1,2,4*, Ebba Mark5,6,7, Elina Dilger1, Patrick Klösel1,2, ratcheting up NDCs under the Paris Agree-
Laura Menicacci1, Daniel Nachtigall8, Felix Pretis5,9, Nolan Ritter1,2, Moritz Schwarz1,5,6,10, ment by providing tangible evidence on which
Helena Vossen1, Anna Wenzel1 policy instruments have the potential to achieve
large emission reductions. At the heart of our
Meeting the Paris Agreement’s climate targets necessitates better knowledge about which climate analysis is a meticulously collated climate pol-
policies work in reducing emissions at the necessary scale. We provide a global, systematic ex post icy database from the Organisation for Eco-
evaluation to identify policy combinations that have led to large emission reductions out of 1500 climate nomic Co-operation and Development (OECD),
policies implemented between 1998 and 2022 across 41 countries from six continents. Our approach which constitutes the most comprehensive,
integrates a comprehensive climate policy database with a machine learning–based extension of internationally harmonized policy inventory
the common difference-in-differences approach. We identified 63 successful policy interventions with to date and addresses important prior data
total emission reductions between 0.6 billion and 1.8 billion metric tonnes CO2. Our insights on effective limitations. It is global, disaggregated by re-

Downloaded from https://www.science.org at Loughborough University on September 24, 2024

but rarely studied policy combinations highlight the important role of price-based instruments in levant economic sectors (buildings, electricity,
well-designed policy mixes and the policy efforts necessary for closing the emissions gap. industry, and transport), covers both policy
adoptions and the tightening of existing pol-

M
icies, and is of high quality, ensured by draw-
eeting the Paris Agreement’s climate sment of climate policy interventions that pays ing on official data verified by countries. Its
objectives necessitates decisive policy careful attention to the diversity among policy consistent, theory-based categorization of 48
action (1). Although the agreement seeks instruments and their mutual complementarity. distinct climate policy instrument types en-
to limit global average temperature in- Assembling such a global stocktake of ef- ables systematic assessments of synergies be-
crease to “well below 2°C above pre- fective climate policy interventions is so far tween different instruments.
industrial levels and pursuing efforts to limit hampered by two main obstacles: First, even The empirical challenge is that the candi-
the temperature increase to 1.5°C,” its success though there is a plethora of data on legislative date pool of effective policy interventions is
critically hinges on the implementation of ef- frameworks and pledged national emission re- too large to tackle for standard evaluation
fective climate policies at the national level. ductions (8–10), systematic and cross-nationally tools with their focus on single, known inter-
However, scenarios from global integrated as- comparable data about the specific types and ventions. For example, controlling for all pos-
sessment models suggest that the aggregated mixes of implemented policy instruments are sible policies from the OECD database in a
mitigation efforts communicated through na- lacking. Second, empirical tools are typically conventional policy evaluation setting would
tionally determined contributions (NDCs) fall tailored to isolate the effect of single policy label all countries as treated and leave us with
short of the required emission reductions (2), instruments and are predominantly applied to very few degrees of freedom and little if any
and the United Nations (UN) estimates quantify policies that researchers subjectively deem par- statistical power. Rather than resorting to a
a median emission gap of 23 billion metric tonnes ticularly relevant. Consequently, only a few subjective selection of particular policies to
(Gt) carbon dioxide equivalent (CO2-eq) by 2030 headline policy instruments receive much at- analyze, we aimed to identify large reductions
(3). The persistence of this emissions gap is not tention. For example, carbon pricing is well- in emissions and subsequently attribute them
only caused by an ambition gap but also a gap studied in high-income countries (11–13), whereas to potential policy interventions. We did so by
in the outcomes that adopted policies achieve in countless alternative policy instruments such applying a machine learning–based extension
terms of emission reductions (4). This raises the as standards remain sparsely evaluated, spe- of the standard difference-in-differences (DID)
fundamental question as to which types of pol- cifically in lower-income countries. Last but approach to evaluate policy.
icy measures are successfully causing mean- not least, there are few tools to empirically First, we exploited methods of break detec-
ingful emission reductions. Despite more than evaluate mixes of multiple, simultaneously com- tion from the time series literature in a gen-
two decades of experience with thousands of bined policy instruments. Thus, although policy- eralized DID setting using well-established
diverse climate policy measures gained around makers heavily rely on policy mixes (14–16), variable selection tools from the machine learn-
the world, there is consensus in neither science assessing which combinations of policies ef- ing literature to generate data-driven hypothe-
nor policy on this question (5–7). This high- fectively unfold complementarities and can ses about previously known or unknown policy
lights the need for a fine-grained global asses- deliver stronger emission reductions is poorly interventions with meaningful emission reduc-
understood. For all these reasons, the emis- tion effects. In comparison with the standard
sion gap is intertwined with an equally notable DID approach, which requires a priori knowl-
1
Potsdam Institute for Climate Impact Research (PIK),
knowledge gap on effective climate policies. edge about where and when a small subset of
Potsdam, Germany. 2Mercator Research Institute on Global This also hampers learning in Intergovern- policies was implemented, we neither made
Commons and Climate Change (MCC), Berlin, Germany. mental Panel on Climate Change (IPCC) as- any assumptions about which country is treated
3
Institute of Physics, University of Potsdam, Potsdam,
Germany. 4IZA Institute of Labor Economics, Bonn, Germany.
sessments that can only draw on descriptive at which point in time, nor did we restrict the
5
Climate Econometrics, Nuffield College, University of reviews of selected studies and instruments number of potential interventions. Allowing
Oxford, Oxford, UK. 6Smith School of Enterprise and the (17) rather than systematic evidence for the for any country to be potentially treated at any
Environment, University of Oxford, Oxford, UK. 7Institute for
entire spectrum of diverse climate policy instru- point in time permits us to identify large re-
New Economic Thinking, University of Oxford, Oxford, UK.
8
Organisation for Economic Co-operation and Development ments at the global scale (18). ductions and reduces concerns around omit-
(OECD), Paris, France. 9Department of Economics, University Here, we provide a global, data-driven causal ting potentially influential policy interventions.
of Victoria, Victoria, BC, Canada. 10Faculty of Economics and impact assessment to identify effective policies Second, we estimated the effect size for the
Management, Technische Universität Berlin, Berlin, Germany.
*Corresponding author. Email: stechemesser@pik-potsdam.de that have led to large emission reductions out agnostically detected country-specific interven-
(A.S.); koch@mcc-berlin.net (N.K.) of a universe of about 1500 climate policy mea- tions using two popular estimators from the

