A state space forecasting model with fiscal and monetary control

In this paper we model the U.S. economy parsimoniously in an a theoretic state space representation. We use monthly data for thirteen macroeconomic variables. We treat the federal deficit as a proxy for fiscal policy and the fed funds rate as a proxy for monetary policy and use each of them as control (exogenous) variables, and designate the rest as state variables. The output (measured) variable is the growth rate of quarterly real GDP which we interpolate to obtain a monthly equivalent. We specify a linear relation between state variables and implicitly allow for time variation of the relationship by using a recursive least squares (RLS) with forgetting factor algorithm to estimate the coefficients. The model coefficients are also estimated using ordinary least squares (OLS) and the resulting forecasts (in-sample and out-of-sample) are compared. The RLS algorithm performs better in the out-of-sample forecasts, particularly for those state variables which exhibit the greatest cyclical variations. Variables which had greater stability were forecasted more precisely with OLS estimated parameters.