The Economics of Prevention Versus the Cost of Crisis

The economic burden of the polycrisis is not limited to the direct cost of individual disasters. It includes indirect consequences, cascading failures, lost productivity, reduced human capital, health spending, damage to infrastructure and the long-term erosion of development potential. A proactive approach reallocates some of these resources toward interventions that reduce the probability, severity or consequences of future crises.

The five propositions are deliberately modest. Prevention can reduce future crisis spending; early intervention can stop small problems from becoming systemic failures; investments that address several risks generate benefits across several sectors; developing economies can gain most from basic infrastructure and human development; and the value of prevention grows when avoided losses accumulate across generations. None says that every preventive intervention pays.

The tests are equally modest. Each proposition is confronted with data or with the published evidence and, where the evidence is thin, with transparent arithmetic. The simulator then lets the reader apply the main sources of doubt, from uncertain estimates and weak implementation to slow benefits and high discount rates, and see what survives. Most of the case for prevention does.

The OECD's Creditor Reporting System records every reported aid flow by purpose. In 2010, official donors spent $13.1 billion on emergency response and reconstruction in developing countries and $0.5 billion on disaster prevention and preparedness, 25 dollars of response for each dollar of prevention. By 2024 response had doubled to $28.1 billion, but prevention had grown nearly eightfold to $4.0 billion, and the ratio had narrowed to 7 : 1. Over the decade 2015–2024 the totals were $263 billion and $21 billion: about 8 cents of prevention for every dollar of response.

The bias is starkest where crises are deepest. The largest recipients of response aid, Syria ($22.9 billion over the decade), Yemen, South Sudan, Ethiopia, Ukraine, Afghanistan, the West Bank and Gaza, DR Congo and Somalia, received between 0.6% and 6% of their disaster-related aid as prevention. Much of that response is unavoidable, since conflict cannot be prevented by disaster preparedness. But it shows how little of the international system is organised to reduce risk before it materialises.

Disasters themselves cost far more than aid flows can cover. EM-DAT records $3.7 trillion of economic damage from natural disasters between 2000 and 2024, an average of $194 billion a year over the last decade, peaking at $371 billion in 2011. These are recorded losses to property, crops and livestock, which understate damage in poor countries where assets are uninsured and records thin, and exclude the costs of disease, pollution, conflict and lost development that the rest of this dashboard counts.

The evidence base draws on peer-reviewed studies and major institutional assessments. In air pollution, the US Environmental Protection Agency estimated that the 1990 Clean Air Act Amendments delivered benefits of about $2 trillion in 2020 against costs of $65 billion, a central ratio above 30 to 1 (3 to 90). In climate and disasters, the Global Commission on Adaptation found ratios from 2 to 10 for adaptation, and that $1.8 trillion invested in 2020–2030 could yield $7.1 trillion in net benefits. The US National Institute of Building Sciences found $11 saved for every $1 spent on modern building codes and $6 for federal mitigation grants. The World Bank's Lifelines study estimated $4 of benefit for each $1 invested in more resilient infrastructure in developing countries.

In health and human development, childhood immunisation in 94 low- and middle-income countries returned 16 times its cost in avoided illness costs (44 times counting the full value of longer, healthier lives). The Copenhagen Consensus project's 2023 assessments, published in the Journal of Benefit-Cost Analysis, put tuberculosis control at 46, malaria nets at 48, maternal and newborn health at 87 and agricultural research and development at 33. Nutrition interventions to reduce stunting had a median ratio of 18 across 17 countries (3.6 to 48), and universal sanitation and drinking water 5.5 and 2.0.

For the hardest domains the evidence is thinner and the dashboard says so. Primary pandemic prevention (surveillance, wildlife-trade management and reduced deforestation) would cost about $20 billion a year against at least $562 billion of annual losses from zoonotic disease; the ratio depends on how much of that loss it prevents, assumed here at a quarter. Drought resilience in the Horn of Africa returned $2.3 to $3.3 per dollar. Peacebuilding returns $16 per dollar in one model, on strong assumptions; the UN and World Bank estimate that effective conflict prevention would save $5 billion to $70 billion a year.

