Two Similarities, One Firm Value: A Natural Language Processing Study of IPOs’ First-Day Returns

Abstract

Conflicting results tie audiences’ valuations to firms’ similarity to category prototypes or to outliers. Explanations of these results generally assume that prototypes and outliers have a strong weight in audiences’ valuations although audiences may use them differently in different contexts. We challenge this assumption and propose that the weight of prototype similarity and outlier similarity in audiences’ valuations is contingent on prototypes’ and outliers’ prominence in audiences’ eyes in a given context. Prototype similarity plays a lesser role in audiences’ valuation when specific knowledge rather than general knowledge is needed to value an entity. Outlier similarity plays a heavier role in audiences’ valuations when recent trends in the market recall outliers’ features to audiences. We measure prototype and outlier similarity by applying a natural language processing technique to 160,000 financial documents and test our hypotheses based on the first-day returns of 2,488 U.S. IPOs from 1996 to 2015. We find that firms with prototype similarity are less underpriced in low-tech industries, but that this effect disappears in high tech industries, where more specific knowledge is required to value any particular firm. Firms with outlier similarity are more underpriced, particularly during periods of hot IPO market, during which highly underpriced IPOs are recalled to investors. Our findings extend understandings of optimal distinctiveness, valuation, market categories, and computational approaches to strategy.