
In 2007 I gave a talk on how we could prevent global climate collapse with some clever financial engineering, and in 2015 I posted the slides on this blog after the Paris Agreement was signed. At the time I said some of the modelling and facts were “a little out of date”. With COP31 running in Antalya this week, I thought it was time to find out just how out of date.
So I went back through every slide and checked every number and every bit of maths against the best data available in November 2026. Some of the talk has held up well. Some of it was wrong, and one error (my fault) changes the conclusion. Here is the update, with the new slides at the end.
The Short Version
The basic idea still stands. We can get a coalition of countries to act now, without first agreeing who pays, by financing the transition with very long-dated bonds that our grandchildren repay. What has changed is the arithmetic. Clean energy is now cheap, so the total cost is far lower than I feared in 2007, at about 1.5% of world GDP a year at the peak. But my 2007 model understated what future generations would have to repay by about three and a half times. With real interest rates where they are today, Carbon Bonds move the bill into the future. They don’t shrink it.
What Has Changed Since 2007
The planet has kept warming. 2024 was the first calendar year more than 1.5 °C above pre-industrial temperatures (1.55 °C), and 2025 came in at 1.43 °C according to the WMO State of the Global Climate 2025. The latest Indicators of Global Climate Change puts human-caused warming at 1.37 °C, rising at 0.27 °C per decade, which means we will cross 1.5 °C of human-caused warming around 2030.
The good news (and it is real good news) is that emissions have almost stopped growing. Total greenhouse gas emissions were 58.4 billion tonnes of CO2 equivalent (Gt CO2e) in 2025, about the same as 2024, according to the UNEP Emissions Gap Report 2026. The Global Carbon Budget 2026 (still a preprint, so treat it with some caution) finds total CO2 grew only 0.3% a year over 2016–2025, down from 1.7% a year over 2006–2015. Emissions are at a record, but they are on a plateau.
The bad news is that the plateau is far too high. At current emissions the remaining carbon budget for 1.5 °C (130–150 Gt CO2) is used up in about four years. The 1.5 °C target is effectively gone. 2 °C is still in reach, with roughly 24 years of current emissions left in the budget.
My 2007 Consequences Were the Worst Case
I should be honest about this slide. In 2007 I listed the consequences as a 6 to 10 °C rise, no plant life between the tropics, both ice caps melting, a 100 m rise in sea level, the oceans turning acid and a 90% extinction rate. These were the far tail of the worst case (sorry).
The 2026 picture is still bad enough:
- Temperature. About 2.8 °C by 2100 on current policies, or 2.1 °C if every country meets every pledge (UNEP 2026).
- Sea level. Greenland and Antarctica together hold about 65 m of sea level (not 100 m), and losing them takes centuries to millennia. The IPCC’s high scenario gives roughly 0.6 to 1.0 m by 2100. Sea level is now rising at about 3.8 mm a year.
- Ice sheets. Armstrong McKay and colleagues put the tipping thresholds for Greenland and West Antarctica at somewhere between 1.5 and 2 °C, so they may already be committed.
- Oceans. Surface pH has fallen from about 8.2 to 8.1 since the industrial revolution, which is a 30 to 40% increase in hydrogen ions, and CMIP6 models project a total fall of 0.16 (low emissions) to 0.44 (high emissions) by the end of the century. The oceans are more acidic, but they are still alkaline.
- Extinction. About 1 million of an estimated 8 million species are threatened with extinction, and the extinction rate is tens to hundreds of times the background rate (IPBES 2019).
- The tropics. I can find no evidence for “no plant life between the tropics”. The real tropical risk is heat stress on people and degraded forests.
The feedbacks I worried about in 2007 are real. The 20 lowest Arctic sea ice minima on record have all been since 2007, and south-east Amazonia has become a net carbon source. But they have not run away. Arctic ice loss has plateaued since 2012, Amazon deforestation is falling, and methane growth has dropped from about 17 ppb a year in 2021 to 4.8 ppb in 2025.
I also suggested that emissions were higher than even the worst-case scenario. That turned out to be wrong. My 2007 model used the IPCC’s A1FI worst case, which had emissions at about 68 Gt a year by 2025. Actual 2025 emissions were 58 Gt. We missed the worst case, and high-emissions pathways like it are now considered implausible. Unfortunately we also missed the Stern Review 550 ppm path, which had emissions at about 51 Gt and falling by now.
