Essays on Liquidity, Liquidity Risk, and Method of Moments Estimation

Because prices are a function of expected future liquidity levels, a component of unexpected returns is due to new information about future costs to trade. The first two chapters of this dissertation explore this relationship. Chapter 1 extends the Campbell (1991) return decomposition to include liquidity news, which captures the component of unexpected returns due to revisions in future expected liquidity levels. For the market portfolio of NYSE and AMEX stocks between 1964 and 2001, liquidity news has approximately 5 times the volatility of contemporaneous liquidity and 1.2 percent the volatility of contemporaneous returns. Liquidity news’s contribution to portfolio volatility is higher for small and illiquid stocks. Systematic liquidity news risk is priced and has explanatory power for the cross-sectional variation in expected returns. However, the price of systematic liquidity news risk is not statistically different than that of dividend and discount rate news risk. Chapter 2 investigates the cross-sectional variation in changes in valuation after the Tokyo Stock Exchange shut down on January 18, 2006 because its daily capacity of 4.5 million trades had been reached. According to simulations using historical data, 1.5 additional closures are expected over the six month period following the event. Stocks traded only on the TSE lost 2 percent of their value relative to those traded on at least one additional Japanese exchange and liquid stocks lost value relative to illiquid stocks. For instance, high turnover stocks lost approximately 6 percent of their value relative to low turnover stocks. Under small samples, numerous or unbounded moments, or mis-specified models, Empirical Likelihood (EL)’s estimator may have restricted support and may be undefined at or near the population parameter value. Chapter 3 proposes and investigates the properties of an estimator named Penalized Method of Moments (PMM) that merges the GMM and EL objective functions. Through a free parameter, PMM’s estimator properties, including its support, may be adjusted with GMM and EL behavior attained at the extremes. For a simulated Schennach (2006) model with small samples and numerous moments, PMM’s estimates are less volatile and its hypothesis tests are less mis-sized than those of EL.