Environmental Policy Instruments

The instrument toolbox

Inventory of instruments

What: Three broad families.

• Law / command-and-control: bans, minimum-emission thresholds, appliance standards (fridges), car emission standards. The law also prepares the groundwork for everything else.
• Market-based (price-based): taxes; subsidies (the mirror image — make desirable behaviour cheaper); ETS (Emissions Trading System), which sets a quantity limit that translates into a price by creating a new market.
• Other: feed-in tariffs (big in renewables), infrastructure/urban planning (public transport), green R&D (research and development), and informational/behavioural tools — education, information campaigns, nudges.

The bad-news framing: Any environmental policy will create costs (higher consumption-good prices; households also bear costs as direct/indirect owners of firms). We do it only if benefits outweigh costs, and we want to minimise the cost — the least-cost principle.

Efficiency: the equimarginal principle

Equimarginal / least-cost principle

What: Cost-effective abatement requires marginal abatement costs (MACs) equal across all emitters and equal to marginal damages.

$$\text{MAC}_i = \text{MAC}_j = p^{\text{CO}_2} = \text{marginal damages}$$

Why: For a ton of CO$_2$, the damage is the same everywhere (it mixes globally), so the marginal cost of emitting should be equalised across the globe — ideally a single global carbon price covering all emissions in all countries.

Remember: In practice this is "close to unimaginable" — some countries don't integrate into the international order, multilateralism is under strain, and it is extremely hard to target all emissions (capturing every cow's methane is impractical). So price-based mechanisms usually only target industrialised processes.

Bans / standards vs. ETS

The comparison

What: An ETS sets an absolute cap (emissions ≤ X/year) — formally equivalent to banning emissions beyond X, so it has a ban-like feature. But:

Why: Bans set a binding constraint on each individual agent, so MACs differ across agents → beneficial trades are left on the table → not cost-effective. An ETS creates a market so permits flow to whoever abates most cheaply → same environmental cap, lower total cost.

Remember — two extra ETS advantages:

1. Negative-emissions room: even if you want a broad category down to zero, an ETS lets negative-emission technologies operate (with a price); a flat ban can't reward them.
2. Revenue: bans only produce costs; an ETS can auction permits → government revenue, which can be recycled to households/firms (Chapter 11).

If you want a single specific technology literally to zero, there's little difference between a ban and an ETS. ETS only pays off for broad enough categories, and where transaction costs of creating a new market aren't too high relative to damages.

ETS vs. taxes — Weitzman and the stock/flow point

Both aim to put one price on CO$_2$

• Taxes: easier to introduce administratively (no new market). But often a negative political connotation; sometimes legally impossible. Uncertainty about the quantity of emission reductions.
• ETS: designed to hit a quantity target. But uncertainty about cost — CO$_2$ prices can be volatile, complicating firms' investment planning.

Weitzman (1974)

What: Studied which instrument dominates under uncertainty. Argued taxes dominate ETS for climate because the slope of marginal damages is likely smaller than the slope of marginal abatement cost. We dislike uncertainty more along the steeper curve.

Remember: Weitzman cast it as a flow pollutant — which is not the right way to think about climate.

Karp & Traeger (2024)

What: ~50 years later, show Weitzman's result holds for flow pollutants (emissions flow each year, don't accumulate). But for a stock pollutant like atmospheric CO$_2$ (cumulative emissions matter), ETS is likely superior.

Remember: Flow → tax; stock (climate) → ETS. This flips Weitzman's headline for the climate case.

Germany 2021 "carbon price" was secretly an ETS: Germany's 2021 transport/heating carbon-pricing scheme looked like a tax but was legally an ETS with one fixed price (a "joke" form, done because a tax was not legally possible). It now moves toward a price collar.

World Bank Carbon Pricing Dashboard: Share of global CO$_2$ emissions under any pricing instrument was tiny and only red (carbon taxes — the Scandinavian pioneers in the early 90s). Picture changes dramatically in 2005 when the EU ETS launched. China's ETS later pushes coverage to ~25% of global emissions.

