Time-Series Analysis for Forecasting Asphalt-Cement Price

Abstract

Variations in the price of asphalt cement have become a serious challenge for Departments of Transportation in proper budgeting of transportation projects. This issue has also pressed contractors in their efforts to develop appropriate cost estimates for transportation projects. Although the price of asphalt cement increases over the long term, it is subject to significant short-term variations. In the current state of practice, the following approaches have been utilized to predict the future price of asphalt cement: (1) adding a fixed percentage of the estimated total cost of asphalt cement to escalate the price of asphalt cement (i.e., adding a risk premium); (2) inflating the estimated cost of asphalt cement to the expected midpoint of construction period to estimate the total cost of asphalt cement; and (3) conducting Monte Carlo simulation analysis to characterize uncertainty about future price of asphalt cement. None of these methods consider autocorrelation in the historical records of asphalt cement price. This paper departs from the existing body of knowledge and challenges the lack of proper treatment of short-term variations in predicting asphalt cement price. The research objectives of this paper are to: (1) identify and characterize variations observed in actual prices of asphalt cement over time; and (2) utilize this knowledge to create time-series forecasting models for asphalt cement price and examine whether and how time-series forecasting models can predict future prices of asphalt cement with higher accuracy compared to the existing approaches. Based on the identified time series characteristics, four univariate time series forecasting models, namely Holt Exponential Smoothing (ES), Holt-Winters ES, Autoregressive Integrated Moving Average (ARIMA), and seasonal ARIMA, are created to take into account the short-term variation of asphalt cement price in forecasting its future values. The forecasting results show that all four time series models can predict future prices of asphalt cement with higher accuracy than the existing methods, such as Monte Carlo simulation. Among the four models, the ARIMA and Holt ES models are the most accurate forecasting models with errors less than 2%. This study can help both owners and contractors improve budgeting process, prepare more-accurate cost estimates, and reduce the risk of asphalt cement price variations in transportation projects.