Recent research has shown that the standard labor matching model has difficulties in reproducing the co-movement patterns observed in US data. This is due to the fact that the standard model lacks sufficient propagation of shocks. This paper shows that refining the informational structure of the model leads to improvements along this dimension: when agents cannot separately identify persistent and transitory technology shocks on impact (so that they must solve a signal extraction problem), shocks are propagated. Under this specification the standard matching model even manages to make recoveries initially jobless, as in the data.