A framework for forecasting project estimate at completion using historical and current performance data

Abstract: Earned value management (EVM) is an integrated project control method that incorporates project scope, budget, and schedule. EVM uses various metrics to report project cost or schedule performance, such as the cost performance index (CPI) and schedule performance index (SPI). In addition to reporting, EVM can also be used to forecast project estimate at completion (EAC) cost. Accurate estimation of the final cost of a project can initiate corrective actions designed to mitigate potential cost overruns. Because of widespread acceptance of EVM by practitioners, academic scholars have explored and recommended methods to improve EVM in practice. In spite of these advancements, the role of EVM in practice remains limited as a reporting tool. Here, a survey was designed and disseminated to practitioners to further understand why EVM has not been adopted as a forecasting method in practice. Findings of this survey indicate that issues such as forecast inaccuracy may lessen practitioners' commitment to EVM application. A factor contributing to inaccurate forecasts is the assumption, in current practice, to consider all activity groups (e.g., concrete, earthwork) as having the same performance, which is not always valid. Indeed, using an ANOVA-based methodology, the present study observed significant differences between performances of various construction disciplines in a historical dataset. To address the challenge of forecast inaccuracy from this perspective, a discipline-level approach for generating project forecasts, which aggregates discipline-level costs to obtain a project-level forecast, was developed. A Monte Carlo simulation modeling approach was implemented to demonstrate the added benefits of this more granular approach. Since Monte Carlo simulation has limitations in project continuity, Markov modeling was used. In addition, Bayesian statistics were used to incorporate current performance data into calculations. The proposed Markov-Bayesian modeling technique was implemented in a real case study, where it was shown to outperform EVM forecasts. To validate the proposed method, a framework based on randomly generated projects was developed and applied. Two sets of randomly sampled project performance data were generated to train the model and to be used for forecasting. The forecast accuracy of the proposed Markov method was compared against the traditional EVM method in various scenarios. The suggested approach was found to be significantly more accurate in the early stages of a project. Notably, this difference decreased as project performance reached stable conditions (i.e., as variation in cost performance decreased). Early detection of deviations from original plans is critical for initiating corrective actions that are effective. The application of randomly-generated performance data as a validation approach can also be used by other scholars to validate their findings in the area of earned value management.