Overview
A new study provides the first rigorous estimate to date of what it would cost to eliminate extreme poverty globally using direct income transfers. Unlike existing back-of-the-envelope calculations, this analysis accounts for the actual information constraints faced by governments when administering social programs. Using modern statistical learning methods and high-quality household survey data from 23 countries that together include roughly half of the world’s extreme poor, the study estimates:
- $170 billion per year to reduce poverty to ~1% in those 23 countries.
- $318 billion per year, or roughly 0.3% of global GDP, to achieve the same target worldwide.
These findings imply that ending extreme poverty, while far from trivial, is financially feasible at the global scale.
Background and Motivation
Global poverty has fallen dramatically over the past three decades, but hundreds of millions of people still live below the international poverty line. Policymakers, donors, and international institutions often ask a deceptively simple question:
What would it actually cost to end extreme poverty?
Until now, answers have relied on back-of-the-envelope calculations, or on untested causal assumptions about the effectiveness of aid programs. This paper introduces a new approach: reframing poverty reduction as a statistical learning problem. By combining modern machine learning methods with nationally representative household surveys, the study provides more realistic, policy-relevant estimates of the costs and tradeoffs of alternative approaches to ending poverty.
Why This Study Matters Now
Recent decades have seen historic reductions in extreme poverty, with the global headcount rate falling from 41% in 1981 to 8% in 2024. This progress makes eradicating the remaining poverty more affordable than at any point in modern history.
Yet momentum is slowing. Many of the countries home to the largest numbers of poor people are projected to grow slowly, if at all, over the next decade. As a result, the world is hundreds of millions of people off track from meeting the Sustainable Development Goal of eradicating poverty by 2030. This makes it all the more urgent to identify ways to accelerate poverty reduction.
Data and Methodology
The study analyzes 23 countries that meet two conditions:
- Substantial poverty: The national poverty rate exceeds 10%, or the country is home to 1% or more of the world’s extreme poor.
- Recent, high-quality data: Country has a recent nationally-representative survey with consumption data that are publicly available.
Together, these countries accounted for 50% of the world’s extreme poor as of August 2025.
The authors use these data to compare the cost of poverty reduction under three approaches:
- Perfect targeting (infeasible benchmark): Assumes policymakers know each household’s exact consumption and can give precisely the “top-up” needed to reach the poverty line.
- Learned Policy (feasible, data-driven targeting): Uses modern statistical learning to allocate transfers based on verifiable household characteristics – the kind of information governments actually observe.
- Universal Basic Income (UBI) or Universal Supplemental Income (USI): Transfers provided universally, at either the poverty line itself or at a country-specific level chosen to achieve a poverty-reduction goal.
Finally, the paper uses these results to estimate what ending poverty would cost in other countries not in the sample based on their current levels of poverty, as measured by their poverty rates and poverty gap indices.
Key Findings
The main analysis yields three main findings:
- Perfect targeting is cheapest – but impossible to implement. If policymakers could give each household exactly the top-up needed to reach the poverty line, it would cost about $31B per year to reduce the poverty rate to 1%, in the sample of 23 countries. However, policymakers never know exactly what each household needs.
- UBI is far more expensive. A flat $2.15/day UBI would cost $895B.
- Learned policies strike a practical middle ground. A data-driven policy that reduces the poverty headcount to 1% would cost $170B per year — roughly 19% of the cost of UBI, but achieving virtually the same poverty reduction.
Taken together, these results illustrate both that realistic policies cost substantially more than hypothetical ones that perfectly target the poor, but also that in many countries there are large efficiency gains available from data-driven targeting. The exceptions are the world’s very poorest countries (e.g., Malawi) in which baseline poverty rates are so high that targeting saves very little relative to universal transfers.
Finally, the global extrapolation (out-of-sample analysis) illustrates what these results imply at a global scale: they correspond to a cost of (approximately) ending extreme poverty equal to 0.3% of global GDP.
Policy Implications
1. Ending extreme poverty is financially achievable with coordinated global action.
The study’s central finding – that global poverty could be nearly eliminated for about 0.3% of global GDP – reframes what is possible. While significant, this cost is small relative to existing global expenditures, such as alcohol (2.2% of global GDP) or cosmetics (0.6%).
2. Smarter, data-driven targeting can save hundreds of billions of dollars.
Universal income schemes are simple, may be appropriate on normative grounds, and may make sense as a poverty reduction strategy in the poorest countries, but on average they are five times more expensive for reducing poverty than data-driven targeting. By leveraging modern statistical learning, governments can deliver transfers more efficiently – capturing most of the poverty-reduction benefits of UBI at a fraction of the cost.
3. Support poverty reduction without undermining high-impact programs.
The analysis quantifies direct transfer costs only. It does not suggest shifting funds away from essential interventions in health, education, or disease prevention – actions that could ultimately worsen poverty, and thus invalidate the analysis, which is based on data collected when those programs were running. As the study notes, defunding effective programs (e.g., malaria prevention) would be a “false savings.” Policymakers should view direct transfers as complements to, not substitutes for, other proven development investments.
4. Global and philanthropic financing can play a transformative role.
Most low-income countries cannot finance full poverty eradication from domestic resources alone. Yet the global numbers imply substantial potential for international cooperation and private philanthropy. With current levels of wealth, over 100 individuals worldwide could eliminate poverty in an entire low-income country using the returns on their assets. For donors and governments alike, the study offers clear, credible benchmarks for what meaningful contributions could achieve.
Limitations
- The study is based on the most recently available high-quality nationally representative survey data, which in some cases are several years old. Obtaining the most accurate possible estimates as of today would require running fresh surveys.
- The study does not account for ways in which the income transfer policies it learns might affect individuals earnings from other sources, either directly (e.g. receiving money lets them expand a business), indirectly (e.g., neighbors receiving money lets them expand a business), or due to incentives (e.g., the transfer policy disincentives effort). Given the available evidence on these points to date the authors anticipate that they would on net tend to further reduce poverty, but this would need to be tested.
- The study does not directly address the details of implementation. Cash transfers are already delivered at comparable scales in many low- and middle-income countries around the world—reaching up to 1.6 billion individuals during the pandemic, for example—so that this does not appear to be a fundamental constraint. Some countries or regions within countries would require significant new investment in infrastructure for identity verification and payment delivery, however.
