Recent results by Baldwin et al. (2003) using an empirical statistical model have demonstrated significant skill in subseasonal forecasts of the monthly-mean Arctic Oscillation (AO). In their paper forecast skill derives from long-lived circulation anomalies in the lowermost stratosphere during Northern Hemisphere winter. Here methodology is presented to combine the statistical model with a dynamical medium range forecast model. This gives a significant increase in skill over the statistical model alone. The dynamical model output used is the Climate Diagnostics Center MRF reforecast project with a 15 member ensemble run from 1979-2002.