Essays on Institutional Ownership and Liquidity in the Corporate Bond Market

Abstract

This dissertation consists of three independent chapters on institutional ownership and liquidity in the corporate bond market. Chapter 1, co-authored with Jian Li, studies the concentration of investor ownership and its impact on liquidity and bond price dynamics. First, we provide evidence that there is considerable dispersion of investor concentration across bonds within a firm. We then show that bonds with lower investor concentration are more liquid, controlling the shares of different investor types. Moreover, during the COVID-19 crisis and 2008 financial crisis, bonds within a firm with lower investor concentration experienced larger initial price drops but faster recovery. We explain our findings by extending the dynamic model of Vayanos (1999) to incorporate multiple risky assets and investors with heterogeneous risk aversion. Chapter 2, co-authored with Jian Li, studies the investor composition and liquidity component of credit spread in the corporate bond market. We show that secondary market frictions have larger effects on credit spreads now than in 2005. We provide a model connecting this to the rapidly growing share of mutual funds in the corporate bond market. The model features investors with different trading needs who choose among a risk-free asset and heterogeneous illiquid bonds. As the risk-free rate declines, short-term investors enter the bond market, reaching for yields. Although they provide liquidity, their greater trading needs amplify the sensitivity of credit yields to bid-ask spreads, leading to larger liquidity components. We test the model's predictions using U.S. investor holdings data and find evidence consistent with the model's prediction. Chapter 3 performs a comparative analysis of state-of-the-art machine learning methods for liquidity prediction in the corporate bond market. I first construct a liquidity measure using the first principal component of 11 liquidity measures, then build a comprehensive data library of corporate bond characteristics for liquidity prediction. I find that tree models such as random foreast and LightGBM outperform the linear models. I trace their predictive gains to their ability to accommodate non-linear effects and interaction among characteristics. Among bond characteristics, characteristics related to past liquidity, inter-dealer trades, and investor composition play the most important roles in liquidity prediction.