Compensating for fuel subsidy cuts works only when people believe it

Bolivia is trying, for the second time in sixteen years, to stop paying people to burn fuel. The IMF's newly published staff report on its $1.9bn program commits the government to a front-loaded fiscal adjustment while protecting vulnerable groups. It also records a setback: higher oil prices after the war in the Middle East pushed the government back toward energy subsidies. The World Bank, which expects the economy to shrink by 2.8% this year, cites subsidy cuts among the causes (as reported). The last attempt, in December 2010, raised fuel prices by as much as 83% and was rescinded within a week after protests. This time, as everywhere, the promise is that cash compensation will soften the blow.

The problem

The question we take up is narrow: when governments cut fuel subsidies, what determines whether cash compensation actually protects poor households and keeps the reform in place? It is puzzling because the two halves of the literature disagree. Simulation studies find that compensating the poor is cheap and feasible. The historical record is full of compensated reforms that were abandoned, from Bolivia in 2010 to Ecuador in 2019, where a decree scrapping subsidies worth $1.3bn a year was repealed after twelve days. Something the models assume is failing in practice.

What the evidence shows

Start with who gains from subsidies. A review of country studies by Javier Arze del Granado, David Coady and Robert Gillingham found that the richest fifth of households captures six times as much subsidy as the poorest fifth in absolute terms, and that more than half of the welfare loss from higher fuel prices comes indirectly, through the prices of food, transport and other goods. A later update confirmed that most benefits go to high-income households. But there is a twist that matters for the poor. Measured as a share of their budgets, losses are roughly even across the distribution. Simulations for 19 countries by Chris Klaiber, Jun Rentschler and Ira Dorband put the richest decile's absolute burden at 13 times the poorest decile's, yet only 1.1 times as large relative to consumption. Subsidies are an expensive way to help the poor, but removing them without compensation hurts poor households about as much, proportionally, as anyone else, and they have far less cushion.

Can cash close that gap? The models say yes. For Ecuador, Franziska Schaffitzel and colleagues estimate that removing all energy subsidies while raising the Bono de Desarrollo Humano, the national cash transfer, by almost $50 a month would raise the real income of the poorest fifth by 10% and still free more than $1.3bn for the budget. They also show that the fuel matters: removing diesel and cooking-gas subsidies is regressive, removing gasoline subsidies progressive. Across Latin America, Adrien Vogt-Schilb and co-authors find that about 30% of carbon-tax revenue, channeled through existing cash transfer programs, would on average compensate poor and vulnerable households. These are simulations. They establish that compensation is affordable, not that it works when delivered.

The evidence on delivery comes from fewer settings, but it is better identified. In Indonesia, Samuel Bazzi, Sudarno Sumarto and Asep Suryahadi studied a large unconditional cash transfer paid as fuel prices rose. Households that received it on time kept their spending level with non-recipients; those paid late cut spending by 7.5 percentage points relative to them. Because poor households cannot borrow against a promised payment, a delay turns compensation into a cut. In Iran, a 2010 reform replaced energy subsidies with monthly deposits to more than 70 million people, worth about 28% of median per capita household income. Using variation in when households began receiving payments, Djavad Salehi-Isfahani and Mohammad Mostafavi-Dehzooei found no reduction in hours worked or labor force participation, and positive effects for women and self-employed men. The early phase was pro-poor on balance. Yet the reform disappointed: transfers exceeded the new revenue, the gap was financed by printing money, inflation eroded both the payments and public support, and Parliament froze further price increases. Compensation that loses its real value stops compensating.

Coverage is a second constraint. Fast payment requires a list of whom to pay, and lists are imperfect. A randomized trial across 640 Indonesian villages by Vivi Alatas and co-authors found that community ranking identified the consumption-poor slightly less accurately than a proxy means test, though not by enough to change poverty outcomes, and produced higher satisfaction, because communities apply their own notion of who is poor. Perceived fairness, not only statistical accuracy, shapes whether a program is accepted.

That points to the third constraint, and the one the simulations omit: belief. In a nationally representative Nigerian survey, Neil McCulloch, Tom Moerenhout and Joonseok Yang found that people who think the government is corrupt, or doubt that it can deliver compensation, are strongly opposed to reform, while those who receive reasonable public services are more accepting. Mariza Montes de Oca Leon, Achim Hagen and Franziska Holz use a difference-in-differences design on subsidy removals in Bolivia and Mexico in the early 2010s and find that both cut presidential approval. Their model, supported by Mexican individual data, predicts that low trust lowers support across all income groups. A transfer that people do not expect to receive cannot buy their acceptance, however well it is designed on paper.

The ODA view

Read together, the literature suggests that compensation fails less often for lack of money than for lack of timing, real value and credibility. These are questions of opportunity and agency as much as of fiscal arithmetic. A household that cannot borrow loses choices, about meals, school or a clinic visit, in the weeks before a delayed payment arrives. And a state that cannot make credible promises loses the room to reform at all. Our position, proportionate to the evidence, is that Bolivia's compensation should be paid before or with each price increase, through channels people already use; it should prefer broad coverage to precise targeting at the start; it should be protected against inflation and financed without money creation; and its delivery should be published in detail so that trust can be earned rather than assumed. Which fuel goes first matters too. Bolivian researchers at INESAD estimate that fully removing the implicit subsidy on gas for power generation would raise electricity tariffs by 55.5% and cut household real consumption by 6.1%, which argues for sequencing and a protected social tariff.

What we still don't know

The best-identified evidence comes from Indonesia and Iran; Latin America is studied mostly through simulations, which cannot capture delays, leakage or distrust. We know little about the longer-run effects of compensated reforms on schooling, health or women's control of household resources, or about whether trust can be built quickly, for example by paying before prices rise. The data that would answer these questions are within reach: panels or high-frequency phone surveys that follow the same households through Bolivia's adjustment, linked to administrative payment records. Collecting them now would turn this reform into evidence for the next one.