Abstract
This paper proposes an explanation for the frequent appearance of a price puzzle in VARs designed for monetary policy analysis. It assumes that the data are generated by a model in which output and output gap are not equivalent, while an econometrician follows the common practice of including only output in the VAR. The omission of the output gap is shown to spuriously produce a price puzzle (and several other incorrect conclusions) in a class of commonly used models. This can happen even if the model admits a triangular identification and if the forecasts produced by the misspecified VAR are optimal. When the model is tested on US data, all predictions are supported. A commodity price index is not needed to solve the puzzle.