Why Is So Little Being Done?
In 2007 I asked why nothing was being done. The simple answer was money, and the complex answer was also money. In 2026 something is being done, just not enough, but the answer is still money. What has changed is the size of the bill.
My 2007 slide claimed abatement would cost US$50 to $500 per tonne, that Australians emit about 20 tonnes each, so stopping it would cost each of us $2,000 to $20,000 a year, while people were only willing to pay about $120 a year. That is a gap of more than 90%.
Every part of this has moved:
- Cost. The IPCC’s 2022 mitigation report says options costing US$100 per tonne of CO2e or less could halve global emissions from 2019 levels by 2030, and that more than half of that potential costs less than US$20 per tonne (SPM C.12.1). Utility solar is 90% cheaper than in 2010, onshore wind 70% cheaper, and battery packs about 93% cheaper (BNEF).
- Emissions per person. Australians now emit about 16.4 tonnes each, including land use, down from 30.3 tonnes in 2005 (DCCEEW, year to March 2026).
- Cost per person. A 95% cut at an average of US$25 to $120 per tonne works out at about US$390 to $1,870 per Australian per year.
- Willingness to pay. A survey of about 130,000 people in 125 countries found that 69% say they would give 1% of their income to fight global warming. My 2007 figure of $120 a year turns out to be about right for its time: a 2010 review of 27 willingness-to-pay studies found a median of US$135 per household per year (range $22 to $437), and a 2013 US survey found US$79 to $89. Note that these are per household, not per person, and I can’t find any support for my claim that it was the same across western countries and stable for 15 years.
At the median cost, the gap between what climate action costs and what people say they will pay is now small, or gone. Stated willingness is not the same as actually paying, but it is a very different world from 2007.
Vested Interests Are Smaller Than I Said
My 2007 slide said the oil industry had extracted 100 trillion barrels since 1860, with another 100 trillion barrels to come over 40 years. That was wrong by a factor of about 70, and was likely a slip for the actual 100 million barrels a day: 100 million × 365 × 40 is 1.46 trillion barrels. Proved reserves are about 1.57 trillion barrels. About 1.25 trillion barrels had been produced by 2011, and adding production since then puts the total at roughly 1.7 trillion barrels today (my estimate).
The value at stake is also much smaller than I suggested. Estimates of fossil-fuel wealth stranded by a 2 °C transition are US$1 to 4 trillion, with more than US$1 trillion in upstream oil and gas. If the incumbents stand to lose a few trillion dollars, it is far cheaper to buy them out than to fight them. That was always part of the Carbon Bank idea, and the new numbers make it much more attractive.
Infrastructure Is Already Being Replaced
In 2007 I said our infrastructure was 98% fossil-based and would cost US$10 to $200 trillion to replace, or 0.5 to 10 years of world GDP. The second half of that sentence was simply bad maths. Even at 2007 world GDP (about US$57 trillion) it is 0.18 to 3.5 years, and at today’s US$117 trillion it is 0.09 to 1.7 years.
The replacement has also started. Fossil fuels made up 57% of world electricity in 2025, and renewables overtook coal for the first time (Ember). Clean energy investment is running at about twice the rate of fossil investment, about US$2.2 trillion against US$1.1 trillion (IEA). Fossil fuels still supply about 86% of total energy, so there is a long way to go, but we are no longer starting from zero. McKinsey estimates that net zero by 2050 needs about US$3.5 trillion a year more capital spending than in 2020, which is about 3% of world GDP. That is gross spending, not the net cost, since much of it pays for itself in cheaper energy.
International Co-operation Is Still the Hard Part
In 2007 I said the USA was out “until ?”, China was out until after 2050 and India was out, and that these three made up 75% of future emissions. Two of those predictions turned out wrong in a good way. China has pledged to cut emissions 7 to 10% below their peak by 2035 and to be carbon neutral by 2060, and its CO2 emissions have been roughly flat since early 2024. India has a 2035 target and a net zero date of 2070. The USA went the other way: it left the Paris Agreement on 27 Jan 2026 and on 7 Jan 2026 announced it will leave the UNFCCC altogether, a withdrawal that takes effect on 27 Feb 2027. The three countries together make up about 53% of today’s fossil CO2 emissions, not 75%. I can’t find a source for the 75%.
