Moving forecasting error: A risk-based cost forecasting approach

The large amounts of historical pricing data the state transportation agencies (STAs) have collected during the last couple of decades, combined with the information technologies available today, have facilitated greater effectiveness in construction cost estimating procedures through the implementation of data-driven procedures. Ten years ago, these procedures would be considered too advanced or impractical, but current computational tools have made them a feasible option for STAs. However, the application of those data-driven procedures has mainly focused on the pre-forecasting phase of the cost estimating process (to develop cost estimates in current dollars). Cost forecasting activities still rely on antiquated techniques developed before the "computer era", or are performed with annual inflation rates recommended by external entities, and whose suitability to the local construction market is unknown. This dissertation presents a complete analysis of the state-of-the-practice of construction cost estimating in the transportation industry. It proposes data-driven methodologies to address knowledge gaps and opportunities for improvement identified from that analysis. Two data-driven methodologies have been proposed to address the two main phases of an ideal construction cost forecasting process: 1) the development of a construction cost index (CCI) that represents past behavior of the construction market for the intended scope of work and 2) analysis of that scope-based CCI to generate effective annual inflation rates. The development of scope-based CCIs was achieved by the careful application of a data collection and cleaning protocol, which allowed the developed and implementation of a Multilevel Construction Cost Indexes (MCCI) system. Other researchers have previously proposed this cost indexing system, but it has been modified in this dissertation to serve the cost forecasting needs of STAs better. The method used to generate reliable inflation rates from scope based CCIs developed with the proposed MCCI system is called Moving Forecasting Error (MFE). This is a novel method designed to maximize the value of the limited available historical pricing data by evaluating several forecasting scenarios within that data. The output of the MFE methodology is yielded in the form of a risk-based forecasting timeline showing a probabilistic estimate of the future costs for the intended construction activities along different forecasting time horizons. The proposed cost forecasting methodologies were developed and validated through three case studies conducted with three different STAs: the Colorado, Minnesota, and Delaware Departments of Transportation. To satisfy the forecasting needs of STAs, the proposed methodologies should be applicable to at least 20-year forecasts, such as those involved in the Long-Range Transportation Plans (LRTPs) required by federal regulations. To ensure their suitability for long-term forecasting, the MCCI and MFE methods were developed and validated with 20 years of historical bid data from each case study agency. To the best knowledge of the author, this dissertation presents the largest data processing effort to assess and improve STA’s cost forecasting procedures.