A compliance obligation is a forward short position in a physical product with a settlement date, which makes it answerable to the same treatment a commodity book already applies to fuel, freight and metals. The shortfall is usually knowable years ahead of the surrender deadline from production plans, the published free allocation schedule and measured emissions intensity, and it goes unmanaged because nobody owns the position. Compliance instruments and voluntary credits belong in one book with their attributes kept separate, since a voluntary credit can lose value after purchase through reversal, methodology revision or a buyer refusing its vintage. Abatement and purchasing are the same decision at the margin, comparing the cost per tonne of a reduction project against the forward price of the instrument that would otherwise cover it.
The surrender deadline is in April. The buying happens in February and March, because that is when the treasury team gets the verified emissions figure, works out the shortfall against the allowances already held, and goes to the market with a number.
The shortfall was knowable three years earlier. Production plans existed, the free allocation schedule was published, the emissions intensity of the assets was measured. What was missing was somebody whose job it was to hold a position against a forward obligation, so the exposure sat unmanaged until it became an invoice.
Anyone who has run a commodity book will recognise the pattern, and the fix is the same one that industry worked out decades ago for fuel, freight and metals. A compliance obligation is a forward short position in a physical product with a settlement date, and it responds to the same treatment.
Size the position before deciding anything else
The position has a straightforward definition and takes real work to compute. For each year over your planning horizon, projected emissions from the production plan, minus free allocation under the rules as they stand, minus allowances or credits already held, gives the volume you must acquire. Do that per regime, since obligations in different jurisdictions are not interchangeable.
Two features make this harder than a fuel position and more important to get right.
The first is that the obligation moves with policy on a published schedule. In the EU, free allocation for sectors covered by the carbon border adjustment mechanism phases down over the period to 2034 as that mechanism phases in from 2026, so a plant with flat emissions has a growing purchase requirement every year with no operational change behind it. Regimes elsewhere are expanding their coverage: China extended its national trading system beyond power generation to steel, cement and aluminium in 2025. A position sheet built on this year's rules understates the exposure of a plant that will still be running in 2032.
The second is that the position is joint with production. Emissions follow output, so a production forecast is an input to the carbon position, and the same uncertainty that sits in the production plan sits in the obligation. That argues for covering a base layer corresponding to volume you are confident about, and leaving the variable layer to be covered closer in.
A coverage policy is the artefact that turns this into a routine. A stated fraction of the obligation covered at twenty four months, a higher fraction at twelve, the balance inside the year, with tolerance bands and a named person who is allowed to sit inside them. The policy matters more than any individual purchase decision, because it moves execution away from the two months a year when everybody with the same deadline is buying.
Two markets, one book
Most organisations end up spanning both markets, and it helps to be clear about how different they are.
Compliance instruments are created by a regulator, are fungible within their jurisdiction, have exchange-traded price discovery and a futures curve, and carry essentially no delivery risk once settled. The scale is substantial: the World Bank's State and Trends of Carbon Pricing report for 2024 put global carbon pricing revenues above 100 billion dollars for the first time, covering around a quarter of global emissions. The risks are price and policy, and both can be reasoned about.
Voluntary credits are issued by standard-setting bodies against project methodologies. Two credits both representing one tonne can differ in price by two orders of magnitude, because the buyer is purchasing a story about a counterfactual along with the tonne. There is no fungibility, no reliable index, and price discovery is bilateral. The market has also contracted sharply, with Ecosystem Marketplace reporting in 2024 that transaction value had fallen by more than half from its 2022 level, largely on the back of integrity concerns.
Both end up in the same book because the obligations that drive them arrive together. Regulated obligations require compliance instruments. Public commitments, customer contract clauses, and sector schemes such as ICAO's CORSIA require something else. Running these as separate spreadsheets owned by separate functions is how an organisation ends up retiring an expensive voluntary credit in the same quarter it buys compliance allowances on a spike.
What can go wrong with a voluntary credit after you buy it
Four risks are worth naming separately, because they need different mitigations.
Non-delivery. Forward purchase agreements with projects that have not yet issued credits carry the risk that the project underdelivers, is delayed, or never issues at all. This is the ordinary risk of buying something before it exists, and the mitigation is ordinary too: replacement obligations in the contract, staged payments against issuance, and a limit on how much of the book sits in forwards.
Reversal. Credits from stored carbon can be undone. Forests burn, land use changes, and the sequestration is released. Standards address this with buffer pools funded by a levy on issuance, which is a reasonable design and one whose sizing depends on the frequency of the events it is meant to absorb. An analysis published by CarbonPlan in 2022 estimated that wildfire losses had already consumed a large share of the buffer pool that California's forest offset programme had set aside to cover a century of risk, within about a decade of the programme starting. Buffer pools are insurance with an actuarial assumption inside them, and the assumption is worth reading.
