Energy price surges and inflation: Non-monetary policy to the rescue?

Drawing inspiration from the fiscal responses deployed by euro area governments amid the 2021–2022 energy price shock, this paper investigates the capacity of fiscal tools to mitigate inflation driven by rising energy costs in economies that rely on energy imports. The analysis employs a calibrated small open economy framework under a fixed exchange rate regime. The findings highlight energy subsidies — both at the retail level and directed at firms' input costs — as particularly potent instruments, as they compress marginal costs and dampen the transmission of the external price shock to domestic inflation, output, and the current account balance. When prices are indexed to inflation, the shock's adverse effects are amplified; however, indexation simultaneously enhances the effectiveness of consumption tax reductions by curtailing the domestic propagation of the original disturbance.

Model misspecification in multivariate econometric models can strongly influence estimates of quantities of interest such as impulse response functions, structural parameters, and forecast distributions, even more so at longer horizons due to parameter convolution. We propose powered Bayesian VARs (pBVARs) to address specification issues. Instead of using the standard posterior, we rely on the power posterior that raises each equation-specific likelihood to a fractional power, thereby protecting posterior inference from the effects of unknown misspecification. Equation-specific learning rates adapt automatically to the degree of misspecification present in each series. In a comprehensive Monte Carlo study covering a large set of misspecified data-generating processes, pBVARs produce appreciably better calibrated impulse response estimates than standard BVARs. Applied to US data, pBVARs yield economically plausible structural impulse responses to a monetary policy shock that are robust to changes in lag length, a form of dynamic misspecification that can substantially distort inference in standard BVARs.