The Polycrisis Bill

The narrowest reading — market losses from disasters and heat — comes to about $1.4 trillion a year, 1.2 per cent of world output. Even this understates market losses, because it omits the output cost of conflict, the productivity cost of pollution-related illness and the disruption that infrastructure failures impose beyond the assets destroyed: the World Bank estimates that infrastructure disruptions alone cost households and firms in low- and middle-income countries hundreds of billions of dollars a year.

Adding the health cost of air pollution raises the total to about $9.5 trillion, 8 per cent of output. This is the largest single entry and the most contested, because it values premature deaths at what people are willing to pay to reduce mortality risk. It is also the entry most directly tied to the energy system: most of it would disappear with clean air, which is why the IMF counts unpriced air pollution as the largest part of the $6.7 trillion of implicit fossil-fuel subsidies it estimated for 2024.

Adding the non-military impact of violence brings the total to about $17 trillion, 14 per cent of output, and the full ledger — with military spending and the climate liability of current emissions — to about $28 trillion, 23.5 per cent. That upper figure mixes present losses, defensive spending and future damage, and should be read as an upper bound rather than a total. The useful conclusion is the range itself: depending on the question asked, the polycrisis costs between one and twenty-odd per cent of world output a year.

Output or welfare. National accounts record what is produced and sold; they do not record a death, except as lost earnings. Welfare measures add the value people place on life and health, which for air pollution is much larger than the lost output. Neither is wrong, but they answer different questions: what the economy loses, and what people lose. Flows or liabilities. A flood destroys assets this year; a tonne of CO2 emitted this year causes damage for decades. Counting the social cost of this year's emissions adds a liability that will not appear in any year's losses until long afterwards.

Market or purchasing-power dollars. The economic impact of violence is published in purchasing-power terms, which value spending in poorer countries more highly. Applied to market output it is smaller in dollars, though the same share. Overlap. The climate-attributable share of extreme weather is already inside disaster losses and health costs; the health cost of pollution overlaps with fossil-fuel subsidies and with the social cost of carbon's health component. A defensible total includes each loss once.

Level or growth. Most estimates treat losses as a share of output in a given year. If a crisis lowers the growth rate rather than the level, the loss compounds: a polycrisis that shaves 0.2 percentage points off growth each year leaves the world economy about 14 per cent smaller by 2100. Discounting. A dollar lost in 2090 counts for less today than a dollar lost now. At a 2 per cent discount rate, the present value of the central century bill is less than two-fifths of its undiscounted total; at 3 per cent, about a quarter. Section 4 lets readers choose.

Three patterns stand out. First, today's largest losses are not climate damages but the health cost of pollution and the economic impact of violence, each between 6 and 12 per cent of output on welfare measures. The polycrisis is already expensive before its climate component reaches full scale. Second, climate losses are the fastest-growing entry: disaster losses, heat losses and the stock of damaging emissions all rise with warming, and the higher-warming projections for 2050 are larger than any current entry.

Third, the projections are not additive either. A world in which climate damage reaches 15 per cent of output is one in which health, migration and conflict costs also rise, partly because of the same warming, and the estimates cannot simply be stacked. That is the polycrisis point: interacting losses can exceed the sum of the separately modelled channels, or overlap with them, and no published model captures the interaction well.

For the century calculator, this evidence suggests a range of starting shares between about 2 per cent (a narrow, mostly market measure) and 8 per cent (a welfare measure without violence), rising by 2100 to between 5 and 20 per cent. The presets in section 4 use exactly those bounds, with a central path from 4 to 10 per cent. Readers who think the evidence points elsewhere can move the sliders.

On the central path, losses rise from about $4.7 trillion in 2025 to about $75 trillion a year by 2100, because both the share and the economy grow. The bill is $205 trillion in 2025–2049, $529 trillion in 2050–2074 and $1.33 quadrillion in 2075–2100: almost two-thirds of the century's losses fall in its last 26 years. The total is about $2.1 quadrillion, or $781 trillion in present value at a 2 per cent discount rate and $514 trillion at 3 per cent.

The working range in the table — $100–300 trillion, $200–700 trillion and $300 trillion to over $2 quadrillion for the three periods, $600 trillion to over $3 quadrillion in total — falls inside the calculator's low-to-high band. A total of $600 trillion corresponds to losing about 2.3 per cent of output every year; $3 quadrillion to about 11.5 per cent. In other words, the range is a statement about the share of output lost, and the calculator makes that statement explicit.

Growth matters almost as much as the loss share. At 1.5 per cent real growth the central path costs $1.25 quadrillion; at 3 per cent, $2.7 quadrillion. Richer futures have more to lose. The quadrillion-dollar figures are therefore not exaggerations but the ordinary consequence of applying single-digit percentage losses to the output of a growing world economy over 75 years — which is also why preventing even a fraction of them is so valuable.

At the defaults, prevention avoids about $800 trillion of losses over the century for about $390 trillion of spending — roughly $2 back for every $1, and a net saving of about $400 trillion. In present value at 2 per cent the return is about 1.8 to 1. These are cautious numbers: the published benefit-cost ratios for adaptation, early warning and resilient infrastructure run from 2 to more than 10, and the calculator counts none of the co-benefits of cleaner air, healthier populations or more stable economies.

Timing matters as much as scale. Because losses grow with the economy and with the share lost, prevention that starts now avoids losses that would otherwise compound; prevention delayed to mid-century faces a larger stock of damage and a shorter period in which to recover its cost. The twenty-year ramp in the calculator is a reminder that most of what will protect the world economy in 2050 — grids, buildings, water systems, institutions — has to be built in the next two decades.

The ledger also points to where the money is. The largest current entries — pollution, violence and the climate liability of emissions — are the ones most amenable to prevention: cleaner energy removes most of the pollution cost and much of the climate liability, and conflict prevention costs a tiny fraction of what violence costs. The polycrisis bill is not a fixed charge. It is the price of the prevention the world has not yet bought.

World output is the World Bank's estimate for 2025, $118.35 trillion at current prices. Ledger entries are published estimates in their own years and currencies, listed with their sources; the economic impact of violence is converted from purchasing-power terms by applying its non-military share of gross world product (55 per cent of 11.6 per cent, as published by the Institute for Economics and Peace) to 2025 world output. The climate liability multiplies the Global Carbon Budget's 2025 total CO2 emissions, 42.2 billion tonnes, by the US EPA's 2023 social cost of carbon of $190 a tonne at a 2 per cent discount rate. Ledger totals are shares of 2025 output and are not adjusted for overlap.

The century calculator grows output at a constant real rate from 2025 and applies a loss share that rises linearly from its 2025 to its 2100 value; periods are 2025–2049, 2050–2074 and 2075–2100. The present value discounts each year's loss at the chosen rate to 2025. The prevention calculator spends a constant share of output each year and avoids a share of each year's losses that ramps up linearly over twenty years. All values are in 2025 dollars at market exchange rates. The working range in the period table is an earlier H Heuristics synthesis, shown for comparison.

The calculators omit growth effects (losses are treated as levels, not as reductions in the growth rate), tipping points and catastrophic tail risks, all of which would raise the bill, and they treat losses and prevention as smooth when both are lumpy. They are tools for making assumptions explicit, not forecasts. Related analysis in this series: the economics of prevention, the Polycrisis Apocalypse and resilience in an age of polycrisis.