How does private consumption react to an exogenous increase in government expenditure? Standard structural vector autoregressions (SVARs) usually report a positive GDP as well as consumption response, while event studies report a negative consumption response. We investigate in a SVAR whether anticipation of the fiscal shock reverses the sign of this dynamic response to a negative one. As a methodological contribution, we model expectation formation within a SVAR framework. We show for the US that consumption falls in reaction to an expenditure shock once the model allows for one-period-ahead anticipation of this shock. Modelling anticipation of fiscal shocks is thus crucial to correctly capture their macroeconomic effects. Differences in results between event studies and VARs can be explained by missing anticipation in VARs. When re-estimating the two models (with and without anticipation) for non-defense related expenditures, we find a positive consumption response for both models. The implications of our results for macroeconomic theory are briefly discussed.