Inference for impulse responses

Poor identification of individual impulse response coefficients does not necessarily mean that an impulse response is imprecisely estimated. This paper introduces a three-pronged approach on how to communicate uncertainty of impulse response estimates: (1) withWald tests of joint significance; (2) with conditional t-tests of individual marginal coefficient significance; and (3) with fan charts based on the percentiles of the joint Wald statistics. The paper also shows how to anchor the impulse response analysis with a priori economic restrictions that can be formally tested and used to tighten structural identification. These methods are universal and do not depend on how the impulse responses are estimated. An empirical application illustrates the techniques in practice. -- impulse response function ; local projections ; vector autoregressions