Jacopo Cimadomo , 2007.
"Fiscal Policy in Real Time,"
CEPII Working Paper 2007- 10 , May 2007 , CEPII.
In this paper we argue that any assessment on the intentional stance of fiscal policy should be based upon all the information available to policymakers at the time of fiscal planning. In particular, real-time data on the discretionary fiscal policy “instrument”, the structural primary balance, should be used in the estimation of fiscal policy reaction functions. In fact, the ex-post realization of discretionary fiscal measures may end up to be drastically different from what intentionally planned by fiscal authorities in the budget law. If this is the case, and if revision errors in the policy indicator are correlated with the ones in the regressors, it is shown that commonly used estimators become biased possibly inducing a misleading judgement on the policy stance. We derive the functional form of that bias and, based on empirical second-order moments, we are able to accurately predict the potential impact of using revised data in the evaluation of the ex-ante stance of fiscal policy. When fiscal policy rules are estimated on real-time data, our results indicate a counter-cyclical stance in OECD countries, especially during economic expansions. This contrasts with conventional findings based on revised data, which point to fiscal policy acyclicality or pro-cyclicality, and with Forni and Momigliano (2005) who employ real-time data for the output gap and find countercyclicality, but just in recessions. Further, we test whether threshold effects might be at play in the reaction of fiscal policy to the economic cycle and to debt accumulation. It emerges that the intentional cyclical behavior of fiscal policy is characterized by two regimes, and that the switch between them is likely to occur when output is close to its equilibrium level. On the other hand, the use of revised data does not allow to identify any threshold effect.
Fiscal policy ; Cyclical stabilization ; Real-time data ; Revision errors ; Endogenous threshold models