Stechemesser et al., Science 385, 884–892 (2024) 23 August 2024 1 of 9

RES EARCH | R E S E A R C H A R T I C L E

A
tightenings per country
Average number of 8
adopted policies & Buildings
6 Electricity
4 Industry
Transport
2

0
2000 2005 2010 2015 2020
B 80

Performance standard (electric motors)

tightenings by policy instrument

Performance standard (cars & trucks)

Number of adopted policies and

Performance standard (appliances)

Ban & phase out (fossil heatings)

Renewable expansion planning

Ban & phase out (coal plants)

60

Renewable portfolio standard

Ban & phase out (fossil cars)

Energy efficiency mandate
Fossil fuel subsidy reform

Fossil fuel subsidy reform

Fossil fuel subsidy reform

Fossil fuel subsidy reform

Public expenditure for rail
Emission trading scheme

Emission trading scheme

Emission trading scheme

Emission trading scheme

Renewable feed in tariff
Financing mechanism

Financing mechanism
Air pollution standard
40 Renewable auction
Label (appliances)

Vehicle tax (cars)

Adoption subsidy

Adoption subsidy
Building code

Label (cars)

Downloaded from https://www.science.org at Loughborough University on September 24, 2024

Speed limit
Carbon tax

Carbon tax

Carbon tax

Carbon tax
20

Road toll
Fuel tax

Fuel tax

Fuel tax

Fuel tax
0

Developed Australia, Austria, Belgium, Bulgaria, Canada, Czech Republic, Denmark, Finland, France, Developing/Transitioning Argentina, Brazil, Chile, China, Colombia,
Germany, Greece, Hungary, Ireland, Italy, Japan, Netherlands, New Zealand, Norway, Poland, Portugal, India, Indonesia, Mexico, Peru, Russia, Saudi Arabia,
economies Romania, Slovak Republic, Spain, Sweden, Switzerland, United Kingdom, United States economies South Africa, South Korea, Turkey

C Americas Europe and Middle East

FIN
NOR
SWE
CAN
RUS
GBR DNK

POL
NLD
IRL DEU
USA CZE SVK
BEL

FRA AUT HUN ROU

CHE

ITA BGR
PRT ESP TUR

MEX
GRC

COL
Number of detected breaks in SAU

CO2 emissions between 2000

and 2020 by sector. Asia and Oceania
PER
Country Buildings JPN
4+
3 Electricity BRA CHN
2 Industry KOR
1 Transport
Two breaks No breaks
in electricity sector in industry sector IND
South Africa
CHL

IDN

ARG
ZAF
4+
3
AUS
2 NZL
1

Fig. 1.

Stechemesser et al., Science 385, 884–892 (2024) 23 August 2024 2 of 9

RES EARCH | R E S E A R C H A R T I C L E

Fig. 1. Increase in climate policy and detected 69 success cases with large emission reductions across sectors and countries. (A) Increase in average
number of adopted policies and policy tightenings per country between 1998 and 2022. A policy tightening is a substantial increase in stringency of an already existing
policy instrument (SM section 3). (B) Visualization of the number of adopted policies and policy tightenings across the whole timeframe for each instrument
type, highlighting the diverse number of instruments used. Developed and developing or transitioning economies are considered separately. (C) Number of large
emission reductions for each country and sector as detected with the break detection DID analysis. Overall, we found 69 breaks.

