The Durbin‐Watson test is shown to reject the hypothesis of independent disturbances in the well known Friedman‐Meiselman study. To cope with this the models have been re‐estimated by maximum likelihood assuming an autoregressive disturbance process. The resulting estimates are consistent with Leijonhufvud's analysis of the economics of Keynes. Large sample methods have been applied in a highly exploratory spirit since for the sample sizes involved such methods may not be superior to ones adopted by Friedman‐Meiselman. Our results imply that the Friedman‐Meiselman estimates must be interpreted with caution. This is particularly acute for the 1933–1938 depres