The core problem has not changed. No country wants to pay unless every other country pays too, and the biggest historical emitter has walked away. This is exactly the problem the Carbon Bank was designed to get around.
The Carbon Bank in One Paragraph
A small group of countries set up a new bank (the Carbon Bank) and jointly underwrite its debt. The bank issues 50-year, sovereign-backed, inflation-indexed, zero-coupon bonds (Carbon Bonds). A zero-coupon bond pays no interest, but is sold at a discount and repays its full face value at the end. The money raised is spent buying and closing emitting assets and building zero-emission replacements. Nobody has to agree today on who pays. When the bonds fall due, each member pays its share based on the size of its economy at that time, with a discount for countries that joined early, and later joiners take on a negotiated share of the debt.
Checking My 2007 Maths
Here is every error I found, most of them minor, one of them not:
- Cost per person. 20 tonnes × US$50–500 is US$1,000–10,000, not $2,000–20,000. If the $50–500 was per tonne of carbon (as the slide’s “C equiv.” suggests), it is only US$273–2,727. The source I cited, Anderson’s 2006 paper for the Stern Review, actually had central costs of US$22–61 per tonne of CO2, and for a much smaller cut.
- Oil. 100 trillion barrels should be about 1.5 trillion (see above).
- Years of GDP. $10–200 trillion was 0.18–3.5 years of 2007 world GDP, not 0.5–10.
- Bond price. A 50-year zero-coupon bond at 2.2% real sells for 34% of its face value, not ~30%. My model actually used 2.5%, which gives 29%. The two slides were inconsistent.
- Growth. I assumed 3.9% real world growth to 2050, which is too high. It implies world GDP tripling by 2050, whereas the IPCC expects it to “at least double”. Oddly, the GDP path inside my own model grew at only about 2.4–2.5% a year.
- The big three. USA, China and India are 53% of current emissions, not 75%.
- Repayment cost. My “Cost % GDP” chart claimed the high case would cost future generations at most 1.27% of world GDP, in 2100. When I worked backwards from the charts, every point on the cost chart equals the spending 50 years earlier multiplied by about 0.29. It should have been multiplied by 3.4 (the 50 years of compound interest) and divided by 50 years of GDP growth. Corrected, the high case repays about 4.5 to 4.8% of world GDP in 2100, not 1.27%, and the median about 1.4%, not 0.39%. I can’t tell exactly what went wrong in the spreadsheet. 0.29 is the price of the bond, so it may have applied the discount the wrong way round, but the GDP path inside the model grew at about 2.5% a year, almost the same as the interest rate, so simply leaving out the 50 years of interest gives the same 0.29. Either way the repayment was understated by about three and a half times, which is the part that matters.
Error 7 matters because it carried the conclusion of the talk. I told people the cost to future generations would be trivial. It wasn’t.
Re-running the Model With 2026 Numbers
I rebuilt the model from scratch and kept it as simple as possible, so you can see every assumption and disagree with it:
- Emissions. Flat at 58 Gt CO2e a year (today’s plateau), instead of the A1FI worst case.
- Cut. 95% by 2050, in a straight line from 2027, then held to 2100.
- Cost. An average of US$25 (low), US$60 (median) or US$120 (high) per tonne of CO2e, held flat in real terms. These are average costs. The last tonnes are much more expensive than the first (the IPCC’s cheap options only get you to a halving), so for a 95% cut the high case may be closer to the truth than the median.
- Window. Like my 2007 model, the bonds pay for every year’s abatement out to 2100. I also ran a version where the bonds pay only for the 2027–2050 build-out and later costs are paid as we go.
- Growth. World GDP of US$117 trillion in 2025, growing at 2.7% a year to 2050 and 2.0% a year after that.
- Bonds. 50-year zero-coupon at real yields of 1.8% (the September 2024 low on US 30-year inflation-linked Treasuries), 2.5% (typical of 2025) or 3.25% (about where they are today: 3.24% on 12 Nov 2026).