causal inference literature that address differ- country-specific interventions without prior the TWFE DID model to estimate effect sizes
ent identification concerns. In the last step of knowledge of their occurrence [supplemen- of emission breaks conditional on their detec-
our methodology, policy attribution, we com- tary materials (SM), materials and methods]. tion (SM section 8).
bined these estimates with our comprehensive The realized emission breaks are unevenly dis-
policy data from the OECD to draw systematic tributed across sectors and countries (Fig. 1C). Association of breaks with known and
inference on the (differential) effectiveness of Most breaks occur in the buildings sector unknown policies
single policies and various policy mixes. The com- (24 cases), followed by transport (19 cases), in- The timing of the identified structural breaks
bination of conservatively controlling for the risk dustry (16 cases), and electricity (10 cases). A matches well with newly adopted or tightened
of spuriously identifying false positives, the use total of 48 and 21 breaks are identified in climate policies, which are visualized as squares
of control groups in a panel setting, and a cross developed and developing or transitioning eco- along the time axes of Figs. 2 and 3. Of 69

Downloaded from https://www.science.org at Loughborough University on September 24, 2024

validation with popular synthetic control meth- nomies, respectively. breaks, 63 are associated with at least one
ods and alternative selection algorithms give this The reductions in emissions for each of the policy adoption or tightening within a 2-year
reverse causal approach (20) credibility. Our policy 69 breaks estimated with our preferred two- interval around the time of the break allowing
data are publicly available to spur further re- way fixed effects (TWFE) DID estimator (SM for lagged or anticipatory policy effects (details
search on ex post climate policy evaluation. Our materials and methods) are shown in Figs. 2 on the policy attribution are available in the
statistical methodology is easily reproducible and and 3. They provide evidence that the de- SM materials and methods). Of the matched
allows for constant updating as new experiences tected breaks are credible and align with breaks, four are associated with the hetero-
with climate policies are gained (21). the timing of the adoption or tightening of geneous effect of the introduction or tight-
meaningful climate policies. Visual compar- ening of a European Union (EU) policy that
Results ison between the time series for observed we control for (SM materials and methods).
Increase in climate policies with diverse policy emissions (Figs. 2 and 3, black lines) and Most policies associated with our 21 breaks in
mixes over time predicted emissions (Figs. 2 and 3, blue lines) developing economies have been rarely studied
The OECD policy data reveal a consistent in- across countries and sectors suggests a good in the literature and highlight the benefit of
crease in the number of implemented climate fit of our model for the log of CO2 emissions our approach to detect hitherto insufficiently
policies across all sectors between 1998 and as a function of socioeconomic developments studied or unknown effective policy interven-
2022 (Fig. 1A). By 2022, the average number [gross domestic product (GDP) and popula- tions that require more research. We show in
of policy adoptions and tightenings ranged tion], weather (cooling and heating days), and fig. S45 how often each policy instrument coin-
between four and eight policies per country. country-specific time trends. The comparison cides with a detected large emission reduction.
At the same time, there is substantial variation with the counterfactual emissions (Figs. 2 and Most breaks are matched to two or more
in the types of policy instruments used across 3, red lines), which would have occurred in the policies (70%). Before describing our system-
sectors and countries (Fig. 1B). With 270 cases, absence of the detected breaks and are de- atic assessment of differential effects of policy
command-and-control measures such as emis- rived from the control group in our DID set- mixes and stand-alone policies, we discuss
sion standards and technology mandates are ting, suggests strong country-level breaks in here one prominent example for each country
the most frequently used policies in all sectors emissions. Our approach targets large emis- group that illustrates the power of our ap-
except transport. Market-based policies are sion reductions (which require a minimum proach to identify both (i) known, headline
primarily concentrated in developed econo- effect size ranging from 4.5 to 13%) (SM sec- policies for which some evidence already exists
mies and most prevalent in the transport sec- tion 6.1), and the average effect size across the and (ii) previously unknown combinations of
tor. Among market-based policies, subsidies detected large breaks is 19.4% (22.7% in build- effective policies.
are popular, whereas carbon pricing (carbon ings, 26% in electricity, 18.4% in industry, In the electricity sector, we detected two
taxes and emission trading schemes) remains and 12.6% in transport). We report in tables adjacent breaks for the United Kingdom in
limited, with a total of 116 cases (88 in devel- S12 to S19 point estimates and standard er- 2015 and 2016. These follow the mid-2013 in-
oped economies). rors for the country-level breaks in emissions. troduction of a carbon price floor that imposed
On the basis of an approximate 95% confi- a minimum price for UK power producers in
Detecting structural breaks in sectoral dence interval (CI) of these estimates, we cal- the EU emission trading system and has been
emissions across the globe culated equivalent total emission reductions shown to have reduced emissions considerably
Although we observed around 1500 policy between 0.6 and 1.8 Gt CO2 (SM section 10). (13, 22–24). Although the existing literature has
adoptions and tightenings in our policy data, To ensure the robustness of our DID model attributed most of this effect to the carbon price
their impact on emissions has so far been against misspecification, which might lead to floor, our attribution method, combined with
highly uncertain. Our break detection DID the detection of spurious breaks, we show in the OECD policy database, reveals that the car-
analysis suggests that large emission reductions the SM that our results are robust to (i) the bon price floor was part of a wide policy mix
have materialized in only 69 cases (Fig. 1C). We selection algorithm used for break detection that included command-and-control measures
identified these successful cases with a machine (SM section 7.2), (ii) alternative model speci- (renewable portfolio standards, renewable ex-
learning–based, data-driven search for struc- fications (SM section 7.3), (iii) omitted varia- pansion planning, stricter air pollution stan-
tural breaks in sector-specific CO2 emissions bles (SM section 7.4), and (iv) the country dards, and the announcement of a phase-out
relative to a control group, separately for de- sample composition (SM section 7.5). We ob- of coal power plants) and other market-
veloped and developing or transitioning eco- tained very similar results when we used gen- based incentives (renewable feed-in tariff and
nomies. The detected breaks identified large eralized synthetic control methods rather than auctions).

