Reference Class Forecasting: The Flyvbjerg Method
Reference class forecasting (RCF) is the only forecasting method with documented evidence of reducing optimism bias in megaprojects. The UK Treasury and Department for Transport have made it mandatory since 2003.
The five steps
- Identify the reference class. Find past projects that are genuinely comparable in scope, technology, scale, and delivery model. Not "similar in name", similar in structure.
- Establish the distribution of outcomes. For each reference project, calculate the cost overrun against its original approved budget (not against any later re-baseline). Build a distribution.
- Compare your project to the distribution. Identify any structural reasons your project should sit lower (proven technology, smaller scope) or higher (less mature supply chain, first-of-a-kind) than the reference class.
- Pick a target percentile. A 50th-percentile uplift accepts a 50/50 chance of overrun. An 80th-percentile uplift gives roughly 80% confidence of staying within budget. UK Treasury Green Book defaults to the 80th percentile for high-stakes projects.
- Apply the uplift. Add the percentile-derived uplift to the bottom-up budget. Document the assumptions; the uplift is a forecasting tool, not a slush fund.
Worked example: a UK light-rail project
Suppose you are forecasting the cost of a 12-mile urban light-rail extension. Bottom-up estimate is 800M GBP.
Reference class: 60 urban rail extension projects in OECD countries, 1990 to 2020, with original budgets between 300M and 2bn GBP equivalent. From the Flyvbjerg dataset:
| Percentile | Cost overrun (real terms) | Uplift to apply |
|---|---|---|
| 50th (median) | +45% | 1.45x = 1,160M |
| 70th | +68% | 1.68x = 1,344M |
| 80th | +90% | 1.90x = 1,520M |
| 90th | +150% | 2.50x = 2,000M |
At the UK Green Book default 80th percentile, the RCF-adjusted budget is roughly 1.5bn GBP, not 800M. Sponsors who want to publish the 800M figure are choosing to accept roughly 80% probability of overrun.
Sources
- HM Treasury Green Book, supplementary guidance on optimism bias
- Flyvbjerg B. (2008). Curbing optimism bias and strategic misrepresentation in planning: reference class forecasting in practice. European Planning Studies 16(1).
- Kahneman D., Tversky A. (1979). Intuitive prediction: biases and corrective procedures. Management Science 12.