The results:
| Case | Peak spend (2050) | Repayment in 2100 at 1.8% / 2.5% / 3.25% real | Peak stock of Carbon Bonds: bonds to 2100 | Peak stock: build-out only (to 2050) |
|---|---|---|---|---|
| Low (US$25/t) | 0.6% of GDP | 0.55% / 0.77% / 1.11% | 21–31% of GDP | 9–14% of GDP |
| Median (US$60/t) | 1.45% of GDP | 1.32% / 1.85% / 2.67% | 51–75% of GDP | 21–33% of GDP |
| High (US$120/t) | 2.9% of GDP | 2.63% / 3.71% / 5.34% | 102–150% of GDP | 42–65% of GDP |
The median case spends 1.45% of world GDP in 2050 (about US$3.3 trillion a year). That is the same order of magnitude as the IPCC’s estimate that 2 °C pathways leave world GDP 1.3–2.7% lower in 2050, although that is a loss of output rather than spending, so it is a rough comparison, not a check.
Carbon Bonds Move the Bill, They Don’t Shrink It
The key number is the gap between the real interest rate (r) and real economic growth (g). A 50-year bond repays (1 + r)^50 times what it raised. If the economy grows by (1 + g)^50 over the same time, then future generations pay a smaller share of their income than we would have, only if r is below g. With growth after 2050 at 2.0% a year:
- At 1.8% real, the burden shrinks slightly (×0.91).
- At 2.5% real, it grows by 28%.
- At today’s 3.25% real, it nearly doubles (×1.84).
In 2007, and through most of the 2010s, long real interest rates were below growth, and Blanchard’s argument that public debt was cheap looked solid. Since the 2022–2024 rate rises that is far less clear. At today’s rates each dollar raised by a 50-year zero-coupon bond costs nearly five dollars, in real terms, to repay.
The other problem is size. If the bonds pay for everything out to 2100, as in my 2007 model, the median stock of Carbon Bonds peaks at 51–75% of world GDP in the 2080s. Most of that (80% of the money) is borrowed after 2050 to keep paying for abatement once the transition is done, which makes little sense. If the bonds pay only for the 2027–2050 build-out, the median peak falls to 21–33% of world GDP. That is still a lot on top of global public debt that is already about 94% of GDP, and issuing it would itself push real interest rates up.
I should add one thing that my 2007 talk glossed over completely. Debt shifts who pays, but it doesn’t change what is built or when. The steel, concrete and labour are used today. Carbon Bonds don’t make the transition free for the present generation. They just change which people alive today pay for it (savers who buy the bonds, rather than taxpayers).
So where does that leave the idea? The financial case for Carbon Bonds is weaker than I thought in 2007, but the political case is as strong as ever. The whole point was to let a coalition act today without first fighting over who pays. That still works. In practice I think it would need to be a mix: bonds for the up-front buyouts and infrastructure when real rates are low, and pay-as-you-go (carbon prices and budgets) when they are high.
Has Anyone Else Suggested This?
Yes, although as far as I can find nobody has proposed exactly this combination. The closest is John Broome and Duncan Foley’s chapter A World Climate Bank (2016, in Institutions for Future Generations, Oxford University Press). They propose a mutual bank, like the IMF or World Bank, that issues very long-dated World Climate Bonds to pay for the switch from fossil fuels to renewables and to compensate the owners of fossil reserves. That is very close to my 2007 idea. Like me, they borrow so that the present generation makes no sacrifice, and future generations inherit the debt but still come out ahead because they inherit a better climate (Broome calls this “efficiency without sacrifice”). The differences are that their bonds pay interest, with maturities discussed in the hundreds of years, and they don’t allocate costs by GDP at maturity or offer a discount for joining early.
A few others are worth knowing about:
- Rezai, Foley and Taylor (2012, Economic Theory) is the economic theory behind the “no generation has to sacrifice” argument.
- Jeffrey Sachs (2015, Oxford Handbook of the Macroeconomics of Global Warming) proposed sharing the burden between generations. Flaherty, Gevorkyan, Radpour and Semmler (2017, Research in International Business and Finance) turned this into climate bonds that pay for mitigation now and are repaid by future generations, which is the closest anyone has come to my Carbon Bonds. The idea was developed further in a 2019 World Bank working paper.