Stechemesser et al., Science 385, 884–892 (2024) 23 August 2024 3 of 9

RES EARCH | R E S E A R C H A R T I C L E

Estimated
Counterfactual
Estim.
Effect Size

Detected Break

Model Fit Attributed Policies

Policy Categories

Downloaded from https://www.science.org at Loughborough University on September 24, 2024

Policy Categories

Fig. 2. Most emission breaks in the electricity and industry sectors can be policy intervention is symbolized by a colored square along the x axis. Policy
associated with policy mixes. For each country and sector, the black line interventions include both newly adopted policies and tightened policies. The
indicates the observed emissions over time, and the blue line indicates the model color of the boxes indicates the policy instrument. If a box falls into the 2-year
fit, which follows the true emissions closely. Detected emission breaks are CI around detected break dates, we attribute the given policy to an emission
indicated with vertical red lines, and counterfactual emissions are indicated in break. Adoptions and tightenings of EU labels, EU performance standards, and
red. Each break is surrounded by a statistical CI (dark gray) and a 2-year CI the EU emission trading scheme are indicated with symbols (tag, gear, and euro
(light gray), which in some cases overlap. The 2-year CI captures both the icons, respectively). For electricity and industry, 67 and 54% of the matched
statistical uncertainty as well as leads and lags in the policy response. Each breaks, respectively, are associated with a policy mix.

Stechemesser et al., Science 385, 884–892 (2024) 23 August 2024 4 of 9

RES EARCH | R E S E A R C H A R T I C L E

Estimated
Counterfactual

Observed CO2
Emissions Model Fit

Detected Break
Attributed Policies

Implemented Policies

Downloaded from https://www.science.org at Loughborough University on September 24, 2024

Policy
Categories

Policy
Categories

Fig. 3. Most emission breaks in the buildings and transport sectors can be indicates the observed emissions over time, and the blue line indicates the model fit,
associated with policy mixes. For each country and sector, the black line which follows the true emissions closely. Detected emission breaks are indicated

Stechemesser et al., Science 385, 884–892 (2024) 23 August 2024 5 of 9

RES EARCH | R E S E A R C H A R T I C L E

with vertical red lines, and counterfactual emissions are also indicated in red. The color indicates the policy instrument. If a box falls into the 2-year CI around
Each break is surrounded by a statistical CI (dark gray) and a 2-year CI detected break dates, we attribute the given policy to an emission break.
(light gray), which in some cases overlap. The 2-year CI captures both the Adoptions and tightenings of EU labels, EU performance standards, and the
statistical uncertainty as well as leads and lags in the policy response. EU emission trading scheme are indicated with symbols (tag, gear, and euro icons,
Each policy intervention is indicated with a colored square along the x axis. respectively). For buildings and transport, 60 and 94% of the matched breaks,
Policy interventions include both newly adopted policies and tightened policies. respectively, are associated with a policy mix.

In the industry sector, the break in China in the most widely used regulatory instruments subsidies. Instead, regulation is the most effec-
2016 occurs with some lag after the launch of and subsidy schemes may require comple- tive stand-alone policy (33%), but pricing is an
seven pilot emission trading schemes begin- mentary instruments to enable substantial equally important element of effective policy
ning in 2013. Again, prior literature has shown emission reductions. The effect sizes of policy mixes because 50% out of all successful policy
that the carbon price reduced emissions (23, 25), mixes that combine these non–price-based in- mixes include pricing. In the industry sector,
but these studies frequently do not consider struments with taxation or reduced fossil fuel pricing plays a prominent role. It is most ef-
the role of the simultaneous reduction of fossil subsidies (Fig. 4A, thick black lines) suggest fective individually in developed economies
fuel subsidies in 2016 and the strengthening that in most cases pricing is the complement (43%) and shows the most synergy with other
of financing mechanisms for energy efficiency that enables effective emission reductions. For policies in developing economies (50%). How-

