A number of recent studies regress a "narratively" identified measure of a macroeconomic shock directly on an outcome variable. In this note, we argue that this approach can be viewed as the reduced-form regression of an instrumental variable approach in which the narrative time series is used as an instrument for an endogenous series of interest. This motivates evaluating the validity of narrative measures through the lens of a randomized experiment. We apply our framework to four recently constructed narrative measures of tax shocks by Romer and Romer (2010), Cloyne (2013), and Mertens and Ravn (2012). All of them turn out to be weak instruments for observable measures of taxes. After correcting for weak instruments, we find that using any of the considered narrative tax measures as an instrument for cyclically adjusted tax revenues yields tax multiplier estimates that are indistinguishable from zero. We conclude that the literature currently understates the uncertainty associated with quantifying the tax multiplier.