Forecasting Infidelity: Why Current Methods for Predicting Costs Miss the Mark

Abstract

High-fidelity forecasts of construction cost indexes and material prices are critical for the successful delivery of infrastructure work projects. Unfortunately, existing models tend to underperform because they either (1) ignore relevant explanatory factors or (2) incorrectly specify system feedback and structure. Through a case study with bitumen, a construction material of prime concern for transportation agencies, this paper presents a novel multivariate cost forecasting approach that overcomes these two gaps. Specifically, based on several diagnostic tests, an autoregressive distributed lag and equivalent error-correction model is specified that correctly captures the feedback structure between bitumen and energy commodities. The study then characterizes the relative merits of the approach by introducing robust deterministic and probabilistic out-of-sample forecast measures. The proposed forecasting approach greatly outperforms conventional methods: 6-month-ahead price projections are at least 25% better across the available deterministic and probabilistic metrics. For state planning agencies, this improved forecasting model will allow decision makers to better predict capital budgeting requirements and resource-planning risks. Furthermore, the proposed performance measures will better equip the construction research community to evaluate future forecasting models.