Downloaded from https://www.science.org at Loughborough University on September 24, 2024

investments in 2015. example, in the electricity sector all mixes that ever, subsidies can be effective complements
were associated with large emission reduc- in both contexts. In buildings, across coun-
Identification of effective policy mixes tions have pricing elements (fig. S46). However, tries, our findings suggest that a broad set of
The major benefit of conducting policy eval- we cannot rule out whether a single policy instruments can be similarly powerful, but sub-
uation for many policy instruments in one dominates the combined effect of a mix or sidies slightly dominate (individually and in
integrated approach is that we can system- whether any policy instrument in the mix is combinations) in developed economies and
atically compare the effectiveness of these weakening the overall effect. In this regard, regulations in developing economies.
policy instruments both when implemented we offer in Fig. 4A a starting point toward
individually and when used as part of a mix. understanding interaction effects among Discussion
To this end, we aggregated instruments with policy instruments, but empirical challenges Identifying effective policies is crucial to guide
equivalent economic mechanisms (such as remain. policy-makers in designing the most mean-
emission trading and carbon and fuel taxes Next, we assessed which specific combinations ingful interventions. We sought to narrow the
to “taxation”) (SM section 9.1) and in Fig. 4A of policy instruments caused large emission policy evaluation gap by applying a machine
compared the average effect sizes of emis- breaks. To this end, we categorized instruments learning–based extension of the DID approach
sion breaks by policy instrument for the case into four groups of policy types (Fig. 4A, x axis, to identify effective policies that have led to
of a stand-alone implementation and an im- and SM section 9.1), whose role is at the center strong emission reductions. We considered
plementation in a policy mix. For non–price- of high-level political discussions about opti- the universe of about 1500 observed policies
based policies, we also show the average effect mal policy design: information, pricing, regu- documented in a comprehensive, high-quality,
sizes of a mix with pricing. lation, and subsidy. We show in Fig. 4B how OECD climate policy database. Across four sec-
In most cases, we found that effect sizes are often each distinct policy combination or single tors, 41 countries, and 2 decades, we found
larger if a policy instrument is part of a mix policy is associated with a break as a percent- 63 successful policy interventions with large
rather than implemented alone. Some policies— age share of all successful interventions (an effects that reduced total emissions between
for example, labels and fossil fuel subsidy alternative pairwise clustering approach that 0.6 and 1.8 Gt CO2.
reforms—are only ever associated with large uses less aggregated policy categories is pro- Our results inform contentious policy de-
emission breaks in a mix, which suggests that vided in SM section 9.5). bates in three main ways. First, we contribute
these types of policy intervention are either Our findings suggest that the combinations empirical evidence for the effectiveness of pol-
never implemented as a stand-alone policy of policy instruments that are complementary icy mixes. Although the assertion is wide-
or do not cause major emission reductions by vary across sectors and country groups. Trans- spread that policy mixes can be beneficial and
themselves. Several popular instruments—such port is generally the sector with the most po- unfold positive synergies (17), controversies
as bans, building codes, energy efficiency man- tential for complementarities. However, the remain because critics argue that policy mixes
dates, and subsidies—are either also only ever dominant sectoral policy differs across coun- may instead be subject to overlapping instru-
detected in policy mixes or have smaller aver- try groups. In developed economies, pricing ments and perform no better than a single
age effect sizes if they are associated as stand- stands out individually, with 20% out of all instrument, which is partly backed by theoret-
alone policy with emission breaks. For example, successful detected interventions being asso- ical studies in economics (26, 27). Even though
the mean effect size of all breaks associated ciated with pricing individually. Yet subsidies we cannot generally rule out such negative
with ban and phase-out policies implemented are the most complementary instrument, es- synergies owing to empirical constraints, we
in the buildings sector as part of a policy mix is pecially in combination with pricing (33%). By identified a number of policy instruments for
around −32%, whereas it is only around −13% contrast, in developing economies regulation which the empirical evidence suggests comple-
if ban and phase-out policies were stand-alone is the most powerful policy. It is highly effec- mentary effects. These include popular sub-
policies. By contrast, taxation is a notable ex- tive as an individual policy (33%) but also in sidy schemes and regulatory instruments such
ception in effectively causing large emission combination with the duo of subsidies and as bans, building codes, energy efficiency man-
breaks alone. It stands out as the only policy pricing (33%) and information (33%). The elec- dates, and labels, for which we found larger
instrument that achieves near equal or larger tricity sector of developing economies shows reduction effects in policy mixes as compared
effect size as a stand-alone policy across all no detected complementarities. In this study, with the case of a stand-alone implementa-
sectors. subsidies alone are the most powerful policy tion. This suggests that some of these most
Overall, these comparisons of relative effect tool (67%). By contrast, in developed econo- widely used policy instruments are comple-
sizes provide suggestive evidence that some of mies we never found successful stand-alone mentary or even reinforcing in policy mixes,

Stechemesser et al., Science 385, 884–892 (2024) 23 August 2024 6 of 9

RES EARCH | R E S E A R C H A R T I C L E

A 0
Avg. Effect Size
Mix Single Policy
−10
−20
−30
−40 Not detected
Policy Mix Fossil fuel subsidy reform Taxation Adoption subsidy Financing mechanism Ban & phase out Building code Performance standard Label
Single Policy Pricing Subsidy Regulation Information
Avg. Effect Size
of Mix with Pricing
0
−10 Mix with Pricing
−20 Not detected
Average effect size (%)

−30
−40
Policy Mix Fossil fuel subsidy reform Taxation Renewable subsidy Air pollution standard Ban & phase out Renewable planning Renewable portfolio standard
Single Policy
Pricing Subsidy Regulation
0
−10
−20
−30
−40
Policy Mix Fossil fuel subsidy reform Taxation Financing mechanism Energy efficiency mandate Performance standard
Single Policy Pricing Subsidy Regulation
0
−10
−20
−30
−40