Hybrid instruments

Price collar / smart cap

• Price collar: an ETS with a flexible minimum and maximum permit price; at those bounds the government steps in (withdraws or issues certificates). Germany's national ETS (transport/heating) moves to a collar; the UK uses a carbon price floor.
• Smart cap / flex cap (Traeger et al. 2020): even better from a welfare perspective. The government doesn't know firms' MACs and firms have no incentive to reveal them, so it learns from market behaviour and dynamically adjusts the collar to cut the welfare cost of asymmetric information on abatement costs.

Remember: The smart cap is dynamic learning about firms' abatement costs (not price discrimination — everyone still pays the same price). It eliminates welfare cost from asymmetric information.

Policy ramp

Remember: In practice governments start with a lower price and ramp it up. Whether future governments actually ramp it is open. Public opposition often forces a cut or stop to increases — most famously France's yellow vests (gilets jaunes).

Should we rely solely on emissions pricing? No.

Why a mix beats a single instrument

Why: Multiple, overlapping externalities (climate and local pollution) plus additional market failures — public-goods character of R&D (→ too little green R&D), too little public transport, information asymmetries. By the theory of the second best, with multiple market failures a single instrument can even make things worse.

Remember: Also acceptability: price-based instruments have less public acceptability than other instruments — relying on them alone risks "spectacular failure" to deliver the needed cuts.

Meckling et al. (2017, Nature Energy) — policy sequencing

What: Smart sequencing prepares the fertile ground for the politically hard instruments (carbon pricing or bans) that the late stages of decarbonisation require.

Remember: Start with education, green industrial policy (green R&D, subsidies, feed-in tariffs) to build a supportive firm coalition and let people see/experience benefits → acceptance for pricing later. The Biden-era Inflation Reduction Act (IRA) wasn't even labelled "environmental" yet that was its main aim — a sequencing move.

Stechenmesser et al. (Science) — assigned reading

What: Documents the proliferation of environmental policy instruments. Panel A: average number of adopted policies + tightenings per country across 4 sectors — buildings, electricity, industry, transport — rising substantially over time, heaviest in transport and electricity. Bottom panel: large heterogeneity of instruments (e.g. transport: EV subsidies, rail public expenditure, labels, low-emission zones, fuel taxation, bans, fossil-fuel-subsidy reforms, vehicle taxes).

Remember: Its second contribution: machine-learning to detect emission breaks (years with a sizeable structural shift) and link them to nearby policies — generating plausible causal hypotheses, not single-policy attribution. The prof explicitly told the class to read it.

Lindenmeyer et al. — "pricing comes last" sequencing visual

Remember: Across ~5 sectors and countries, carbon pricing is the LAST link in the policy chain. Typical order: start with regulatory instruments (sometimes cheap end-of-pipe filters), then information/education (often early), and only toward the end do you start pricing. Complements Meckling et al.

Evaluating effectiveness: the counterfactual problem

Do CO$_2$ prices cut CO$_2$?

Remember: Never use naïve absolute before/after comparisons. The income elasticity of demand means getting richer pushes emissions up over time (longer trips, more flying), masking policy effects. Counterfactual = (actual outcome) − (estimated counterfactual outcome) at each horizon.

Three quasi-experimental methods

1. DID (Difference-in-Differences) / two-way fixed effects: compare treated vs. control, before vs. after. Needs the parallel-trends assumption (similar trends, not levels, before treatment). Threshold-based policies (e.g. EU ETS regulates only above a capacity MW) make treated and controls systematically differ → matching is hard.
2. SCM (Synthetic Control Method, Abadie): reweight donor (control) units to build a "clone" that matches the treated unit's levels pre-treatment. It essentially forces the pre-period to be parallel by construction (puts weights like France 40%, Germany 10%, …).
3. SDID (Synthetic Difference-in-Differences, Arkhangelsky et al. 2021): combines both — assigns unit weights (à la SCM, can be 0) and time weights (extra weight on periods where pre-trends were parallel).

