This paper develops a new methodology for estimating both the automatic and discretionary components of fiscal policy in one reaction function using the differences between real-time and ex post data. Discretionary policy should respond to information available to the policy maker at the time (real-time data), whereas automatic fiscal policy should respond to the true state of the economy at the time (proxied by the final data). We find that the intended discretionary response of fiscal policy to the cycle is counter-cyclical. Our estimates suggest that the automatic stabilizers are at the lower end of the range found in the related literature. This new methodology reduces the risk present in the conventional CAB approach that part of the discretionary actions may be wrongly attributed to automatic stabilizers. In that sense, automatic stabilizers are typically not as strong as usually claimed. This could be of particular use in countries where insufficient data exist to estimate structural budget sensitivities directly.