Applying Probability Management to Cost Overruns: Reference Class Forecasts on Steroids

Large projects are famously behind schedule and over budget. The late behavioral economist Daniel Kahneman (2011) suggested a simple approach he dubbed reference class forecasting to correct this, which simply consists of looking at the outcomes of similar projects. Bent Flyvbjerg has developed a set of reference class forecasts as a database of the statistics of cost overruns for thousands of large projects (Flyvbjerg and Gardner, 2023). The discipline of probability management converts such statistics to data arrays called stochastic information packets (SIPs) that obey the laws of both arithmetic and probability. This allows reference class forecasts to be combined in arithmetical expressions to aggregate the costs of larger projects from subprojects. We have created a SIP library of a few of Flyvbjerg’s project categories using Tom Keelin’s metalog distribution. We have developed a proof-of-concept model in Excel that uses this library to predict the cost overrun for the upcoming Sepulveda Transit Corridor Project in Los Angeles. It is our hope that such interactive models based on increased adoption of SIP libraries of reference class forecasts will lead to more realistic cost estimates on large projects.