Construction Forecasting Using Time Series Volatility Models

Abstract

Most linear time series models, including the univariate time series models, assume a time series has a constant variance over time. However, many construction time series data have not shown a constant variance. The volatility of a construction time series variable over time is challenging for accurate forecasting and risk management. This chapter discusses two time series volatility models (i.e., ARCH and GARCH) to forecast the variance of a construction time series. The ARCH and GARCH models are developed for modeling the volatility of the total federal construction spending time series published by the US Census Bureau. Then, the ARCH and GARCH models are combined with the ARIMA model to jointly estimate the mean and error variance of the total federal construction spending time series. The results show that the ARIMA-ARCH and ARIMA-GARCH models assuming a time-varying variance outperform the ARIMA model assuming a constant variance in terms of accuracy. R code examples are provided to develop time series volatility models and forecast a time series considering its time-varying variance. Exercise problems are presented at the end of the chapter for readers to review and practice the time series volatility models.