Modeling and forecasting U.S. public construction

This paper extends Lin's flexible accelerator model of dynamic investment behavior of U.S. public construction by relaxing two of the underlying basic assumptions: (1) the coefficient of adjustment is allowed to vary with the level of government expenditure and (2) the regression coefficients are treated as randomly changing over time rather than being viewed as fixed. The new models afford a better explanation of the behavior of U.S. public construction. Importantly, the forecasting ability of the variable-coefficient-of-adjustment model is tested for the three-year period beyond the sample period and compared to both the Lin's original and ARIMA models. It is found that this new model gives better forecasts of public construction for ten quarters ahead. On the basis of the chi-square test of model stability and the mean squared error, it is concluded that the model with variable adjustment coefficients is a better abstraction of economic reality and improves forecasting accuracy.