This paper examines the periodic net discount rate using genetic programming (GP) techniques to build better short-term forecasts. Standard GP techniques require human judgment as to which data window to use, which may be problematic due to structural breaks and persistence (or long memory) in the net discount rate. We use a recently developed extension of GP to overcome this problem. While our results show no significant out-of-sample forecast improvement relative to the linear alternative or random walk model over the full sample, they do provide evidence as to the stochastic nature of the net discount rate considering the AR(3) model yielded lower forecasting errors in the post-1982 sample.
©2011 Walter de Gruyter GmbH & Co. KG, Berlin/Boston