Downloaded from https://www.science.org at Loughborough University on September 24, 2024

Fossil fuel subsidy reformTaxation (driving) Taxation (vehicle) Adoption subsidy Public expenditure Ban & phase out Performance standard Label
Policy Mix for rail
Single Policy
Pricing Subsidy Regulation Information

B
Developed economies Developed economies
Subsidy
27.3%

Subsidy Regulation
Pricing
33.3% 33.3%
18.2% of detected 16.7%
18.2% 18.2% Regulation
successful interventions
were policy mixes 27.3%
27.3% of detected
of subsidies and pricing Pricing successful interventions 16.7%
9.1% were regulation alone

Developing economies Developing economies

Pricing only detected as successful
intervention in isolation

Information Subsidy Subsidy

28.6% 14.3% 66.7%
Regulation14.3% Pricing Regulation
28.6% 33.3%
14.3%

Developed economies Developed economies

13.3% Regulation
Subsidy
13.3% 13.3%
Pricing
Pricing Regulation Subsidy
Information 20%
42.9% 14.3% 14.3% 28.6%
6.7%
33.3%

Developing economies Developing economies

Pricing Regulation
33.3%

Regulation Subsidy
Subsidy
25% 25% 25% 25% Information
33.3% 33.3% Pricing

Fig. 4. Effective policies and policy mixes. (A) On the basis of point estimates for combinations of policy types [definitions of categories are provided in (A), x axis] are
country-specific breaks in emissions (tables S12 to S19), we compared the average effective in each sector separately for developed and developing economies. For
effect sizes of all breaks in which a policy instrument appears individually with each circle area, the percentage indicates which share of successful interventions in
that of all breaks in which this policy instrument appears in a mix. For non–price- this sector was made up by a specific individual policy type or a specific combination
based policies, the black thick line also indicates the average effect size of a mix of policy types. An individual policy type encompasses breaks that match a single
with a given policy instrument and pricing (through taxation or reduced fossil policy instrument (for example, one subsidy scheme) or a combination of policy
fuel subsidies). (B) Euler diagrams (SM materials and methods) show which instruments of the same type (for example, two or more different subsidy schemes).

Stechemesser et al., Science 385, 884–892 (2024) 23 August 2024 7 of 9

RES EARCH | R E S E A R C H A R T I C L E

which is in line with the theoretical under- Our approach identifies country-specific pol- 3. UN Envrironment Programme (UNEP), Emissions Gap Report
standing that these specific instruments alone icy interventions that have led to large emission 2022, (2022); https://www.unep.org/resources/emissions-
gap-report-2022 (accessed July 2023).
often have a limited scope (for example, only reductions. Thus, we are unable to quantify 4. T. Fransen et al., Nat. Clim. Chang. 13, 752–755 (2023).
new cars or new appliances) and are subject the effect of policies with minor effects (a dis- 5. T. Sterner et al., Nat. Sustain. 2, 14–21 (2019).
to rebound effects (28). Additional instruments cussion on minimum effect size is available in 6. D. Rosenbloom, J. Markard, F. W. Geels, L. Fuenfschilling,
Proc. Natl. Acad. Sci. U.S.A. 117, 8664–8668 (2020).
such as pricing can effectively address both SM section 6.1). However, regarding the size of 7. O. Blanchard, C. Gollier, J. Tirole, Annu. Rev. Econ. 15, 689–722
factors and thus generate positive synergy the emissions gap and the commitment to the (2023).
(15, 29). Further explanations for the comple- Paris Agreement, it is arguably most impor- 8. S. M. Eskander, S. Fankhauser, Nat. Clim. Chang. 10, 750–756
(2020).
mentarities include that policy mixes can ad- tant to identify combinations of policy instru- 9. G. Martin, E. Saikawa, Nat. Clim. Chang. 7, 912–919 (2017).
dress a multitude of market failures (7) and ments that have large effects. In addition to 10. C. Le Quéré et al., Nat. Clim. Chang. 9, 213–217 (2019).
may be more successful in increasing the over- the detected effects of national climate poli- 11. P. Bayer, M. Aklin, Proc. Natl. Acad. Sci. U.S.A. 117, 8804–8812
(2020).
all policy stringency (30) and maximizing pol- cies, there might also exist impacts of inter-
12. J. J. Andersson, Am. Econ. J. Econ. Policy 11, 1–30 (2019).
icy credibility, which shapes the expectations national or regional policies that are only 13. M. Leroutier, J. Environ. Econ. Manage. 111, 102580 (2022).
of consumers and investors (31). controlled for in this analysis. Furthermore, 14. M. Linsenmeier, A. Mohommad, G. Schwerhoff, Nat. Clim. Chang.
Second, our findings highlight that success- even though we are able to provide an analysis 12, 1107–1110 (2022).
15. J. van den Bergh et al., Clim. Policy 21, 745–764 (2021).
ful policy mixes vary across sectors and that for 41 countries on 6 continents, our analysis 16. N. Koch, L. Naumann, F. Pretis, N. Ritter, M. Schwarz,
policy-makers should focus on sector-specific still suffers from regional imbalance. This is Nat. Energy 7, 844–853 (2022).

