How Well Do Adverse Selection Components Measure Adverse Selection?

Abstract

The performance of five adverse selection models are examined by comparing their component estimates to other measures of information asymmetry and informed trading. The models produce mixed results. Adverse selection components correlate with various volatility measures, but appear unrelated to measures of uncertainty. Only three of the five models have the expected relation with informed trader proxies, suggesting that the adverse selection models measure adverse selection weakly at best. Spread also relates to many of the volatility measures, suggesting that some adverse selection components might be measuring some other cost of trading. One of the significant recent advancements in the market microstructure literature is the development of models that decompose the bid-ask spread into various components. In these models, the spread generally has three cost components: order processing, inventory holding, and adverse selection (or asymmetric information). Even though these microstructure models provide an important development in empirical finance, we know little about how well these models measure adverse selection and perform relative to each other. In the corporate finance literature, variables such as market-to-book, volatility, and institutional ownership are often used to measure the asymmetric information present in a stock. Recent papers also use adverse selection components as a direct measure of information problems. 1 However, little is known about how well adverse selection components measure information asymmetries. In this paper, we test the performance of five commonly used methods of computing adverse selection components. To determine the usefulness of the adverse selection models in measuring information problems, the relation between adverse selection components and measures of information asymmetries and various proxies for the presence of informed traders are examined. As a benchmark for the analysis, we also examine the performance of spread as a measure of information asymmetry. Using a three-stage simultaneous equation framework, adverse selection models are found to relate inconsistently to the various information variables. The major determinant of adverse selection appears to be volatility. Other measures of information asymmetries, such as analyst forecast errors and market-to-book, are not related to adverse selection. The proxies for informed traders produce mixed results. Overall, the relation for four of the five models are similar to those