A 1200-day perpetual January integration of the Berlin troposphere-stratosphere-mesosphere general circulation model (GCM) has been performed with constant boundary conditions. The long-term mean climatology represents the response to the different forcing terms in the zonal-mean momentum budget: the winter stratospheric mean climatology is determined by the balance between the mean meridional circulation (m.m.c.), acting via the Coriolis force, and the Eliassen–Palm flux divergence. The winter stratospheric mean state oscillates between periods of strong wind (undisturbed conditions) and weaker winds (following minor warmings) on periods of several hundred days; this is the response to transient eddy forcing from the troposphere which has an almost-red power spectrum, with maximum power on time-scales of tens of days. As previously found in observations, there is a near cancellation between the wave- and the m.m.c.-induced forcing of the mean flow, but the acceleration is well correlated with the wave forcing. The dominant spatial modes of variability of the GCM are isolated using principal-component analysis. The dominant orthogonal modes of wind and temperature extend from the winter into the summer hemisphere. They represent structures close to thermal-wind balance, with a dipolar structure in the lower stratospheric temperature field associated with variations in the strength of the polar-night jet and weaker wind anomalies in the summer subtropics (22°S, 5hPa). There is some evidence of long-period fluctuations in the vertical velocity in the lower tropical stratosphere associated with the high-latitude velocity variations; these are consistent with the equatorial temperature anomalies. Time-delayed point correlations with the zonal velocity at 62°N, 1 hPa reveal that the weak anticorrelation of velocity in the northern subtropics (28°N, 1 hPa), which is weak at zero-lag, increases with time to about 58% at 20-days lag. Singular-spectrum analysis of the zonal velocity at the three reference points, which isolates orthogonal modes of temporal variability, reveals that the correlations are increased when only the low-frequency velocity variations are considered.