We develop a structural vector autoregressive framework that combines external instruments and heteroskedasticity for identification of monetary policy shocks. We show that exploiting both types of information sharpens structural inference, allows testing the relevance and exogeneity condition for instruments separately using likelihood ratio tests, and facilitates the economic interpretation of the structural shock of interest. We test alternative instruments and find that narrative and model‐based measures are valid, while high‐frequency data instruments show signs of invalidity. Finally, we document that monetary shocks identified with both a valid instrument and heteroskedasticity have larger effects on production and prices than monetary shocks identified via an instrument only.
