Essays in Empirical Asset Pricing

Abstract

This dissertation explores three topics in empirical asset pricing, with a focus on cross-sectional anomalies, factor model evaluation, and information infrastructure in shaping cross-sectional returns and institutional investor demand.

In the first chapter, co-authored with colleagues, I show that merger announcement returns account for virtually all of the measured size premium. An empirical proxy for ex ante takeover exposure positively and robustly relates to cross-sectional expected returns. The relation between size and expected returns becomes positive or insignificant, rather than negative, conditional on this takeover characteristic. Asset pricing models that include a factor based on the takeover characteristic outperform otherwise similar models that include the conventional size factor. We conclude that the takeover factor should replace the conventional size factor in benchmark asset pricing models.

The second chapter, co-authored with a colleague, critiques the prevailing methods used to evaluate asset pricing models. Many popular models incorporate factors motivated by previously documented cross-sectional return patterns. We argue that popular "out-of-sample" methods for evaluating and comparing such models do not adequately protect against biases driven by the data-instigated nature of the models. Empirically, we show that maximum Sharpe ratio estimates fall substantially for many models when computed using validation samples designed to mitigate data-instigated model bias. Our lower estimates are easier to reconcile with leading risk-based economic models. However, it is also less clear to what extent popular multifactor models actually outperform the classic capital asset pricing model.

The final chapter investigates the role of financial data vendors in capital markets. These vendors collect, aggregate, and process data on clients' behalf. I show that data vendors' coverage decisions affect institutional investor demand. The focal vendor in this study, Standard and Poor's ('SandP') Compustat database, provides subscribers with decades of 10-K and 10-Q data; however, it does not cover every public firm in every period. I show that institutional investment in firms with no Compustat coverage is over 36% below its unconditional mean, even controlling for other firm characteristics. A novel quasi-natural experiment establishes a plausibly causal connection: a technology improvement at SandP in the 1990s causes a discrete reduction in missing data. This change in data coverage is followed by a significant increase in institutional investment for treated firms relative to control firms. I then show that missing Compustat data is associated with lower informational efficiency of equity prices. I conclude that data vendors' actions can exert a material influence on capital markets because they affect firms' access to institutional capital.