Milestone Markets: Software Cost Estimation through Market Trading

Software cost estimation remains a difficult challenge despite decades of attention by both researchers and practitioners. Predictions are often inaccurate and characterized by very wide confidence intervals. Direct approaches base "expert" estimates on detailed requirements, along with the experience and intuition of the estimator. The Delphi method seeks a consensus estimate among a group of expert estimators. Still other approaches use historical project data to fit estimation models, such as COCOMO. While hybrid techniques like COBRA combine aspects of several methods. This paper proposes a very different approach, the use of information markets to continually aggregate the individual estimates of diverse software project stakeholders. Information markets have been applied successfully in several areas with the market consensus often outperforming individual experts. The paper describes a market mechanism for software cost estimation, explores the characteristics that make such an approach possible, and presents initial experiments based on a simple estimation task.