Large infrastructure projects go over budget far more often than they come in under it, and researchers at the University of Oxford and elsewhere argue that the main problem lies in the estimate approved before construction begins.

A 2018 paper in the journal Transportation Research Part A, led by Oxford professor Bent Flyvbjerg, reported that 86% of the transportation infrastructure projects in the authors’ data had a cost overrun, with an average overrun of 28% in real terms. The authors said average overruns typically exceed median overruns, a sign that a minority of projects go very far over budget.

The pattern extends beyond transport. A study of large hydropower dams by Atif Ansar, Flyvbjerg and two colleagues concluded that budgets were systematically set below actual costs and left out inflation, substantial debt servicing, and environmental and social costs.

Governments see the problem across whole portfolios. The International Monetary Fund reported in 2015 that cross-country comparisons reveal “average inefficiencies in public investment processes of around 30 percent,” and that the most efficient public investors get twice the growth return of the least efficient.

In the United Kingdom, the National Infrastructure and Service Transformation Authority (NISTA) reported on 213 major government projects with a whole-life cost of £996 billion as of March 2025. It rated 31 projects, or 15%, Red and 30, or 14%, Green, with 135 at Amber. The whole-life cost of Red projects rose to £198 billion from £97 billion a year earlier, mostly because a few large projects moved from Amber to Red.

Why estimates go wrong

The Oxford-led authors argue that scope changes, complexity, geology and bad weather are causes of overruns but not root causes. In their account, the root cause is human bias: planners underestimate those risks through overconfidence and the planning fallacy, and sometimes through what the paper calls strategic misrepresentation.

That view is contested. The paper responds to researchers Love and Ahiaga-Dagbui, who emphasize scope changes and complexity and favor measuring overruns from the contract stage. It also discusses an Australian study by the Grattan Institute, in which a hand-collected set of 51 projects showed overruns on 65% with an average of 52%, while a 542-project consultancy database showed overruns on 34% with an average of 24%. The Oxford-led authors reject the second dataset as not independently verified.

What investors check first

Taken together, the research suggests several questions to ask before committing capital.

  • Which budget is the baseline? The 2018 paper says a contract-stage budget suits judging contractors, while the budget at the decision to build shows whether decision makers were well informed. Measured from contract award, overruns on average look smaller.
  • Are the numbers in constant prices? Overruns should be compared in local currency at one price level. The authors note that financing costs, such as loan interest, are typically left out, so they need separate scrutiny.
  • How did similar projects perform? Reference class forecasting bases an estimate on the actual results of comparable completed projects rather than on the project’s own plan.
  • What confidence level does the budget represent? A P50 budget accepts a 50-50 risk of overrun. A P80 budget aims for an 80% chance of staying within budget, which still means one project in five would exceed it.
  • Who is delivering it, and on what incentives? The authors recommend incentives that encourage everyone involved to stay on budget and a delivery team with a proven record on that type of project.

An article on the website of Mustafa Egemen Sener makes a related case for early planning, arguing that financing, pre-construction conditions and coordination among institutions shape a project’s outcome.

Why it matters

The 2018 paper warns that overruns can bring outsize losses for investors and taxpayers and can even lead to bankruptcy. Because the distribution of overruns has a long upper tail, an average can understate the risk on a single large project, which is why the authors say clients with one or a few megaprojects often budget at P80 rather than at the median or mean.

A poor early rating is not by itself a sign of failure. NISTA says it is “not unusual for projects to be rated as Red earlier in their lifecycle,” when scope, benefits, costs and delivery methods are still being explored. The dam study, for its part, advises policymakers, especially in developing countries, to favor energy options that can be built over shorter periods.

The caveats

The 86% and 28% figures come from the authors’ own transport dataset, and the 2018 paper is a defense of that work against published criticism. Different baselines and datasets produce different results, as the Australian figures show, and transport findings may not apply equally to every sector. The IMF estimate is a cross-country average from 2015.

Corrections and tips: email the editor using the address on the Contact page.