This (early-stage) project develops a formal, empirically tractable framework to study narratives. I define a story as a list of events in chronological order, and a narrative as a collection of stories similar in meaning. I axiomatize story-similarity, and embed it into an empirically-implementable algorithm. My method follows a three-stage process: large language models extract stories from a corpus of natural text; stories are compared using a multi-stage similarity measure that exploits semantic and structural features of stories; and network-based clustering aggregates similar stories into endogenously emerging narratives. (The algorithm avoids introducing demand effects regarding the number, size, themes, or other key features of narratives.) Adopting a revealed-preference paradigm and using a database of news articles, I find preliminary evidence of markets pricing in narratives on American tariffs. In other words, markets behave as if participants on aggregate believe in certain narratives at particular moments in time. I also show that the interpretation of narratives and the attention paid to each one are time-varying. Directions for future research, theoretical conjectures, and the empirical hypotheses they generate will also be discussed.