Forecasting Total and Type-Specific Non-Residential Building Construction Spending: The Case Study of the United States and Lessons Learned

Abstract

Forecasting construction spending is important for civil engineering practitioners to make business decisions. Currently, the main body of forecasting literature pertains exclusively to aggregate construction investment, such as total construction spending (TTLCON), private construction spending, or residential construction spending. But type-specific construction spending, such as that for education, healthcare, and religion, had yet to be explored using forecasting techniques. This case study presents a viable procedure by which aggregate and type-specific non-residential construction can be forecasted. The procedure that involves the use of the Granger causality test and the Vector Autoregression (VAR) model proved to be able to provide an accurate forecast pre-COVID-19, with some accuracy even during the COVID-19 pandemic period. Lessons learned include the following: (1) effort should be diverted towards model interpretation, as the impulse–response trial yields results conforming to current well-established empirical evidence; (2) a type-specific approach should be adopted when analyzing construction spending, as different types of construction spending react differently to potential indicators; and (3) complex models incorporating multiple indicators should be used to generate a forecast, as a complex model has a higher chance of containing parameters explanatory of the target variable’s features during the testing period.

Keywords

• construction spending
• time-series analysis
• forecasting