Practical application of reference class forecasting for cost and time estimations: Identifying the properties of similarity

• Published in: European journal of operational research Vol. 295; no. 3; pp. 1161 - 1179
• Main Authors: Servranckx, Tom, Vanhoucke, Mario, Aouam, Tarik
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 16.12.2021
• Subjects: Project forecasting; Project management; Project similarity; Real project data; Reference class forecasting; Project similarity; Reference class forecasting; Project forecasting; Project management; Real project data
• ISSN: 0377-2217; 1872-6860
• DOI: 10.1016/j.ejor.2021.03.063

•Identify six key properties for project similarity based on in-depth interviews.•Determine reference class forecasting accuracy using inter- and intraaccuracy with cross validation.•Validate the performance of reference class forecasting based on empirical study of 52 real-life projects.•Accuracy depends on the number, combination and interaction effects of properties.•Trade-off exists between number of properties and size of the reference classes. Many project managers underestimate that todays projects are subject to various risks, which might lead to enormous cost overruns and project delays. In the domain of project forecasting, Reference Class Forecasting has been introduced as a method to bypass human judgement by applying an uplift that is based on the forecast errors of similar historical projects. In this research study, we aim to identify the possible drivers of project similarity based on interviews with 76 project managers. Also, we evaluate the performance of Reference Class Forecasting as a project forecasting technique from both a cost and time perspective. Based on an empirical study of 52 real projects, the accuracy of Reference Class Forecasting is successfully demonstrated by implementing a method that considers both the intra- and inter-accuracy of reference classes. The accuracy increases when more project properties are considered, however, a higher number of project properties decreases the size of the reference classes and thus reduces the reliability of the results.