Peering into Compensation Disclosure : Semantic Similarity and Peer Selection

This paper studies economic determinants of semantic similarity in executive compensation disclosures and its implications for compensation peer selection. We employ a novel measure based on a natural language processing algorithm with machine learning---document embeddings---to capture nuanced semantic aspects of compensation disclosure. Consistent with our predictions, we demonstrate that firms facing similar economic conditions, specifically size and industry, and firms with common compensation consultants are all associated with greater semantic similarity in their compensation disclosures. We then show that similar compensation disclosures can predict future compensation peer selection incremental to economic factors that explain both disclosure similarity and peer selection, suggesting that our measure captures distinct explanatory information about peer selection. This semantics-based measure is widely applicable to the literature on narrative disclosures as it addresses some of the shortcomings of traditional textual measures, thus could enable more research into cross-firm comparisons and disclosure clustering, among other subjects