Research

We identify supply and demand shocks to the real price of natural gas leveraging exogenous temperature variation and high-frequency price movements around gas-supply events. Gas demand is less elastic than supply, leading to larger price fluctuations following supply shocks. These shocks have notable macroeconomic effects: strongly inflationary in the Euro Area, amplified by inventory dynamics and financial volatility, suggesting a transmission channel driven by expectations and uncertainty. They were a major driver of the post-pandemic inflation surge, though effects were more muted in the U.S. Aggregate real effects appear short-lived, though we document notable sectoral heterogeneity and a substantial decline in German activity.

This paper investigates the dynamic effects of weather shocks on economic activity in Europe’s three largest economies: Germany, France, and Italy. We develop a novel approach to measuring country-level exposure to abnormal weather, based on grid-level data weighted by economic activity. We construct five harmonized weather indices—heat, cold, drought, precipitation, and wind—and, using a Bayesian SVAR framework, assess their impact on output and prices across major sectors: energy, construction, manufacturing, and services. The results show that weather shocks have both direct and indirect effects on economic activity, with substantial heterogeneity across shock types and production sectors.

When macroeconomists use instruments, do they seek to identify latent structural shocks or exogenous shifts in observed variables? This paper clarifies how normalization makes this economic choice operational and traces its implications for identification and inference. In SVAR-IV and LP-IV, impact normalization yields the dynamic effect of an exogenous movement in an observed variable. It admits a reduced-form IV interpretation that need not invoke latent shocks; in VARs, it need not require shock recovery or invertibility. A shock-scale normalization, such as one standard deviation, instead targets a latent disturbance and uses the additional structural assumptions that recover its scale. The alternatives have different relevance and inference problems. Impact-normalized responses are ratio estimators with normalization-specific first stages. I propose a shock-based relevance diagnostic that targets the primitive instrument–shock condition and show that shock-scale responses avoid the additional weakness caused by selecting a poorly affected normalization variable, though they remain vulnerable to weak relevance for the shock itself. A natural-gas application and Monte Carlo evidence illustrate the implications for confidence intervals and weak-identification diagnostics.