Many crises grow by compounding. An outbreak's cases multiply until transmission is interrupted; a drought turns into hunger as herds die and prices rise; a neglected bridge deteriorates faster as damage accumulates; grievances harden into violence and violence into displacement. When the cost of responding is proportional to the size of the problem, every doubling time of delay doubles the bill. A problem that doubles every two weeks costs four times as much to address after four weeks and 64 times as much after twelve.

The evidence confirms the logic where it has been measured. The World Meteorological Organization notes that 24 hours' notice of a storm or heatwave can cut damage by 30%, which is why the Global Commission on Adaptation found that $800 million of early warning systems in developing countries could avoid $3 billion to $16 billion of losses a year. The USAID study of drought in Ethiopia, Kenya and Somalia found that earlier response would cut humanitarian costs by about 30% and, with investment in resilience, save $4.2 billion over 15 years, $2.3 to $3.3 per dollar.

Escalation is also why the reactive bias is so costly. Emergency response buys the most expensive version of a crisis, after it has grown, with surge prices, logistics and short-term contracts. Anticipatory action, which releases pre-arranged finance when a forecast crosses a trigger, aims to act inside the doubling time rather than after it. It is the cheapest form of response, and a bridge between response and prevention.

Clean air reduces heart and lung disease, protects crop yields and ecosystems, and often comes with lower carbon emissions. Sanitation prevents diarrhoeal disease, protects water sources, saves time and reduces the spread of infections. Mangroves protect coasts against storm surge, store carbon and sustain fisheries; a 2020 study in Scientific Reports valued their flood protection alone at more than $65 billion a year. Primary pandemic prevention, by slowing deforestation and managing the wildlife trade, also conserves biodiversity and avoids emissions. This is what it means for prevention to address the polycrisis rather than a single crisis.

The dashboard classifies where each intervention's benefits fall across the ten domains. On that classification, 13 of the 17 interventions deliver 40% or more of their benefits outside their lead domain, and seven deliver more than half. A budget process in which each ministry appraises only the benefits that accrue to its own mandate would therefore see a fraction of the return: the climate-adaptation portfolio would appear to return about 2 to 1 rather than 4.9, and drought resilience would fall just below break-even.

This is a classification, not a measurement, and published BCRs differ in how many cross-domain benefits they already include. But the direction is robust and it matters for practice. Investments whose benefits are spread across sectors are systematically under-funded when no one ministry captures enough of the return to justify them. Whole-of-government appraisal, pooled funds and finance ministries that count avoided crisis costs are the institutional remedies.

The interventions evaluated in low- and middle-income settings have a median benefit-cost ratio of 16, against 11 for those evaluated in high-income settings. The highest ratios, for maternal and newborn health, malaria, tuberculosis and agricultural research, are in the basic services and infrastructure that are least complete in developing economies. Where diseases are common, coverage low and infrastructure missing, the marginal dollar prevents more harm.

The aid system does not reflect this. Low-income countries received $117 billion of emergency response and reconstruction aid in 2015–2024 and $4.6 billion for disaster prevention and preparedness, a ratio of 25 to 1. Lower-middle-income countries received $43 billion and $8.2 billion (5 to 1), and upper-middle-income countries $39 billion and $3.0 billion. The countries where prevention pays most are also those whose disaster-related aid is most heavily weighted to response, largely because they are where conflict and protracted crises concentrate.

Recorded disaster damage, as a share of GDP, is similar across income groups in EM-DAT, but those records undercount poor countries' losses, and a given loss is far harder for a poor household or government to absorb. The development case for prevention therefore rests less on the size of recorded damage than on what basic infrastructure and human development protect: lives, schooling, health and the productive assets that drive convergence.

Building codes protect buildings for their whole lives; a well-nourished child earns more for decades; agricultural research raises yields for generations; a forest kept standing keeps protecting. Standard project appraisals typically count 20 to 30 years of benefits at discount rates of 5% or more. For durable prevention, that truncates most of the value. At 2%, a benefit stream that lasts a century is worth 2.6 times as much as the same stream counted for 30 years at 5%; at 1%, more than 3.5 times.

The choice of discount rate is an ethical as well as an economic judgement. A high rate gives little weight to people born in the second half of this century, who will bear the accumulated consequences of today's under-investment in prevention. A low rate, applied consistently, makes durable prevention among the best investments available. The simulator lets the reader choose and see how much of the portfolio's value depends on it.

