We propose a method for indentifying discretionary fiscal policy with real time data. The starting point is the observation that automatic stabilizers should depend on true GDP, while discretionary fiscal policy depends on the information that policy makers have in real time. We approximate the information set of policy makers with GDP data released in real time. True GDP is approximated using the last GDP release. Accordingly, we can compute a real time measurement error. Discretionary fiscal policy can be expected to react to this measurement error, whereas automatic fiscal policy will not. We apply this identification approach in order to test the central identifying assumption of Blanchard and Perotti's (2002) seminal structural VAR. According to this assumption, fiscal policy makers do not react to GDP evolutions contemporaneously in a discretionary fashion. We find that government expenditure is adjusted upward if GDP in real time is lower than true GDP. This suggests that fiscal policy makers can use short-term funds to buy goods and services in response to GDP updates. Our results therefore call the identifying assumption of Blanchard and Perotti's (2002) SVAR into question.