Remember: All are "quasi-experimental" because the ideal — an RCT — is impossible at macro scale. All rest on assumptions (above all, that parallel pre-trends continue post-treatment).

The parallel-trends assumption + PM2.5 overestimation

Remember: The core DID identifying assumption is parallel pre-trends. For CO$_2$ in the German eco-tax case, pre-trends were nicely parallel → standard DID and SDID give similar answers. But for PM2.5 (particulate matter) the pre-trends were NOT parallel → naïve two-way-fixed-effects DID would considerably overestimate the eco-tax's PM-reduction effect; SDID corrects it. Classic "which assumption does this rely on?" question.

Case study: Sweden's carbon tax (Andersson 2019)

Remember: Sweden was among the earliest (early-1990 carbon tax). Build a synthetic clone of Sweden with SCM from ~20 structurally similar countries, using CO$_2$ data back to the 1960s up to 1990. Post-policy, true Sweden and the clone diverge substantially → ~−11% transport-sector CO$_2$ relative to the counterfactual. Note: Sweden's emissions didn't fall much absolutely — the reduction is relative to the counterfactual.

Case study: the German 1999 Eco-Tax (Basaglia et al.)

The policy

Remember: The 1999 ecological tax reform — to date the largest environmental tax reform in the world. Raised taxes on petrol, diesel, natural gas, electricity in yearly steps 1999–2003, ~€0.03/litre per step → ~€0.15/litre of gasoline total. Implicit price ~ $70/tCO$_2$ — second only to Sweden's tax at the time, but a much larger tax base. Revenue recycled to cut non-wage labour costs (lower pension contributions). It is the predecessor of the 2021 carbon price on transport.

The regional SDID analysis

Remember: Goes from country to regional level — ~1200 small EU regions; the 400 NUTS-3 regions in Germany are treated, controls built from other EU regions. Uses gridded pollution maps for CO$_2$, PM, and NOx. Donor pool prunes: late-joining EU countries, border regions (fuel tourism), and countries that introduced similar policies simultaneously. France, Italy, the Netherlands get the largest synthetic weights.

Results: CO$_2$ −15% vs. the synthetic baseline (close to Sweden's −11%); PM and NOx also reduced significantly. Because CO$_2$ pre-trends were parallel, standard DID ≈ SDID ≈ SCM.

Mechanisms

Remember: Three margins, only the first two show up:

1. Fleet renewal: big spike in share of passenger cars ≤2 years old just after the reform. People bought cleaner, more efficient cars.
2. Less driving: road passenger-km fell ~6% for gasoline cars.
3. De-growth story: estimated zeros — no discernible GDP effect. Not a de-growth mechanism.

Health co-benefits dominate

Remember: Aggregating avoided CO$_2$/PM/NOx (1999–2009) × official German Environment Agency per-ton cost estimates gives >€100 billion of avoided damages, of which ~2/3 are local air-quality (PM, NOx) benefits, not CO$_2$. These local benefits accrue to the same people who bear the costs — important for Chapter 11's distribution analysis.

Tax saliency: the 4–5× wedge

Real-price vs. eco-tax elasticity

Remember: Raw real-price elasticity for gasoline ≈ −0.5 (a 1% price rise → ~0.5% lower demand). But the eco-tax elasticity ≈ −2.2 to −2.5 → 4–5× larger. This is the tax saliency ratio (also found in Sweden). Implication: standard ex-ante simulation models that plug in real-price elasticities (as the German government did, ~0.5, to forecast the 2021 carbon price's effect) underestimate eco-tax effectiveness — a much smaller explicit carbon-price rise (~8%) could deliver the same 15% demand cut.

Three mechanisms behind the wedge

Remember: Not a pure income effect. The three are:

1. Moral / signal effect — the government signals it wants you to consume less of the good.
2. Persistence — taxes are perceived as permanent (rarely repealed), whereas market-price spikes are expected to fade → taxes drive long-run investment decisions (which car to buy).
3. Salience — eco-tax introductions get heavy media coverage; a salient, much-discussed price rise registers where a quiet 10-cent market fluctuation doesn't.