Not every benefit is durable. Emergency preparedness and early warning must be maintained; a vaccine campaign protects one cohort; peace agreements can unravel. The dashboard therefore assigns each intervention a persistence share, the fraction of its benefits that continue beyond the standard appraisal horizon, from 0.3 for early warning to 0.9 for building codes, resilient infrastructure, nutrition and agricultural research.

For each draw, each intervention's ratio is sampled from a triangular distribution on the logarithm of its range, peaking at the central estimate. It is then multiplied by the implementation-capacity factor and discounted for any extra delay. It is scaled by a horizon factor: for the persistent share of benefits, the value of the stream over your horizon at your discount rate relative to 30 years at 5%. Under lead-domain accounting, only the largest domain share is counted. The portfolio ratio is the average across the selected interventions, since the budget is split equally.

With the defaults (all 17 interventions, 70% capacity, two years of extra delay, 5% over 30 years, whole-of-government accounting), the portfolio's median ratio is about 13 and its 10th percentile about 11: even the pessimistic draws return many times the cost. Cutting capacity to 40%, adding ten years of delay and counting only lead-domain benefits lowers the median to about 3, still well above break-even. Only when every assumption is set to its harshest (30% capacity, 15 years' delay, an 8% discount rate over 20 years and siloed accounting) does the portfolio fall to about break-even. A 2% rate over a century raises the median to about 28.

Caveats apply. Published ratios measure average returns over the scale studied; spending far more on one intervention would meet diminishing returns. Draws are independent, so correlated errors, such as a shared over-estimate of the value of a life, are not captured. And the estimates come from different methods and settings. The simulator is a way to see which conclusions are robust to doubt, not a precise forecast.

Count the counterfactual. Appraisals should compare prevention not with doing nothing, but with the expected cost of the crises it avoids, including response, reconstruction and lost development, and count benefits across all the domains it affects. Finance ministries can publish contingent-liability estimates for disasters, epidemics and infrastructure failure, so that avoided costs become visible on the balance sheet that pays for them.

Pre-arrange the money. Contingency funds, catastrophe bonds, risk pools and contingent credit lines with triggers linked to forecasts move part of crisis spending before the peak. Anticipatory action converts a share of humanitarian budgets into earlier, cheaper interventions. Development banks can attach crisis-resilient debt clauses and scale grant finance for prevention in the poorest countries, where the response-to-prevention ratio is highest.

Invest in the foundations. The highest ratios are in health, nutrition, water, sanitation, agricultural research and resilient infrastructure: the basics of development. Protecting their budgets in downturns, maintaining what is built, enforcing building codes, extending early warning to everyone, and funding pandemic surveillance and conflict prevention as public goods are the core of a prevention agenda for the polycrisis.

Aid: OECD Creditor Reporting System via the OECD Data Explorer (dataflow DSD_CRS@DF_CRS 1.7), all official donors (ALLD) minus Türkiye, ODA (measure 100), gross disbursements, constant 2024 prices, sectors 720, 730, 740, 700 and all sectors, 2010–2024, for developing countries in aggregate and by recipient. Disasters: EM-DAT economic damages via Our World in Data (current US$; country damages as % of GDP, 2000–2024). Countries with at least 0.5 million people; World Bank income groups.

Evidence: 17 benefit-cost estimates from the sources listed, with published ranges where available and otherwise ×/÷ 2; early warning and pandemic prevention ratios are derived as described in the table. Domain weights, lags and persistence are H Heuristics classifications. Simulator: log-triangular draws; adjusted BCR = BCR × capacity × (1 + r)^−delay × [(1 − p) + p × A(r, H) ÷ A(5%, 30)], where A is the annuity factor and p persistence; lead-domain accounting multiplies by the largest domain weight; 2,000 draws; equal budget shares.

Caveats: published BCRs use different methods, discount rates and benefit definitions and are not strictly comparable; they are averages that do not hold at unlimited scale; the domain classification is judgemental; aid data capture only reported official flows and not domestic budgets; EM-DAT undercounts damages in poor countries; the escalation calculator is arithmetic, not a forecast.