UC San Diego

This dissertation studies the effect of parameter uncertainty on the return predictability and volatility of the stock market. The first two chapters focus on the decomposition of market volatility, and the third chapter studies the return predictability. When facing imperfect information, the investors tend to form a learning scheme that encompasses both historical data and prior beliefs. In the variance decomposition framework, the introducing of learning directly impacts the way that return forecasts are revised and consequently the relative component of market volatility based on these forecasts, namely the price movements from revision on future discount rates and those from future cash flows. According to the empirical study in Chapter 1, the former is not necessarily the major driving force of market volatility, which provides an alternative view on what moves stock prices. Learning is modeled and estimated by Bayesian method. Chapter 2 follows the topic in Chapter 1 and studies the role of persistent state variables in return decomposition in order to provide more robust inference on variance decomposition. In Chapter 3 we propose to utilize theoretical constraints to help predict market returns when in sample data is very noisy and creates model uncertainty for the investors. The constraints are also incorporated by Bayesian method. We show in the out-of- sample forecast experiment that models with theoretical constraints produce better forecasts