Instrumental Variable Identification of Dynamic Variance Decompositions

Macroeconomists increasingly use external sources of exogenous variation for causal inference. However, unless such instruments (proxies) capture the underlying shock without measurement error, existing methods are silent on importance that macroeconomic fluctuations. We show that, in a general moving-average model with instruments, variance decompositions instrumented interval-identified, informative bounds. Various additional restrictions guarantee point identification both and historical decompositions. Unlike structural vector autoregression analysis, our do not require invertibility. Applied to US data, they give tight upper bound monetary shocks inflation dynamics.