Downloaded from https://www.science.org at Loughborough University on September 24, 2024

best practices when designing climate policy largely caused by a sparsity of data for devel- 17. J. Hoppe, B. Hinder, R. Rafaty, A. Patt, M. Grubb, Annu. Rev.
Environ. Resour. 48, 615–650 (2023).
rather than following a one-size-fits-all ap- oping countries, especially in Africa and Asia. 18. J. C. Minx, M. Callaghan, W. F. Lamb, J. Garard, O. Edenhofer,
proach. In line with theoretical expectations, Last, the ex post analysis identifies past cases Environ. Sci. Policy 77, 252–259 (2017).
we have identified pricing as a particularly of successful policy interventions (2000 to 19. M. Crippa et al., CO2 emissions of all world countries—2022
report (2022); https://publications.jrc.ec.europa.eu/
effective policy in those sectors dominated by 2020). If policy-makers focused more on the repository/bitstream/JRC130363/co2_emissions_of_all_
profit-maximizing firms—namely, industry— identified best practices, we can plausibly ex- world_countries_online_final_2.pdf.
but also the electricity sector in developed eco- pect more substantial emission reductions in 20. A. Gelman, G. Imbens, “Why ask Why? Forward Causal
Inference and Reverse Causal Questions”(working paper 19614,
nomies. By contrast, for the building and partly the future. However, we also acknowledge that National Bureau of Economic Research, 2013); https://www.
also the transport sectors, which both include the workings of policies can be context spe- nber.org/papers/w19614.
a large share of private consumers subject to cific. Therefore, we caution against general- 21. J. Elliott et al., Nature 600, 383–385 (2021).
22. J. Abrell, M. Kosch, S. Rausch, J. Environ. Econ. Manage. 112,
documented behavioral factors such as myo- izing our country-specific effect estimates to
102589 (2022).
pia (32, 33), we found most potential for com- average treatments effects for particular policy 23. N. Döbbeling-Hildebrandt et al., Nat. Commun. 15, 4147 (2024).
plementarities between policy instruments. instruments or mixes. In addition, a better 24. J. L. Castle, D. F. Hendry, Found. Trends Econom. 10, 145–322
This is consistent with the hypothesis that understanding of the optimal design of cli- (2020).
25. J. Cui, C. Wang, J. Zhang, Y. Zheng, Proc. Natl. Acad. Sci. U.S.A.
broad incentives with a particular focus on mate policy mixes in a sound welfare frame- 118, e2109912118 (2021).
adoption decisions (such as renewal of heat- work that accounts for additional dimensions 26. S. Fankhauser, C. Hepburn, J. Park, Clim. Change Econ.
ing systems or cars) are needed in these sec- than environmental effectiveness remains a (Singap.) 1, 209–225 (2010).
27. G. Perino, R. A. Ritz, A. Van Benthem, “Overlapping climate
tors (34). major research need to guide policy-making. policies,” working paper 25643 (National Bureau of Economic
Third, our results stress that effective pol- Our results provide a clear yet sobering Research, 2019); https://www.nber.org/papers/w25643.
icies vary with economic development. For ex- perspective on the policy effort necessary for 28. K. Gillingham, M. J. Kotchen, D. S. Rapson, G. Wagner, Nature
493, 475–476 (2013).
ample, in sharp contrast to that of developed closing the remaining emissions gap of 23 Gt 29. E. Dimanchev, C. R. Knittel, Energy Econ. 122, 106697 (2023).
economies, we did not find any successful CO2-eq by 2030 (3). Using the average (or highest) 30. J. Meckling, N. Kelsey, E. Biber, J. Zysman, Science 349,
pricing intervention with large emission re- effect sizes of the detected breaks, we com- 1170–1171 (2015).
31. G. Dolphin, M. Pahle, D. Burtraw, M. Kosch, Nat. Clim. Chang.
ductions in the electricity sector of develop- puted a hypothetical scenario in which all 13, 1033 (2023).
ing economies, even though around 13% of 41 countries in our sample achieve emission 32. K. T. Gillingham, S. Houde, A. A. van Benthem, Am. Econ.
policy adoptions or tightenings are pricing reductions the size of the average (as well J. Econ. Policy 13, 207–238 (2021).
interventions. This finding is consistent with as highest) detected sectoral effect size once 33. E. Myers, Am. Econ. J. Econ. Policy 11, 165–188 (2019).
34. S. T. Anderson, J. M. Sallee, Annu. Rev. Resour. Econ. 8,
claims that the lack of liberalized markets and before 2030 (SM section 10). We estimate that 157–180 (2016).
existence of other price distortions can limit this would close the emissions gap by 26% (or 35. F. Teng, F. Jotzo, X. Wang, Interactions between market reform
the effectiveness of price-based instruments by 41% for the highest effects). Thus, scaling and a carbon price in China’s power sector, Econ. Energ.
Environment. Pol. 6, 39–54 (2017).
(35). It is also in line with the theory of policy up good-practice policies identified in this study
36. J. Meckling, T. Sterner, G. Wagner, Nat. Energy 2, 918–922
sequencing, which states that in a first stage to each sector of other parts of the world can in (2017).
of climate policy-making, regulations and the short term be a powerful climate mitigation 37. M. Jakob, J. C. Steckel, The Political Economy of Coal:
subsidies are effective in building economic strategy. However, even if all countries in our Obstacles to Clean Energy Transitions (Taylor & Francis, 2022).
38. S. J. Davis et al., Science 360, eaas9793 (2018).
interest in green technology and reducing sample were able to replicate past success, more 39. OECD, Climate Actions and Policies Measurement Framework
the cost of technologies (36). In this respect, than four times (one and a half times) the effort (CAPMF) (2023); http://oe.cd/dx/5if.
the observed differences in effective policies witnessed so far would have to be exerted to 40. A. Stechemesser, Replication code and materials for “Climate
policies that achieved major emission reductions: Global
may partly reflect the climate policy stage. close the emissions gap. This also highlights
evidence from two decades”. Zenodo (2023); https://zenodo.
We may not observe some policies because the need for research providing systematic evi- org/records/12773811.
they are not implemented owing to interest dence on which climate policy mixes are most
AC KNOWLED GME NTS
group opposition or limited state capacity in powerful in spurring the necessary deploy-
We thank C. Bertram, F. Creutzig, G. Dolphin, O. Edenhofer,
developing countries (37). However, for inter- ment and development of low-carbon tech-
D. Hendry, C. Hepburn, A. B. Martinez, M. Pahle, R. Rafaty, and
pretation it is important to consider that our nologies for a future net-zero economy (38). participants of the Berlin Workshop on Machine Learning in
analysis for developing countries is based Economics, the pre-conference event of the European Association
on a small set of estimates given the lim- of Environmental and Resource Economists; the Conference on
RE FERENCES AND NOTES Econometric Models of Climate Change; the EcoTransit Workshop
ited number of policies detected, particularly 1. H. D. Matthews, S. Wynes, Science 376, 1404–1409 (2022). at the University of Siena; the Workshop on Mapping Effects of
in the electricity sector. 2. M. Roelfsema et al., Nat. Commun. 11, 2096 (2020). Environmental Policy at the University of Hamburg; and the WIP