- Orlov, Rovenskaya, Puaschunder and Semmler (2024, AIMS Environmental Science) tested green bonds repaid by future generations in the DICE climate-economy model. They found bonds reduce but don’t eliminate the unfairness between generations, and that lower interest rates help, which matches my r versus g result.
- Kotlikoff, Kubler, Polbin, Sachs and Scheidegger (2021, International Economic Review) show that a carbon tax combined with transfers between generations can leave every generation better off.
- Bård Harstad’s Buy Coal! (2012, Journal of Political Economy) makes the case for a climate coalition simply buying fossil deposits and leaving them in the ground.
- William Nordhaus’s Climate Clubs (2015, American Economic Review) tackles the same free-rider problem with trade penalties instead of a discount for early joiners.
There are also real-world versions of parts of it. The Asian Development Bank’s Energy Transition Mechanism (2021) buys out coal plants for early retirement, though on a tiny scale. The EU has issued about €84 billion of NextGenerationEU green bonds, and the aligned green and sustainable bond market is now issuing over US$1 trillion a year. Austria, Argentina and Mexico have all sold 100-year bonds, so very long-dated sovereign debt is not science fiction (although Argentina’s shows how badly it can go).
What I haven’t found anywhere is the mechanism that I think makes the Carbon Bank politically workable: zero-coupon bonds, with each country’s share decided only when the bonds fall due, based on the size of its economy at that time, and a discount for joining early. If anyone knows of a paper that proposes this, please let me know.
Objections
Isn’t this just passing the buck to our grandchildren?
Yes and no. Yes, that is the point: they get most of the benefit and they will be much richer than us (world GDP should at least double by 2050 even with climate action). No, because the alternative is to pass them a broken climate instead, which is a much bigger bill that nobody can repay.
Why not just use a carbon tax?
Because it hasn’t happened at the scale needed. Carbon pricing now covers about 29% of global emissions, at an average price of about US$21 per tonne. A carbon tax asks voters to pay today. Carbon Bonds don’t, which is why they are worth considering alongside a tax, not instead of one.
What if a member country defaults in 2077?
This is a real risk, and I don’t have a full answer. The bonds are only as good as the coalition behind them, which is why the founding members would need to be large, stable economies that jointly guarantee the debt.
Won’t the debt count against governments today anyway?
Yes. Ratings agencies, bond markets and fiscal rules will count a sovereign guarantee of Carbon Bonds as debt now, not in 2077, so the bonds don’t let governments escape today’s fiscal limits. What they do change is who pays and when: the cash cost to today’s budgets is zero.
Doesn’t allocating costs by GDP at maturity load the bill onto today’s poorer countries?
Yes and no. India, Indonesia and Africa will make up a much larger share of world GDP in 2077 than today, so they would pay a larger share of the bill than their current size suggests. That is partly the point, since they will also be much richer, but it gives them a strong incentive to walk away before the bonds fall due. The discount for joining early helps, but on its own it doesn’t stop anyone free-riding. There may need to be some sort of penalty for joining late such as trade restrictions or tariffs.
Will anyone buy a 50-year bond that pays nothing until the end?
Pension funds and insurers do buy very long-dated bonds to match their liabilities, and Austria, Argentina and Mexico have all sold 100-year bonds. But a zero-coupon bond backed by a coalition that has to survive for 50 years is a much bigger ask, and the price would include a premium for that risk. I should also add that my model treats growth as fixed, when unchecked climate change would itself lower growth, which makes doing nothing look cheaper than it really is.
Your 2007 numbers were wrong, so why should I trust these?
Fair question. These numbers are better sourced, the model is far simpler, and I have published the assumptions so you can check them. Of course I could be wrong again, so if you find an error please leave a comment.
What Do I Think Will Actually Happen?
Probably nothing, but why let that get in the way of a good idea.
If anyone wants them, here are the updated slides and the original 2007 slides. I am also happy to share the model code. If you can see a fatal flaw in the idea (not just that step X might be hard), I would really like to hear about it.