We don't yet know which mechanism dominates.

Low taxes are statistically undetectable — Canada

Remember: Many carbon/fuel taxes are introduced at very low levels → impossible to statistically detect an effect amid the noise. Canada's estimated effect spanned roughly +1% to −4% ("likely reduced emissions, can't statistically confirm"). A 1-cent rise shouldn't be expected to move demand detectably.

EU ETS effectiveness

The world's largest supranational ETS

Remember: Started 2005. Regulates almost half of EU CO$_2$ emissions; cap dynamically reduced to reach ~0 around 2040. First period 2005–2008 was a trial run — almost all permits handed out free (grandfathering); since then increasing auctioning → government revenue. Prices: long stretch of low €5–10/tCO$_2$, then a large spike toward €80–100 in 2020–21 (firms convinced the EU is serious about net zero).

Did it actually cut emissions?

Remember (spillover subtlety): "Of course, it sets a cap" — but you must check spillovers: regulated→unregulated firms, regulated/unregulated units within one firm, and leakage to other countries.

• Colmer et al. (2024, REStud): French administrative micro-data → EU ETS induced regulated manufacturing firms to cut CO$_2$ by ~14–16%, with no detectable contraction in activity and no detectable leakage abroad. So it cut emissions even from a global view.
• Bayer & Aklin (2020, PNAS): coarser sector/country data → ~−8 to −10% over 2005–2016 (even at low prices).
• Basaglia et al. (2024, PNAS): EU ETS also cut air pollution → health benefits in the hundreds of billions of euros.

Attribution under policy mixes

Remember: Policies rarely arrive in isolation — they co-occur (pricing + regulation + information), making it hard to attribute an emission change to one policy. Stechenmesser et al.'s machine-learning approach flips the logic: find emission breaks first, then link nearby policies → plausible causal hypotheses rather than clean single-policy effects.

Key formulas & one-line takeaways

Key formulas

Equimarginal / least-cost: $\text{MAC}_i = \text{MAC}_j = p^{\text{CO}_2} = \text{marginal damages}$.

Counterfactual effect at horizon $t$: (actual outcome)$_t$ − (synthetic/counterfactual outcome)$_t$.

Demand reduction: $\%\Delta Q = \varepsilon_p \cdot \%\Delta P$ (use real-price elasticity ≈ −0.5 for market prices, eco-tax elasticity ≈ −2.2 to −2.5 for explicit taxes).

One-line takeaways

• Bans and ETS both set a cap, but the ETS equalises MACs (cost-effective) and can raise revenue; bans leave beneficial trades on the table and only impose costs.
• Taxes (Weitzman) vs. ETS (Karp–Traeger): flow pollutant → tax; stock pollutant like climate → ETS.
• Hybrids — price collar, smart/flex cap (Traeger et al. 2020) — manage the price-vs-quantity uncertainty trade-off.
• Don't rely on pricing alone: multiple market failures, too little green R&D, and weak acceptability → use a sequenced mix (Meckling et al. 2017; "pricing comes last," Lindenmeyer et al.).
• Proving effectiveness needs a counterfactual: DID (parallel trends), SCM (level-matched clone), SDID (unit + time weights). RCTs impossible at macro scale.
• For PM2.5 the pre-trends are not parallel, so naïve DID overestimates and SDID corrects — know which assumption fails.
• Sweden (Andersson 2019): −11% transport CO$_2$; German eco-tax (Basaglia et al.): −15% CO$_2$, ~2/3 of benefits are local air quality, no GDP hit.
• Tax saliency: eco-tax elasticity is 4–5× the real-price elasticity (signal + persistence + salience), so simulation models using market elasticities underestimate tax effectiveness.
• EU ETS works: Colmer et al. (2024) −14–16% with no contraction/leakage; Bayer & Aklin (2020) −8–10%.
• Read Stechenmesser et al. (Science): proliferation of instruments + ML emission-break detection → plausible causal hypotheses.