Re-rating. A credit that was acceptable when purchased can be judged inadequate later. Work published in Science in 2023 by West and colleagues found that most of the avoided deforestation projects they examined had not reduced deforestation by anything like the amount credited. The Integrity Council for the Voluntary Carbon Market published its Core Carbon Principles in 2023 and has since assessed methodologies against them, declining to approve several categories including certain grid-connected renewable energy methodologies in 2024. A stockpile bought on the previous consensus can lose its usability without any physical event occurring.
Claim risk. The reputational exposure attaches to the claim rather than the credit. The Voluntary Carbon Markets Integrity Initiative published a Claims Code of Practice in 2023 setting out what a company can say having bought credits, and the gap between what a credit is and what people say about it is where most of the public criticism has landed.
The mitigations that address several of these at once are diversification across project type, geography, vintage and developer, a preference for issued credits over forwards where the price difference allows, retirement close to the date of the claim rather than years of holding, and a documented rationale for each purchase that will still read sensibly in five years.
Vintage, additionality and the fungibility problem
A tonne is a tonne physically. Commercially, credits carry attributes that determine who will accept them, and the attributes are where the price dispersion comes from.
Vintage is the year in which the reduction occurred. Buyers increasingly want recent vintages, schemes impose their own vintage eligibility windows, and old inventory can become unsaleable while remaining perfectly valid. Holding a large position in aged vintages is a bet on future acceptability rules.
Additionality is the claim that the reduction would not have happened without the credit revenue. It cannot be measured, only argued from a counterfactual, and the strength of that argument depends on the methodology and on the economics of the underlying activity at the time. This is the property that has driven most of the re-ratings, because a project type that was plausibly additional when the technology was expensive stops being so once that technology is the cheapest option on its own merits.
Authorisation is the newest attribute and the one creating hard supply constraints. Under Article 6 of the Paris Agreement, with rules agreed in 2021 and operational details settled at the 2024 conference, a credit used towards another country's target requires the host country to make a corresponding adjustment to its own accounts. CORSIA-eligible units need that authorisation, and host governments have been slow to grant it, since authorising a credit means giving up a reduction they could have counted themselves. An airline holding units without authorisation holds something that does not satisfy its obligation.
The practical conclusion is that a credit portfolio needs attribute-level inventory management. Each holding has a project type, a vintage, a registry, an authorisation status, a standard, and a set of buyers or regimes that would currently accept it. Managing it at the level of total tonnes held loses exactly the information that determines whether the position covers the obligation.
Abate or buy, as a marginal comparison
The decision to reduce your own emissions and the decision to buy coverage are the same decision viewed from two ends, and the comparison is marginal cost against market price.
Build the internal side as a cost curve. For every abatement option available to you, compute the cost per tonne avoided over the life of the asset, discounted, alongside the volume it delivers and the year it starts delivering. The global cost curve published by McKinsey in 2009 popularised this presentation, and the corporate version is more useful than the global one because your options are real and priced.
Then compare against the forward price of the obligation, not today's spot. This is where most comparisons go wrong. An abatement project delivers every year for its remaining life, while a purchased credit or allowance covers a single year's tonne. Comparing a project's cost per tonne against one year's price systematically favours buying. The correct comparison puts the project's cost against the discounted stream of prices you would otherwise pay over the same horizon, using a forward curve where one exists and a stated policy assumption where it does not.
Lead time then does the rest of the work. Abatement projects take years to deliver, so the decision has to be taken well before the tonne is needed, which means the buy decision for the near years is largely already made. The realistic shape is a plan where internal abatement covers the later years, purchasing covers the earlier ones, and the crossover point moves as project costs and carbon prices change.
Two cautions on the cost curve itself. Options interact, so a set of projects each computed independently will not sum to their individual figures once implemented, particularly where several reduce the same energy stream. And the cost per tonne is sensitive to the discount rate and to the assumed asset life, which means the ranking is more stable than the absolute figures. Use it to order the options and to find the rough crossover point, then examine the projects near that point in more detail.
The limit
Credit prices and eligibility rules move with policy, and a portfolio built on a single regulatory assumption carries a risk that nothing in this market hedges well.
There is no instrument that pays out when a methodology loses its approval, when a host country declines to authorise, when free allocation phases out faster than legislated, or when a scheme's eligibility criteria change between purchase and surrender. Futures cover price within a regime and assume the regime. Insurance products for reversal and invalidation exist and are improving, and their coverage is narrower than the risk.
What is available is structural rather than financial. Diversify across regimes and instruments so no single rule change takes out the whole book. Stage purchases across time so that a rule change caught midway leaves you with a partial position rather than a full one at the wrong specification. Prefer instruments whose value derives from a physical reduction you can point to, since those have proved more durable through successive rounds of scrutiny. And keep the internal abatement pipeline funded even when purchasing looks cheaper, because a tonne you no longer emit is the only position immune to a change in the rules.
Build the position sheet first: projected obligation by year and by regime, against what you hold today, with the attributes of each holding listed rather than summed.
More on this in energy on the DemandLabs platform.