Stechemesser et al., Science 385, 884–892 (2024) 23 August 2024 8 of 9

RES EARCH | R E S E A R C H A R T I C L E

Seminar of OECD for valuable comments on a draft of this and N.K. wrote the manuscript, with contributions from all authors. 12773811 (40). All results can be explored in more detail in an
paper. Funding: A.S., P.K., H.V., E.D., L.M., and A.W. acknowledge Competing interests: The authors declare that they have no interactive web dashboard available at http://climate-policy-
the financial support by the Werner Siemens Foundation. F.P. competing interests. M.S. is employed by the Austrian Ministry of explorer.pik-potsdam.de. License information: Copyright © 2024
acknowledges support from the Robertson Foundation and the Finance. The views expressed here are those of the authors and the authors, some rights reserved; exclusive licensee American
SSHRC. E.M. gratefully acknowledges funding from the Robertson not those of the Ministry of Finance or the Austrian government. Association for the Advancement of Science. No claim to original
Foundation. M.S. acknowledges support from the Clarendon M.S. was paid for consulting services for a government client US government works. https://www.science.org/about/science-
Foundation and from the Einstein Stiftung Berlin. Author on implementing carbon taxes in 2021. D.N. is employed by licenses-journal-article-reuse
contributions: N.K., F.P., M.S., E.M., P.K., and A.S. designed the OECD. The views expressed here are those of the authors and not
analysis. D.N. contributed the Climate Actions and Policies those of OECD or its member countries. Data and materials
SUPPLEMENTARY MATERIALS
Measurement Framework (CAPMF) data. E.M., E.D., L.M., and availability: The CAPMF data are publicly available from the
A.W. collected further data. F.P. and M.S. wrote the getspanel OECD Data Explorer (https://oe.cd/dx/capmf) (39). Data for the science.org/doi/10.1126/science.adl6547
package for the break detection method. E.M. and A.S. wrote the United States are not publicly available from the OECD Data Materials and Methods
program code for the analysis and ran the break detection models, Explorer. They were downloaded in March 2023 from the IPAC Supplementary Text
with support from N.K., F.P., M.S., and P.K.; A.S. processed the Dashboard from https://www.oecd.org/climate-action/ipac/ Figs. S1 to S47
policy data, with contributions from P.K., H.V., E.D., L.M., and dashboard and from the data visualization tool from https:// Tables S1 to S43
A.W. and analyzed it with support from P.K. and H.V. Main oecd-main.shinyapps.io/climate-actions-and-policies. The EDGAR References (41–79)
figures were designed by all authors and realized by A.S.; P.K. emissions data are available from https://data.jrc.ec.europa.eu/
conducted the supplementary analysis using the synthetic control collection/EDGAR. The code and data to replicate the study is Submitted 1 November 2023; accepted 19 July 2024
method. All authors discussed and interpreted the results. A.S. publicly available from Zenodo under https://zenodo.org/records/ 10.1126/science.adl6547

Downloaded from https://www.science.org at Loughborough University on September 24, 2024

Stechemesser et al., Science 385, 884–892 (2024) 23 August 2024 9 of 9