Predicting construction cost index using the autoregressive fractionally integrated moving average model

Abstract

The construction cost index (CCI) is a quantitative construction-cost indicator proposed by Engineering News-Record (ENR). Because predicting CCI is crucial to the investment planning, bidding, and profitability of construction projects, considerable effort has been devoted to CCI estimation, and substantially accurate results have been obtained. However, these findings were based on the assumption of a Gaussian distribution of data, which limits the estimation accuracy for fluctuating data. To overcome the limitation of such conventional methods, the present study aimed to refine CCI prediction performance by applying the concept of long memory. First, the existence of long memory in CCI is examined by performing rescaled range (range/scale or R/S) analysis. Second, a time-series model was developed: the autoregressive fractionally integrated moving average (ARFIMA) model, which reflects the characteristics of long memory. Finally, the prediction performance of the ARFIMA model was compared with that of the conventional autoregressive integrated moving average (ARIMA) model. CCI data from January 1990 to August 2016 were used to develop these models. The results showed that ARFIMA outperformed ARIMA in terms of prediction performance, on average, by 9.5%. The ARFIMA model achieved higher CCI prediction performance by incorporating the properties of long memory. In summary, the results confirmed the importance of applying long memory to CCI predictions. Furthermore, the developed model could play a key role in improving the accuracy of cost estimation in the construction market. (c) 2017 American Society of Civil Engineers.